JP2016046695A - Acoustic quality evaluation device, acoustic quality evaluation method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To estimate a MOS value from a PESQ value performing a little amount of calculation.SOLUTION: An acoustic quality evaluation device: obtains a first PESQ value for a first reference acoustic signal and a first evaluation object acoustic signal based on a signal including the first reference acoustic signal; and obtains a first MOS value by linearly converting the first PESQ value on the basis of linear relation between a second PESQ value corresponding to a second reference acoustic signal and a second evaluation object acoustic signal based on a signal including the second reference acoustic signal, and a second MOS value based on five-stage evaluation about the difference between reference sound corresponding to the second reference acoustic signal and evaluation sound corresponding to the second evaluation object acoustic signal.SELECTED DRAWING: Figure 13

Description

  The present invention relates to a technique for evaluating call quality, and more particularly to a quality evaluation test technique for a loudspeaker communication system.

  Conventionally, when a conversation MOS (Mean Opinion Score) value or a listening MOS value is estimated using a PESQ (Perceptual Evaluation of Speech Quality) value that is an objective evaluation value, a PESQ value and a conversation MOS value are based on a reference signal. Alternatively, it is necessary to formulate a nonlinear function representing the correspondence relationship with the listening MOS value, and to perform nonlinear conversion based on the function (see, for example, Non-Patent Document 1).

Information and Communication Technology Committee: “IP Phone Call Quality Evaluation Method”, JJ-201.001, 5th edition, August 2008.

  According to this method, complicated non-linear processing is required for the conversion from the PESQ value to the estimated value of the conversation MOS or listening MOS, and there is a problem that the calculation becomes complicated.

  An object of the present invention is to provide a technique for estimating a MOS value from a PESQ value with a small amount of calculation.

  In the present invention, a first PESQ value for the first reference acoustic signal and the first evaluation target acoustic signal based on the signal including the first reference acoustic signal is obtained, and the signal including the second reference acoustic signal and the second reference acoustic signal is obtained. 5 levels of differences between the second PESQ value corresponding to the second evaluation target acoustic signal based on the reference sound corresponding to the second reference acoustic signal and the evaluation sound corresponding to the second evaluation target acoustic signal Based on the linear relationship with the second MOS value based on the evaluation, the first PESQ value is linearly converted to obtain the first MOS value.

  In the present invention, the PESQ value and the MOS value can be approximated to a linear relationship by adopting the MOS value based on the five-step evaluation of the difference between the reference sound and the evaluation sound. Therefore, the MOS value can be estimated from the PESQ value with a small amount of calculation.

FIG. 1 is a block diagram illustrating a functional configuration of the data generation apparatus according to the first embodiment. FIG. 2 is a conceptual diagram for explaining a data structure generated by the data generation apparatus according to the first embodiment. FIG. 3 is a diagram for illustrating a data structure generated by the data generation apparatus of the first embodiment. FIG. 4 is a block diagram illustrating a functional configuration of the data generation device according to the second embodiment. FIG. 5A is a block diagram illustrating the communication environment simulation processing unit of FIG. FIG. 5B is a block diagram illustrating the signal processing unit of FIG. FIG. 6 is a block diagram illustrating a functional configuration of the sound quality evaluation apparatus according to the third embodiment. FIG. 7 is a diagram illustrating display contents in the sound quality evaluation test of the third embodiment. FIG. 8 is a diagram for illustrating the acoustic quality evaluation method. FIG. 9 is a diagram for illustrating the acoustic quality evaluation method. FIG. 10 is a diagram for illustrating the acoustic quality evaluation method. FIG. 11 is a diagram for illustrating the acoustic quality evaluation method. FIG. 12 is a diagram for illustrating the acoustic quality evaluation method. FIG. 13 is a block diagram illustrating a functional configuration of the sound quality evaluation apparatus according to the fourth embodiment. FIG. 14 is a diagram illustrating the relationship between the DMOS value and the PESQ value. FIG. 15 is a block diagram illustrating a functional configuration of an acoustic quality evaluation apparatus according to a modification of the fourth embodiment.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[First Embodiment]
<Evaluation test simulating conversational MOS test in loudspeaker communication system>
First, an evaluation test simulating a conversation MOS test in a loudspeaker communication system will be conceptually described. In this evaluation test, a near-end speaker and a far-end speaker have a conversation through a loudspeaker communication system, and an evaluator located on the near-end speaker side evaluates the quality of the loudspeaker communication system. Note that the loudspeaker communication system is a communication system that transmits and receives an acoustic signal between terminal devices including a microphone and a speaker, and at least a part of the sound output from the speaker of the terminal device is a microphone of the terminal device. The sound received by the sound (the sound wraps around). An example of a loudspeaker communication system is an audio conference system or a video conference system.

  In the loudspeaker communication system illustrated in FIG. 2, the near-end speaker's voice is received by the near-end speaker's microphone, and an acoustic signal obtained based on the sound is transmitted to the far-end speaker via the network. The sound represented by the acoustic signal is output from the far-end speaker. Further, the far-end speaker side sound is received by the far-end speaker side microphone, and the acoustic signal obtained based on the received sound is transmitted to the near-end speaker side via the network, and the sound represented by the acoustic signal is Output from the near-end speaker. However, at least part of the sound output from the speaker on the far end speaker side is also received by the microphone on the far end speaker side. That is, the far-end speaker's sound received by the far-end speaker's microphone is obtained by superimposing the near-end talker's voice (acoustic echo) on the far-end talker's voice. The acoustic signal transmitted to the near-end speaker is derived from a processed signal obtained by performing predetermined “signal processing” on the signal representing the sound received by the far-end speaker's microphone. It may be obtained without performing such signal processing. “Signal processing” may be any processing. An example of “signal processing” is processing including at least one of echo cancellation processing and noise cancellation processing.

The evaluator listens to the direct sound from the near-end speaker with one ear (for example, the ear that is not the dominant ear—for example, the right ear) using a binaural sound reproduction device such as headphones or earphones. The sound output from the speaker on the speaker side is heard with the other ear (for example, the dominant ear—for example, the left ear), and the speech quality is subjectively evaluated (opinion evaluation). In this embodiment, the channel on the direct sound side from the near-end speaker is denoted as “Rch”, and the channel on the sound side output from the near-end speaker is denoted as “Lch”. As described above, the sound output from the speaker on the near-end speaker side is the sound on the far-end speaker side where the acoustic echo of the near-end speaker sound is superimposed on the far-end speaker sound. The sound signal received based on the microphone on the speaker side and transmitted based on the sound is transmitted to the near-end speaker side and output from the speaker on the near-end speaker side. Therefore, the acoustic echo component of the near-end speaker's voice included in the sound output from the near-end speaker's speaker is delayed from the direct sound of the near-end speaker's voice (the acoustic signal is near Delay of one round trip between the end speaker and far end speaker). In addition, the far-end speaker's voice component included in the sound output from the near-end speaker's speaker is delayed from the time when the far-end talker's voice is emitted (the acoustic signal is far away). The delay in time transmitted from the end speaker side to the near end speaker side). Here, an acoustic signal representing the direct sound from the near-end speaker and an acoustic signal representing the sound output from the speaker on the near-end speaker side when there is a sound wraparound on the far-end speaker side, The set is called a “degraded signal”. In particular, a “degraded signal” that has not been subjected to the “signal processing” is denoted as “degraded signal D ₁ ”, and a “degraded signal” that has been subjected to “signal processing” is denoted as “degraded signal D ₂ ”. For reference, the sound signal representing the direct sound from the near-end speaker and the sound output from the near-end speaker when assuming that there is no sound wraparound at the far-end speaker A set of acoustic signals to be expressed is referred to as a “reference signal”. The evaluator subjectively evaluates the call quality by comparing any set of “degraded signal D ₁ ”, “degraded signal D ₂ ”, and “reference signal”, for example.

<Data generation device>
Next, a data generation apparatus that generates a data structure for performing an evaluation test simulating a conversation MOS test in a loudspeaker communication system will be exemplified. As illustrated in FIG. 1, the data generation apparatus 1 according to the present embodiment includes a near-end speaker acoustic signal storage unit 101, a far-end speaker acoustic signal storage unit 102, playback units 103 and 104, speakers 105 and 106, and a microphone. 107, a time adjustment processing unit 108, a recording processing unit 109, a near-end terminal unit 110, a far-end terminal unit 120, output units 131, 132, 141, 142, 151, 152, and a data storage unit 180. The far-end terminal unit 120 includes a signal processing unit 121, and the near-end terminal unit 110 and the far-end terminal unit 120 are configured to be able to communicate through a network (NW). At least the speakers 105 and 106 and the microphone 107 are arranged in the same room. The data generator 1 is connected to a speaker or a microphone, for example, a processor (hardware processor) such as a CPU (central processing unit), a memory such as a random-access memory (RAM), a read-only memory (ROM), or the like. Is a device configured by executing a predetermined program by one or more general-purpose or dedicated computers. Each computer may include one processor or memory, or may include a plurality of processors or memories. This program may be installed in a computer, or may be recorded in a ROM or the like in advance. In addition, some or all of the processing units may be configured using an electronic circuit that realizes a processing function independently instead of an electronic circuit (circuitry) that realizes a functional configuration by reading a program like a CPU. . In addition, an electronic circuit constituting one device may include a plurality of CPUs.

<Data generation processing>
Next, the data generation process of this embodiment is demonstrated.
As pre-processing, a near-end speaker acoustic signal (first acoustic signal on the first end side of the system) representing a sound corresponding to the direct sound (near-end speaker voice) of the near-end speaker that the evaluator listens to The data is stored in the near-end speaker acoustic signal storage unit 101, and the far-end speaker acoustic signal (the second end side of the system) representing the sound corresponding to the far-end speaker's direct sound (far-end speaker's voice) is stored. The data of the second sound signal is stored in the far-end speaker sound signal storage unit 102. Both the near-end speaker sound signal and the far-end speaker sound signal of this embodiment are time-series sound signals, and are obtained based on, for example, sound recorded in a soundproof room. However, this does not limit the present invention, and at least one of the near-end speaker sound signal and the far-end speaker sound signal may be recorded in a normal indoor environment. In this embodiment, the speech timing between the near-end speaker sound represented by the near-end speaker sound signal and the far-end speaker sound represented by the far-end speaker sound signal (that is, when the near-end speaker sound is uttered). There is no restriction on the relative time when the far-end speaker voice is uttered with respect to (for example, the fogging of the near-end talker voice and the far-end talker voice). However, this does not limit the present invention, and some restrictions may be placed on the speech timing between the near-end speaker speech and the far-end speaker speech. Moreover, there is no restriction | limiting in a near end speaker and a far end speaker, These may be persons other than an evaluator, and at least one of these may be the same person as an evaluator.

Based on the above assumptions, a data structure for performing the above-described evaluation test is generated as follows. The reproduction unit 103 extracts the near-end speaker sound signal data from the near-end speaker sound signal storage unit 101 and outputs the near-end speaker sound signal. The near-end speaker sound signal output from the reproduction unit 103 is sent to the output units 131, 141, 151 and the near-end terminal unit 110. The output units 131, 141, and 151 output the near-end speaker acoustic signals (first acoustic signals on the first end side of the system) to “degraded signal D ₁ ”, “degraded signal D ₂ ”, and “reference signal”, respectively. Rch data (first channel data including the first acoustic signal on the first end side of the system). Further, the near-end terminal unit 110 transmits the sent near-end speaker sound signal to the far-end terminal unit 120 via the network. The far-end terminal unit 120 sends the transmitted near-end speaker sound signal (a signal derived from the first sound signal) to the speaker 105, and the speaker 105 generates a sound represented by the near-end speaker sound signal (second end side of the system). (A reproduction signal derived from the first acoustic signal sent to).

The reproduction unit 104 extracts far-end speaker sound signal data from the far-end speaker sound signal storage unit 102 and outputs a far-end speaker sound signal. The far-end speaker sound signal output from the reproduction unit 104 is sent to the time adjustment processing unit 108 and the speaker 106. The time adjustment processing unit 108 delays the sent far-end speaker sound signal and sends it to the output unit 152. The delay amount τ in the time adjustment processing unit 108 simulates the transmission delay amount B from the far-end terminal unit 120 to the near-end terminal unit 110, and is determined based on the transmission delay amount B, for example. For example, the transmission delay amount B from the far-end terminal unit 120 to the near-end terminal unit 110, the predicted value of the transmission delay amount B, the average value of the transmission delay amount B, or any approximate value or correction value thereof ( (Function value) is the delay amount τ in the time adjustment processing unit 108. The “approximate value of α” means a value belonging to a range of α−β ₁ or more and α + β ₂ or less. β ₁ and β ₂ are positive values (for example, constants), and β ₁ = β ₂ may be satisfied, or β ₁ ≠ β ₂ may be satisfied. Further, the transmission delay amount B is a round-trip delay amount C (a near-end speaker acoustic signal is transmitted from the near-end terminal unit 110 to the far-end terminal unit 120, and a sound representing it is output from the speaker 105. The signal obtained by receiving the sound is about half of the time until the signal is further transmitted from the far-end terminal unit 120 to the near-end terminal unit 110. Therefore, the delay amount τ may be determined based on the delay amount C. For example, ½ value of the delay amount C, ½ value of the predicted value of the delay amount C, ½ value of the average value of the delay amount C, or any one of these function values is used as the delay amount τ. Also good. The delay amount τ may be a fixed value or may be determined based on the actually measured transmission delay amount B. However, depending on the network environment, the amount of delay between the forward path and the return path may be different. Further, if the near-end terminal unit 110, the far-end terminal unit 120, the signal processing unit 121, and the network environment change, the transmission delay amount B and the delay amount C change, so the delay amount τ is determined according to such change. It is desirable. The output unit 152 converts the far-end speaker acoustic signal (reference acoustic signal, second comparison signal based on the second acoustic signal) delayed by the time adjustment processing unit 108 into Lch data (reference acoustic signal) of the “reference signal”. 2nd channel data representing

  The speaker 106 outputs a sound (a reproduction signal derived from the second acoustic signal on the second end side) represented by the transmitted far-end speaker acoustic signal (second acoustic signal on the second end side of the system). The sound output from the speaker 105 and the sound output from the speaker 106 are superimposed in the indoor space and received by the microphone 107. A sound reception signal (a signal based on the first sound signal and the second sound signal) obtained by receiving the sound with the microphone 107 is sent to the signal processing unit 121 of the far-end terminal unit 120. The signal processing unit 121 can control whether or not signal processing is performed on the received sound reception signal. When signal processing is executed, the signal processing unit 121 performs signal processing on the received sound reception signal to obtain a processed signal, and the far-end terminal unit 120 transmits the processed signal to the near-end terminal unit 110 (the first terminal) via the network. 1 end side). For this signal processing, a near-end speaker sound signal (a near-end speaker sound signal input to the speaker 105) transmitted from the near-end terminal unit 110 to the far-end terminal unit 120 via the network is used. Good. On the other hand, when the signal processing is not executed, the far-end terminal unit 120 transmits the received sound signal sent to the signal processing unit 121 to the near-end terminal unit 110 (first end side) via the network. For example, the signal processing unit 121 sends information indicating the presence or absence of signal processing to the recording processing unit 109. In addition, the signal processing unit 121 performs signal processing on the received sound signal to obtain a processed signal, the far-end terminal unit 120 transmits the processed signal to the near-end terminal unit 110 via the network, and The received sound signal that is the same as this signal processing or the received sound signal that can be regarded as the same obtained under the same conditions may be transmitted to the near-end terminal unit 110 via the network. That is, a series of processes when signal processing is performed on one of two received sound signals that can be regarded as the same or the same may be performed, and a series of processes when signal processing is not performed on the other may be performed. The “same condition” means that at least the data generation device 1, the near-end speaker sound signal, the far-end speaker sound signal, and the speech timing are the same. “Signal processing” may be any processing, and an example of “signal processing” is processing including at least one of echo cancellation processing and noise cancellation processing. Note that the echo cancellation processing means processing by an echo canceller in a broad sense for reducing echo. The processing by the echo canceller in a broad sense means all processing for reducing echo. The processing by the broad echo canceller may be realized only by a narrow sense echo canceller using an adaptive filter, may be realized by a voice switch, may be realized by echo reduction, or these It may be realized by a combination of at least some techniques, and may also be realized by a combination with other techniques (for example, “Knowledge Base Knowledge Forest, Group 2-6, Chapter 5,“ Acoustic Echo ”). (See Canceller, IEICE). The noise canceling process means a process for suppressing or removing a noise component caused by any environmental noise other than the voice of the far-end speaker that occurs around the microphone of the far-end terminal. Environmental noise refers to, for example, office air-conditioning sound, in-car sound while driving, car traffic sound at intersections, insect sounds, keyboard touch sounds, voices of multiple people (gray noise), etc. It doesn't matter whether it's large / small or indoor / outdoor.

A signal (a superimposed signal based on a signal derived from the first acoustic signal and a second acoustic signal on the second end side of the system) transmitted from the far end terminal unit 120 via the network is input to the near end terminal unit 110. Are sent to the recording processing unit 109. Here, when the signal processing is performed in the signal processing unit 121 (when the signal processing is ON), the recording processing unit 109 is based on the transmitted signal (the signal derived from the first acoustic signal and the second acoustic signal). The superimposition signal derived from the processed signal obtained by performing signal processing on the signal is sent to the output unit 142. The output unit 142 outputs the transmitted signal (evaluation target acoustic signal T ₂ ) as Lch data (second channel data including a superimposed signal) of the “degraded signal D ₂ ”. On the other hand, when the signal processing is not performed in the signal processing unit 121 (when the signal processing is OFF), the recording processing unit 109 transmits the transmitted signal (the first obtained by sending the received sound signal to the first end side). The comparison signal is sent to the output unit 132. The output unit 132 outputs the transmitted signal (evaluation target acoustic signal T ₁ ) as Lch data (second channel data including a superimposed signal) of the “degraded signal D ₁ ”.

A set of Rch near-end speaker acoustic signal data output from the output unit 131 and Lch evaluation target acoustic signal T ₁ data output from the output unit 132 is stored as “deteriorated signal D ₁ ”. Stored in the unit 180. The pair of the Rch near-end speaker acoustic signal data output from the output unit 141 and the data of the Lch evaluation target acoustic signal T ₂ output from the output unit 142 is stored as “deteriorated signal D ₂ ”. Stored in the unit 180. A set of Rch near-end speaker acoustic signal data output from the output unit 151 and Lch reference acoustic signal data output from the output unit 152 is stored in the data storage unit 180 as a “reference signal”. The The Rch near-end speaker acoustic signals of “deteriorated signal D ₁ ”, “deteriorated signal D ₂ ”, and “reference signal” corresponding to the same time interval are the same. Therefore, it is not always necessary to store the same Rch near-end speaker acoustic signal data in the data storage unit 180 for each of the “degraded signal D ₁ ”, “degraded signal D ₂ ”, and “reference signal”. Of course, for each of the “degraded signal D ₁ ”, “degraded signal D ₂ ”, and “reference signal”, the data of the near-end speaker acoustic signal of the same Rch may be stored in the data storage unit 180.

The “reference signal”, “degraded signal D ₁ ”, and “degraded signal D ₂ ” obtained as described above are illustrated using FIG. In the example of FIG. 3, a series of processing is performed when signal processing is performed on one of the two received sound signals that can be regarded as the same or the same, and a series of processing when signal processing is not performed is performed on the other. Both “degraded signal D ₂ ” when processing is performed and “degraded signal D ₁ ” when signal processing is not performed are obtained. In the example of FIG. 3, processing including echo cancellation processing is used as “signal processing”.

The data structure of the “reference signal” of the present embodiment includes Rch data including the above-mentioned near-end speaker acoustic signal (first channel data including the first acoustic signal on the first end side of the system) and the above-described data structure. Lch data including a reference sound signal based on the far-end speaker sound signal (second channel data including a second comparison signal based on the second sound signal on the second end side). The data structure of the “degraded signal D ₁ ” of the present embodiment includes Rch data including the above-mentioned near-end speaker acoustic signal (first channel data including the first acoustic signal on the first end side of the system), and and a data Lch including evaluation target sound signal T ₁ of the above (data of the second channel including a superimposed signal based on a second audio signal of the second end side of the derived signal and the system to the first acoustic signal) . Evaluated acoustic signal T ₁ is obtained without performing signal processing "first comparison signal". The data structure of the “degraded signal D ₂ ” of the present embodiment includes Rch data including the above-mentioned near-end speaker acoustic signal (first channel data including the first acoustic signal on the first end side of the system), and the containing superimposed signal derived from the Lch data (signals and processing signal obtained by performing signal processing on a signal based on a second sound signal from the first acoustic signal comprising evaluated acoustic signal T ₂ of the above 2 channel data). Note that “the data of the Lch including the evaluation target acoustic signal T ₁ ” and “the data of the Lch including the evaluation target acoustic signal T ₂ ” are both “the signal derived from the first acoustic signal and the second end side of the system”. This corresponds to “second channel data including a superimposed signal based on the second acoustic signal”. In particular, the “Lch data including the evaluation target acoustic signal T ₂ ” is a signal processing based on “a signal derived from the first acoustic signal and a second acoustic signal among the data including the“ superimposed signal ”. Data derived from the processing signal obtained by performing "."

As illustrated in FIG. 3, the same near-end speaker acoustic signal (first acoustic signal) is used in the time interval ab of the Rch data of “reference signal”, “degraded signal D ₁ ”, and “degraded signal D ₂ ”. Signal). The time interval ed ′ of the Lch data of “degraded signal D ₁ ” and “degraded signal D ₂ ” includes the acoustic echo component of the near-end speaker acoustic signal. The acoustic echo component is a signal derived from the above-mentioned near-end speaker acoustic signal (a signal derived from the first acoustic signal), but only in the time interval ae (delay amount C) compared to the near-end speaker acoustic signal. There is a delay. This delay amount C is a signal obtained by transmitting a near-end speaker acoustic signal from the near-end terminal unit 110 to the far-end terminal unit 120, outputting a sound representing it from the speaker 105, and receiving it by the microphone 107. Corresponds to the time until the data is further transmitted from the far-end terminal unit 120 to the near-end terminal unit 110.

The time interval cd of the Lch data of the “reference signal” includes a far-end speaker sound signal component based on the far-end speaker sound signal (a 22nd component based on the second sound signal). The far-end speaker acoustic signal component based on the far-end speaker acoustic signal (the 21st component based on the second acoustic signal) is superimposed on the time interval c′-d ′ of the Lch data of “D ₁ ”. The far-end speaker acoustic signal component based on the far-end speaker acoustic signal (the first component based on the second acoustic signal) is superimposed on the time interval c′-d ′ of the Lch data of the signal D ₂ ″. . There is a time difference a−c ′ from the start time a of the Rch near-end speaker acoustic signal of the “deterioration signal D ₁ ” and the “degradation signal D ₂ ” to the start time c ′ of the Lch far-end speaker sound signal component. To do. Also, there is a time difference a−c from the start time a of the Rch near-end speaker sound signal of the “reference signal” to the start time c of the Lch far-end speaker sound signal component. Here, the time difference a−c ′ between the “degraded signal D ₁ ” and the “degraded signal D ₂ ” is the time difference A between the start timing of the near-end speaker acoustic signal and the start timing of the far-end speaker acoustic signal, This corresponds to the sum A + B of the transmission delay amount B from the far end terminal unit 120 to the near end terminal unit 110. On the other hand, the time difference ac in the “reference signal” corresponds to the sum A + τ of the time difference A and the delay amount τ in the time adjustment processing unit 108. Since the delay amount τ is determined based on the transmission delay amount B as described above, the delay amount τ and the transmission delay amount B match or approximate, and the time difference a−c matches or approximates the time difference a−c ′. Can be made. In the evaluation test using such a data structure, the time from the output of the near-end speaker acoustic signal at the Rch of the “degraded signal D ₂ ” to the output of the far-end speaker acoustic signal component at the Lch, The time from when the near-end speaker acoustic signal is output at the Rch of the “reference signal” to when the far-end speaker acoustic signal component is output at the Lch can be matched or approximated. Similarly, the time from the output of the near-end speaker acoustic signal on the Rch of the “degraded signal D ₁ ” to the output of the far-end speaker acoustic signal component on the Lch and the near-end talk on the Rch of the “reference signal” It is possible to match or approximate the time from the output of the speaker audio signal to the output of the far-end speaker audio signal component on the Lch. Furthermore, the time from the output of the near-end speaker acoustic signal at the Rch of the “degraded signal D ₁ ” to the output of the far-end speaker acoustic signal component at the Lch, and the near-end at the Rch of the “degraded signal D ₂ ”. It is possible to match or approximate the time from when the speaker acoustic signal is output to when the far-end speaker acoustic signal component is output at Lch. That is, the superimposed signal includes a first component based on the second acoustic signal, and the comparison signal includes a second component (21st component or 22nd component) based on the second acoustic signal, and the first channel uses the first component. The time from the output of the acoustic signal to the output of the first component on the second channel and the time from the output of the first acoustic signal on the first channel to the output of the second component on the second channel , Can be matched or approximated. FIG. 3 illustrates the situation where the near-end speaker speaks before the far-end speaker, but the far-end speaker speaks before the near-end speaker, or the time difference is ac−c ′. In some cases, ≈0. For example, the time difference A between the start timing of the near-end speaker sound signal and the start timing of the far-end speaker sound signal, and the transmission delay amount B until the signal is transmitted from the far-end terminal unit 120 to the near-end terminal unit 110 May be equal to time difference a−c ′ = difference A−B≈0. Further, when the far-end speaker starts speaking to the near-end speaker earlier than the transmission delay amount B, the positional relationship of the waveforms is reversed, and the start time c ′ of the Lch far-end speaker acoustic signal component is “ In some cases, the deterioration signal D ₁ ”“ deterioration signal D ₂ ”may be before the start time a of the Rch near-end speaker acoustic signal. Even in such a case, the time adjustment can be similarly performed.

In the above data structure, the Rch near-end speaker acoustic signal data and the Lch reference acoustic signal data are associated as the “reference signal”, and the Rch near-end speaker is represented as the “degraded signal D ₁ ”. The acoustic signal data and the Lch evaluation target acoustic signal T ₁ data are associated with each other, and the Rch near-end speaker acoustic signal data and the Lch evaluation target acoustic signal T ₂ data are represented as “degraded signal D ₂ ”. Are associated. In an evaluation test using such a data structure, while outputting a near-end speaker sound signal at Rch and outputting a reference sound signal at Lch, and outputting a near-end speaker sound signal at Rch, Lch in it is possible to perform the control for outputting the evaluated acoustic signal T _1. Similarly, while outputs the near-end talker audio signals Rch, and a control for outputting a reference sound signal Lch, while outputs the near-end talker audio signals Rch, outputs the evaluated acoustic signal T ₂ in Lch Control can also be performed. Furthermore, while outputs the near-end talker audio signals Rch, and a control for outputting the evaluated acoustic signal T ₁ in Lch, while outputs the near-end talker audio signals Rch, evaluated by Lch target sound signal T ₂ Can also be controlled. That is, a control for outputting a comparison signal on the second channel while outputting the first acoustic signal on the first channel, and a superimposed signal on the second channel while outputting the first acoustic signal on the first channel Control is possible.

In the evaluation test, “reference signal”, “deteriorated signal D ₁ ”, and “deteriorated signal D ₂ ” are reproduced in some order. The reproduced sound of the Rch signal of “reference signal”, “degraded signal D ₁ ”, and “degraded signal D ₂ ” is output from, for example, the right speaker of the binaural-type sound reproducing device, and the reproduced sound of the Lch signal is For example, the sound is output from the left speaker of this binaural-mounted sound reproduction device (stereo reproduction). The evaluator wears the binaural sound reproducing apparatus on both ears and listens to these sounds reproduced in stereo to subjectively evaluate the call quality. At this time, the evaluator preferably listens to the reproduced sound of the Lch signal with the dominant ear (for example, the left ear) and listens to the reproduced sound of the Rch signal with the ear that is not the dominant ear (for example, the right ear). Details of the evaluation test will be described in a third embodiment.

[Modification of First Embodiment]
In the first embodiment, the far-end speaker sound signal delayed by the delay amount τ is used as the “reference signal” Lch standard sound signal. This is between the “reference signal” and the “degraded signal D ₁ ” and “degraded signal D ₂ ”, when the near-end speaker acoustic signal (Rch) starts and when the far-end speaker acoustic signal component (Lch) starts. This is for making the time interval between and coincide with each other (for example, coincidence or approximation between the time interval ac and the time interval ac ′ in FIG. 3). However, such an object can be realized by other means. For example, the far-end speaker sound output from the playback unit 104 is output from the output unit 152 as the Lch standard sound signal of the “reference signal” without delaying the far-end speaker sound signal output from the playback unit 104. A signal obtained by raising the signal by time τ (a signal shifted by a time shift opposite to the delay) may be used as the Rch near-end speaker acoustic signal of the “reference signal”. Alternatively, the far-end speaker sound signal output from the playback unit 104 is delayed by the time τ-T and output from the output unit 152 as the Lch standard sound signal of the “reference signal” and output from the playback unit 103. Alternatively, the near-end speaker sound signal of the Rch of the “reference signal” may be obtained by raising the near-end speaker sound signal by time T. However, the value of T is, for example, 0 ≦ T ≦ τ. Alternatively, by the processing during the evaluation test, the near-end speaker acoustic signal (Rch) starts and the far-end speaker acoustic signal between the “reference signal” and the “degraded signal D ₁ ” and the “degraded signal D ₂ ”. It may be a data structure that can match or approximate the time interval from the start of the component (Lch). For example, a data structure having file names of “reference signal”, “degraded signal D ₁ ”, and “degraded signal D ₂ ” and time information of signals constituting them may be used. The data structure may further have information for specifying the delay amount τ. In such a case, between the “reference signal” and the “degraded signal D ₁ ” and “degraded signal D ₂ ” stored in the data storage unit 180, when the near-end speaker acoustic signal (Rch) starts and far The time interval from the start of the end speaker audio signal component (Lch) may not be the same or approximated. In short, in some way, between the “reference signal” and the “degraded signal D ₁ ” and “degraded signal D ₂ ”, the beginning of the near-end speaker acoustic signal (Rch) and the far-end speaker acoustic signal component ( Any data structure that can match or approximate the time interval from the start of (Lch) can be used. Further, depending on the environment, between the “reference signal” and the “degraded signal D ₁ ” and “degraded signal D ₂ ”, the near-end speaker acoustic signal (Rch) starts and the far-end speaker acoustic signal component (Lch). The evaluation test may be performed without adjusting the time interval between the start of the first time and the first time. In such a case, between the “reference signal” and the “degraded signal D ₁ ” and “degraded signal D ₂ ”, the near-end speaker acoustic signal (Rch) is started and the far-end speaker acoustic signal component ( The data structure may not be able to match or approximate the time interval from the start of (Lch). Further, the time between the start of the near-end speaker acoustic signal (Rch) and the start of the far-end speaker acoustic signal component (Lch) between the “deteriorated signal D ₁ ” and the “degraded signal D ₂ ”. A data structure in which the sections do not match may be used.

[Second Embodiment]
The second embodiment is a modification of the first embodiment, and is a data generation device that electrically simulates a communication environment and an indoor environment, and generates a data structure for performing an evaluation test. Below, it demonstrates centering on the difference with the matter demonstrated so far. About the already demonstrated matter, the reference number used for them is diverted and description is simplified.

<Data generation device>
As illustrated in FIG. 4, the data generation device 2 of the present embodiment includes a near-end speaker acoustic signal storage unit 101, a far-end speaker acoustic signal storage unit 102, a time adjustment processing unit 208, and a communication environment simulation processing unit 260. , A signal processing unit 270, output units 131, 132, 141, 142, 151, 152, and a data storage unit 180. The data generation device 2 is a device configured by, for example, one or more general-purpose or dedicated computers capable of processing audio signals executing a predetermined program. Further, a part or all of the processing units may be configured using an electronic circuit that realizes a processing function independently.

  The communication environment simulation processing unit 260 performs communication environment simulation processing that electrically simulates the communication environment and the surrounding environment (space transmission system). The communication environment simulation process includes at least a signal obtained by performing a process including a first time adjustment process on the near-end speaker sound signal (first sound signal), and a far-end speaker sound signal (second sound signal). Includes a process of superimposing a signal obtained by performing a process including the second time adjustment process. Furthermore, the communication environment simulation process may include a process of superimposing at least one of a pseudo echo and a pseudo noise. For example, as illustrated in FIG. 5A, the communication environment simulation processing unit 260 includes time adjustment processing units 264 and 266, a pseudo echo generation unit 265, an addition unit 267, input units 261 and 262, and an output unit 263. Further, the communication environment simulation processing unit 260 may include a pseudo noise source 268. The pseudo noise source 268 is for simulating any environmental noise generated around the microphone of the far end terminal unit other than the voice of the far end speaker.

  The signal processing unit 270 performs predetermined signal processing on the input signal and outputs it. As in the first embodiment, “signal processing” may be any processing, and an example of “signal processing” is processing including at least one of echo cancellation processing and noise cancellation processing. The echo cancellation processing is processing by an echo canceller in a broad sense for reducing echo. For example, as illustrated in FIG. 5B, the signal processing unit 270 includes input units 271 and 272, an output unit 273, an addition unit 274, an adaptive filter 275, and a time adjustment processing unit 276. The signal processing unit 270 may further include a noise removal unit 278 and a multiplication unit 277. 5B, the echo canceller is configured by using the adaptive filter 275. However, the echo canceller may be configured by a voice switch, echo reduction, other techniques, or a combination thereof and the adaptive filter 275.

Next, the data generation process of this embodiment is demonstrated.
As in the first embodiment, first, as a pre-process, the data of the near-end speaker sound signal (first sound signal) is stored in the near-end speaker sound signal storage unit 101, and the far-end speaker sound signal (second sound) is stored. Signal) data is stored in the far-end speaker sound signal storage unit 102. Based on the above assumptions, a data structure for performing the above-described evaluation test is generated as follows.

  Near-end speaker sound signals are extracted from the near-end speaker sound signal storage unit 101 and sent to the output units 131, 141, 151, the input unit 262 of the communication environment simulation processing unit 260, and the input unit 272 of the signal processing unit 270. It is done. The far-end speaker sound signal is extracted from the far-end speaker sound signal storage unit 102 and input to the time adjustment processing unit 208 and the input unit 261 of the communication environment simulation processing unit 260.

The output units 131, 141, and 151 output the RCH data (first signal) of the “deterioration signal D ₁ ”, “deterioration signal D ₂ ”, and “reference signal” to the transmitted near-end speaker acoustic signal (first acoustic signal), respectively. 1st channel data including an acoustic signal).

  The communication environment simulation processing unit 260 uses the “communication environment simulation process” described above for the far-end speaker acoustic signal (second acoustic signal) and the near-end speaker acoustic signal (first acoustic signal) input to the input units 261 and 262. And the simulation signal obtained thereby is output from the output unit 263. In the case of the example of FIG. 5A, the far-end speaker sound signal input to the input unit 261 is input to the time adjustment processing unit 266, and the near-end speaker sound signal input to the input unit 262 is input to the time adjustment processing unit 264. Entered. The time adjustment processing unit 266 gives a delay amount B ′ to the far-end speaker sound signal, and sends the signal obtained thereby to the addition unit 267 (first time adjustment processing). The time adjustment processing unit 264 gives a delay amount C ′ to the near-end speaker sound signal, and sends the delayed near-end speaker sound signal to the pseudo echo generation unit 265 (second time adjustment process). The pseudo echo generation unit 265 creates a pseudo echo using the delayed near-end speaker sound signal (for example, reproduces the near-end speaker sound signal (first sound signal) on the far-end speaker side speaker. Then, a signal that simulates the spatial transmission system and the waveform distortion at the time of sound collection when the sound is collected by the microphone on the far end speaker side is generated as a pseudo echo), and the signal obtained thereby is sent to the adder 267. The adder 267 superimposes the signal obtained by the first time adjustment process and the signal obtained by the second time adjustment process. When the pseudo noise source 268 exists, the adding unit 267 may further superimpose the pseudo noise signal output from the pseudo noise source 268. The signal obtained by the adding unit 267 is sent to the output unit 263, and the output unit 263 outputs it as a simulation signal.

  Note that the delay amount B ′ described above simulates the transmission delay amount B (transmission delay amount from the far-end terminal unit 120 to the near-end terminal unit 110) of the first embodiment, for example. On the other hand, the delay amount C ′ is, for example, the delay amount C of the first embodiment (a signal is transmitted from the near-end terminal unit 110 to the far-end terminal unit 120, and a sound representing it is output from the speaker 105. Time until the signal obtained by receiving the sound is further transmitted from the far-end terminal unit 120 to the near-end terminal unit 110). Therefore, it is desirable that B ′ <C ′ (for example, C ′ = 2 × B ′). However, this is not a limitation of the present invention, and B ′ = C ′, B ′> C ′ or B ′ = C ′ = 0.

The simulation signal output from the output unit 263 is input to the output unit 132 and the input unit 271 of the signal processing unit 270. The output unit 132 outputs the transmitted simulation signal (evaluation target acoustic signal T ₁ , first comparison signal) as Lch data (second channel data including a superimposed signal) of the “deterioration signal D ₁ ”.

The signal processing unit 270 uses the simulated signal input to the input unit 271 and the near-end speaker acoustic signal input to the input unit 272 to perform signal processing on the simulated signal to obtain a superimposed signal. In the case of the example of FIG. 5B, the signal obtained by applying the adaptive filter 275 to the signal obtained by delaying the near-end speaker acoustic signal by the time adjustment processing unit 276 and the simulated signal are superimposed by the adding unit 274. When canceling processing is performed and the noise removing unit 278 and the multiplying unit 277 are included, noise canceling processing is further performed, thereby obtaining a superimposed signal. Note that the noise cancellation processing method is, for example, that noise estimation is performed on the steady noise level of the pseudo noise transmitted from the pseudo noise source 268 in FIG. 5A in the state where neither the near-end speaker nor the far-end speaker has an acoustic signal. The multiplication unit 277 multiplies the gain value by the multiplication unit 277 so that the amplitude is suppressed by the estimated steady noise level for the output signal from the addition unit 274. U, Yoichi Haneda, Masafumi Tanaka, Junko Sasaki, Akitoshi Kataoka, “Acoustic Echo Canceller with Noise Suppression and Echo Suppression”, IEICE Transactions Vol.J87-A, No.4, pp.448-457 (See April 2004)). The obtained superimposed signal is output from the output unit 273. The output unit 273 sends a superimposed signal (a superimposed signal derived from a processed signal obtained by performing signal processing on a signal based on the signal derived from the first acoustic signal and the second acoustic signal) to the output unit 142. The output unit 142 outputs the transmitted superimposed signal (evaluation target acoustic signal T ₂ ) as Lch data (second channel data including the superimposed signal) of the “degraded signal D ₂ ”.

  Further, the time adjustment processing unit 208 delays the input far-end speaker sound signal by the delay amount τ ′, and sends the delayed far-end speaker sound signal to the output unit 152. The delay amount τ ′ in this embodiment corresponds to, for example, the delay amount B ′ described above. For example, the delay amount τ ′ is the delay amount B ′ or an approximate value or correction value (function value) of the delay amount B ′. Alternatively, the delay amount τ ′ may correspond to the delay amount C ′. For example, τ ′ may be a function value of C ′ / 2 or C ′ / 2. Alternatively, the delay amount τ ′ may correspond to the delay amount B ′ and the delay amount C ′. The output unit 152 converts the far-end speaker acoustic signal (reference acoustic signal, second comparison signal based on the second acoustic signal) delayed by the time adjustment processing unit 208 into Lch data (reference acoustic signal) of the “reference signal”. 2nd channel data representing

  The data structure as illustrated in FIG. 3 can also be obtained by the above processing. The obtained data structure is stored in the data storage unit 180.

[Modification of Second Embodiment]
In the second embodiment, the near-end speaker acoustic signal (denoted between the “reference signal”, “degraded signal D ₁ ”, and “degraded signal D ₂ ”) by the delay processing of each of the time adjustment processing units 208, 264, 266, 276. Rch) coincides with or approximates the time interval between the start of the far-end speaker acoustic signal component (Lch) (coincidence or approximation of the time interval ac and the time interval ac ′ in FIG. 3). ) However, like the modification of the first embodiment, such an object can be realized by other means. For example, the far-end speaker sound signal read from the far-end speaker sound signal storage unit 102 is output from the output unit 152 as the Lch standard sound signal of the “reference signal” without delay, and the near-end speaker sound signal is output. The near-end speaker sound signal of the Rch of the “reference signal” may be obtained by temporally raising the near-end speaker sound signal read from the signal storage unit 101 by the time τ ′. In short,
(1) After the Rch near-end speaker acoustic signal (first acoustic signal) of “deteriorated signal D ₂ ” is output, the far-end speaker included in the evaluation target acoustic signal T ₂ (superimposed signal) of the Lch Included in the reference sound signal of the Lch after the time until the sound signal component (first component) is output and the Rch near-end speaker sound signal (first sound signal) of the “reference signal” is output Match or approximate the time until the far-end speaker acoustic signal component (the 22nd component) is output, and
(2) After the Rch near-end speaker acoustic signal (first acoustic signal) of the “deteriorated signal D ₁ ” is output, the far-end speaker acoustic signal component included in the evaluation target acoustic signal T ₁ of the Lch ( The time until the (21st component) is output, and the far-end speech included in the Lch reference acoustic signal after the Rch near-end speaker acoustic signal (first acoustic signal) of the “reference signal” is output Match or approximate the time until the human acoustic signal component (the 22nd component) is output,
One or more time adjustment processing units that perform at least one of the above may be provided. In addition, by the processing at the time of the evaluation test, between the “reference signal”, the “degraded signal D ₁ ”, and the “degraded signal D ₂ ”, the near-end speaker acoustic signal (Rch) starts and the far-end speaker acoustic signal component A data structure that can match or approximate the time interval from the start of (Lch) may be used. The point is that the start of the near-end speaker acoustic signal (Rch) and the far-end speaker acoustic signal component (Lch) between the “reference signal”, “degraded signal D ₁ ”, and “degraded signal D ₂ ” by some method. Any data structure capable of matching or approximating the time interval between the start and the start of the data may be used. Furthermore, depending on the environment, between the “reference signal”, the “degraded signal D ₁ ”, and the “degraded signal D ₂ ”, the start of the near-end speaker acoustic signal (Rch) and the start of the far-end speaker acoustic signal component (Lch) An evaluation test may be performed without adjusting the time interval between times. In such a case, between the “reference signal”, “degraded signal D ₁ ”, and “degraded signal D ₂ ”, the near-end speaker acoustic signal (Rch) starts and the far-end speaker acoustic signal component (Lch). It may be a data structure in which it is impossible to match or approximate the time interval between the start time of and the start time.

[Third Embodiment]
In the third embodiment, a quality evaluation method using the data structure generated as described above will be described.

<Sound quality evaluation device>
As illustrated in FIG. 6, the sound quality evaluation apparatus 3 according to the present embodiment includes a data storage unit 180, a totaling result storage unit 305, a reproduction control unit 301, a display control unit 302, a totaling unit 303, a control unit 304, and a sound output. The processing unit 310-n, the display unit 320-n, and the input unit 330-n are included. However, n = 1,..., N, and N is an integer of 1 or more (for example, N is 1 or more and 4 or less). The sound quality evaluation apparatus 3 is an apparatus configured by, for example, one or more computers including a display device (display, etc.) and an input device (keyboard, mouse, etc.) executing a predetermined program. is there. Further, a part or all of the processing units may be configured using an electronic circuit that realizes a processing function independently.

<Sound quality evaluation process>
The sound quality evaluation apparatus 3 uses the data structure described above and performs an evaluation test that simulates the conversation MOS test in the above-described loudspeaker communication system under the control of the control unit 304.

  For n = 1,..., N, the Rch (first channel: for example, the right channel) that is one channel of the binaural-type sound reproducing device 340-n is connected to the output unit 311-n of the sound output processing unit 310-n. ) Is connected, and the Lch (second channel: for example, the left channel), which is the other channel of the binaural sound reproducing device 340-n, is connected to the output unit 312-n. The binaural-mounted sound reproducing device 340-n includes a speaker dedicated to one ear that outputs sound of one channel Rch and a speaker dedicated to the other ear that outputs sound of the other channel Lch. This is a stereo sound reproduction apparatus equipped with stereo reproduction. Specific examples of the binaural-mounted sound reproducing device 340-n include headphones and earphones. The evaluator 350-n wears the binaural-type sound reproduction device 340-n, and according to the display content output from the display unit 320-n, the sound output from the binaural-type sound reproduction device 340-n. Subjective evaluation is performed, and the evaluation result is input to the input unit 330-n. Note that the evaluator 350-n wears a speaker on the side that outputs the sound of the channel Lch in the dominant ear (for example, the left ear), and the channel Rch in the ear (for example, the right ear) that is not the dominant ear. It is desirable to attach a speaker that outputs sound. Hereinafter, these processes will be described in detail.

The reproduction control unit 301 extracts any one of “reference signal”, “degraded signal D ₁ ”, and “degraded signal D ₂ ” from the data structure described above from the data storage unit 180 in accordance with the control of the control unit 304 (the control content will be described later). And sent to the sound output processing unit 310-n (where n = 1,..., N). At this time, a process for matching or approximating the time interval between the start time of the near-end speaker sound signal (Rch) and the start time of the far-end speaker sound signal component (Lch) may be performed. The sound output processing unit 310-n performs the following processing according to the transmitted signal. Note that the sound represented by the reference acoustic signal of the “reference signal” is referred to as “reference sound”, the sound represented by the evaluation target acoustic signal T _{1 of} the “degraded signal D ₁ ”, and the evaluation target acoustic signal of the “deteriorated signal D ₂ ”. the sound T ₂ represents will be referred to as "evaluation sound".

≪When “reference signal” is sent≫
When the “reference signal” is transmitted, the sound output processing unit 310-n (where n = 1,..., N) transmits the near-end speaker sound signal (first sound signal) of the “reference signal”. While outputting from the output unit 311-n to the Rch (first channel) which is one channel of the binaural-mounted sound reproduction device 340-n, the reference acoustic signal of the “reference signal” is output from the output unit 312-n to both ears. Output to Lch (second channel) which is the other channel of the wearable sound reproducing device 340-n (first processing).

≪When “deterioration signal D ₁ ” is sent≫
When the “degraded signal D ₁ ” is sent, the sound output processing unit 310-n (where n = 1,..., N), the near-end speaker acoustic signal ( _first signal) of the “degraded signal D ₁ ”. while outputting a sound signal) from the output unit 311-n to the Rch (first channel) of two earset sound reproducing apparatus 340-n, evaluated acoustic signal _{T 1} (first acoustic signal of "degraded signal _{D 1"} Is output from the output unit 312-n to the Lch (second channel) of the binaural-type sound reproduction device 340-n (second processing). ).

«If the" deterioration signal D ₂ "has been sent»
When the “degraded signal D ₂ ” is sent, the sound output processing unit 310-n (where n = 1,..., N) transmits the near-end speaker acoustic signal (first signal) of the “degraded signal D ₂ ”. (Acoustic signal) is output from the output unit 311-n to the Rch (first channel) of the binaural-mounted sound reproducing device 340-n, and the evaluation target acoustic signal T ₂ (first acoustic signal) of the “deteriorated signal D ₂ ” is output. A superimposed signal representing an evaluation sound based on the signal derived from the second acoustic signal and the second acoustic signal, which is obtained by performing signal processing on a signal derived from the signal derived from the first acoustic signal and the second acoustic signal. Is output from the output unit 312-n to the Lch (second channel) of the binaural-type sound reproducing device 340-n (second processing).

  The display control unit 302 sends display information to the display unit 320-n (where n = 1,..., N) in accordance with the control of the control unit 304 (details of control will be described later). The display unit 320-n has three or more levels consisting of a combination of whether or not the difference between the reference sound and the evaluation sound is known and the degree of two or more levels of difficulty in hearing the evaluation sound according to the display information sent. The evaluation category including the category is displayed. The evaluator 350-n subjectively evaluates the sound output from the binaural sound reproduction device 340-n according to this display. Here, the “reference sound” corresponds to an acoustic signal received from the far-end speaker in an ideal state. An ideal state of a loudspeaker communication system can be simulated by presenting it together with a “near-end speaker sound” corresponding to a direct sound from the near-end speaker. By presenting the “near-end speaker sound” at the same time as the “reference acoustic signal”, it becomes easy to distinguish between the near-end speaker's voice wraparound (acoustic echo) and the far-end speaker's voice. By comparing "evaluation sound" with "reference sound" at all times, objectively and subjectively evaluate how close or different the communication system is to be evaluated is. can do. When only the “evaluation sound” is presented and evaluated, the far-end speaker's stagnation, the far-end speaker's ambient noise, etc. are judged as degradation factors and are likely to be evaluated low. By always comparing with the “reference sound”, deterioration factors other than the communication system are excluded from the evaluation target, and an accurate evaluation value with little variation can be obtained. This evaluation category defines not only the deterioration of the evaluation sound with respect to the reference sound but also the evaluation standard for difficulty in hearing the evaluation sound (easy to hear). In this way, by displaying an evaluation category that combines the degree of deterioration of the evaluation sound from the reference sound and the degree of ease of hearing, an evaluation category that focuses only on deterioration, such as conventional DCR (deterioration category evaluation), is displayed. Compared to the case, it becomes clear what criteria should be used for evaluation, and the evaluation variation can be reduced even in an environment where a plurality of factors are intertwined in a complicated manner. In addition, by displaying the evaluation standard (negative evaluation standard) for the evaluation sound listening “Nikusa”, the evaluation standard (positive evaluation standard) for the evaluation sound listening “ease” is displayed. In comparison with the above, the selection of the evaluator 350-n becomes stricter, and the evaluation accuracy is improved. This is based on the natural laws of physiology.

Preferably, the evaluation category includes a category of four or more levels composed of a combination of whether or not a difference between the reference sound and the evaluation sound is known and a degree of three or more levels of difficulty in hearing the evaluation sound. The evaluation accuracy can be further improved by determining the evaluation criteria for the degree of three or more levels of difficulty in hearing the evaluation sound. In particular, the evaluation category is a one-step category indicating that the difference between the reference sound and the evaluation sound is not known, and a four-step degree indicating that the difference between the reference sound and the evaluation sound can be understood and the evaluation sound is difficult to hear. It is desirable to include a four-stage category consisting of Specific examples of evaluation categories are shown below.
In addition, “I don't know the difference from the reference sound”, “I have a difference” or “I have a difference” means “I can understand the difference between the reference sound and the evaluation sound”, and “There is no problem with listening” “Difficult to hear a little”, “Difficult to hear” and “Very difficult to hear” represent “degree of difficulty in hearing the evaluation sound”. Each evaluation category in this example is associated with a value representing an evaluation of 1 to 5, and the larger this value, the higher the quality. Here, the category is set assuming that the “reference sound” is in an ideal state, but the “evaluation sound” has a higher evaluation than the “reference sound” due to the effect of the noise canceller of the communication system to be evaluated. Is also possible. In this case, “there is a difference, but easy to hear” may be included as a higher category.

The following is an evaluation category focusing only on the degradation used in the conventional DCR (degradation category evaluation). It can be seen that there are more subjective and internal expressions than the evaluation categories in Table 1.

Furthermore, the display information output by the display control unit 302 includes information for instructing evaluation of the ease of hearing of the evaluation sound, and the display unit 320-n further instructs evaluation of the ease of hearing of the evaluation sound. (Display indicating “what to evaluate”) may be performed. For example, the display unit 320-n may display “Please rate the ease of hearing of the evaluation sound“ female voice (left side) ””. In this example, the left side indicates the output of the speaker on the Lch (second channel) side in the “reference signal”, “degraded signal D ₁ ”, and “degraded signal D ₂ ”. As described above, the evaluation category includes a combination of whether or not the difference between the reference sound and the evaluation sound is known and the degree of difficulty in hearing the evaluation sound. Physiologically, humans are sensitive to the difference and can evaluate the difference between the reference sound and the evaluation sound without particular attention. On the other hand, appropriate evaluation cannot be performed unless attention is paid to the ease of hearing. Based on such a natural law, the display part 320-n can further improve the evaluation accuracy or reduce the evaluation variation by performing display for instructing the evaluation of the ease of hearing of the evaluation sound. In addition, when the “display for instructing the evaluation of listening to the evaluation sound“ Nikusa ”” is performed as a display indicating what is evaluated, the evaluator 350-n pays too much attention to details in terms of physiology. Therefore, there is a tendency to evaluate even a small deterioration having an influence on “easy to hear”. As a display indicating what is to be evaluated, “display for instructing evaluation of listening to“ ease of evaluation sound ”” makes evaluation of the evaluator 350-n appropriate and can improve evaluation accuracy, Evaluation variation can be reduced.

Further, the display information output by the display control unit 302 may include information for displaying what is focused on, and the display unit 320-n may display “what to focus on”. For example, the display unit 320-n performs a display indicating an instruction to pay attention to the reference sound at the time of the above-described “first process”, and an instruction to pay attention to the evaluation sound at the time of the “second process”. You may perform the display to represent. For example, the display unit 320-n displays “reference sound (1): pay attention to“ female voice (left side) ”during“ first processing ”and outputs“ deterioration signal D ₁ ”. When “second processing” is performed, “evaluation sound (1): pay attention to“ female voice (left side) ”is displayed and“ deterioration signal D ₂ ”is output. "Evaluation sound (2): Pay attention to" Female voice (left side) "" may be displayed. As a result, the evaluation target is clarified so that the evaluator 350-n is focused on the evaluation target acoustic signal (far-end speaker acoustic signal side) and the evaluator 350-n is not focused on the near-end speaker acoustic signal side. Can be. In addition, depending on the signal output from the sound output processing unit 310-n, the display of “what to focus on” and “what to evaluate” displayed from the display unit 320-n is changed. The generation timing of the acoustic signal can be visually recognized.

  The evaluator 350-n who performed the subjective evaluation inputs an evaluation value In, which is information representing the category selected from the evaluation categories (information representing the evaluation result), to the input unit 330-n. FIG. 7 illustrates a display screen 321 displayed by the display unit 320-n. The display screen 321 includes an attention content presentation unit 3211 that displays “what to focus on”, an evaluation instruction presentation unit 3212 that displays “what to evaluate”, an evaluation category presentation unit 3213 that displays an evaluation category, and an evaluation The icons 3214 to 3218 that are touched or clicked to input the values “1” to “5” (evaluation value In) that represent the values “3” and the icon 3219 that is touched or clicked to confirm the input are included. The evaluator 350-n subjectively evaluates the sound output from the binaural-equipped sound reproduction device 340-n according to the display of the attention content presentation unit 3211, the evaluation instruction presentation unit 3212, and the evaluation category presentation unit 3213, and evaluates it. One of the corresponding icons 3214 to 3218 is touched or clicked, and the icon 3219 for confirmation is touched or clicked. Until the icon 3214 to 3219 is active and the icon 3219 is touched or clicked, the evaluator 350-n can perform a touch or click operation to reselect the icons 3214 to 3218 many times. Thereby, the evaluation value In representing the category selected from the evaluation categories is input to the input unit 330-n. In addition, in order to make evaluation conditions the same, it is desirable that the above-described evaluation test is simultaneously executed by all the evaluators 350-n (where n = 1,..., N). When there is an evaluator who does not confirm the evaluation for a certain time or more, a screen display that prompts the evaluator to confirm and a screen display that waits for other evaluators may be displayed.

The evaluation value In input to the input unit 330-n is sent to the counting unit 303. The tabulation unit 303 tabulates the evaluation value In and stores the tabulation result obtained thereby in the tabulation result storage unit 305. For example, the tabulation result is stored together with an ID representing the evaluator 350-n, an acoustic signal such as “deterioration signal D ₂ ” used in the evaluation test, and its conditions. The aggregation result of the evaluation values In may be a set of the evaluation values In, or may be a maximum value, a minimum value, an average value, a variance value, etc. for each acoustic signal used in the evaluation test. Also good. The maximum value, the minimum value, the average value, the variance value, and the like obtained after excluding the evaluation value In corresponding to the evaluator 350-n whose suspicion is in the evaluation content may be used as the aggregation result. In addition, further detailed analysis may be performed by another processing apparatus.

<< Control contents of control unit 304 >>
Next, the control contents of the control unit 304 will be illustrated using FIGS. 8 to 12. The horizontal axis of these figures represents the time axis, and represents the later time as it goes to the right of the page. In these figures, the “Lch” row represents the sound output from the speaker on the Lch side of the binaural-mounted sound reproducing device 340-n, and the “Rch” row represents the binaural-mounted sound reproducing device 340-n. Represents the sound output from the speaker on the Rch side. In these figures, the column “3211” represents the presentation content of the focus content presentation unit 3211 (what to focus on), and the column “3212” represents the content of the evaluation instruction presentation unit 3212 (what to evaluate). The column “3213” represents the presentation content (evaluation category) of the evaluation category presentation unit 3213.

≪Example of FIG. 8≫
In the example of FIG. 8, the reproduction control unit 301 first reads a “reference signal” from the data storage unit 180 and sends it to the sound output processing unit 310-n (where n = 1,..., N). The sound output processing unit 310-n outputs the reference sound signal of the “reference signal” from the output unit 312-n, and outputs the near-end speaker sound signal of the “reference signal” from the output unit 311-n. As a result, the “reference sound” represented by the reference acoustic signal is output from the Lch of the binaural-mounted sound reproducing device 340-n, and the “near-end speaker sound” corresponding to the direct sound from the near-end speaker is output from the Rch. Is output. At this time, the display control unit 302 sends a display information indicating the attention content _{F 1} and evaluation categories on the display unit 320-n. Note that the focus content F ₁ means content indicating an instruction to focus on the reference sound (Lch) (for example, “focus on the reference sound (1):“ female voice (left side) ”). In addition, the evaluation category includes an evaluation including three or more categories consisting of a combination of the above-mentioned “whether or not the difference between the reference sound and the evaluation sound is known and the degree of difficulty of hearing the evaluation sound in two or more levels. Category ". Display unit 320-n presents the focused content _{F 1} to the target content presentation unit 3211 presents the evaluation category rating category presentation unit 3213 (step S1).

Next, the reproduction control unit 301 reads “deterioration signal D ₂ ” from the data storage unit 180 and sends it to the sound output processing unit 310-n (where n = 1,..., N). Sound output processing unit 310-n outputs the evaluated acoustic signal _{T 2} of the "degraded signal _{D 2} 'from the output unit 312-n, the near-end speaker sound" degraded signal _{D 2'} from the output unit 311-n Output a signal. As a result, the “evaluation sound” represented by the evaluation target sound signal T ₂ of the “deterioration signal D ₂ ” is output from the Lch of the binaural-mounted sound reproduction device 340-n, and the near-end speaker sound signal is output from the Rch. A “near-end speaker sound” is output. At this time, the display control unit 302 sends the attention content F ₂ , the evaluation instruction S ₁ , and display information representing the evaluation category to the display unit 320-n. In addition, attention contents F _2, the contents (for example, "evaluation sound (1):" female voice (please focus on the left side), ""), which represents an instruction to focus on the evaluation sound (Lch) means. The evaluation instruction S ₁ means an instruction for evaluating the ease of hearing of the evaluation sound (Lch) (for example, “evaluate the ease of hearing of the“ female voice (left side) ”of the evaluation sound”). Display unit 320-n presents the focused content _{F 2} to the target content presentation unit 3211 presents the evaluation instruction _{S 1} to the evaluation instruction presentation unit 3212 presents the evaluation category rating category presentation unit 3213 (step S2) .

  Next, step S1 is executed once again (step S3), and step S2 is executed again (step S4). Step S1 and step S2 may be repeated three or more times.

  Thereafter, the icons 3214 to 3219 are activated, and the evaluation value In and the input of confirmation are received from the input unit 330-n (step S5).

Furthermore, even if the process in which “degraded signal D ₂ ” in steps S1 to S5 is replaced with “degraded signal D ₁ ” and “evaluation target acoustic signal T ₂ ” is replaced with “evaluation target acoustic signal T ₁ ” is executed. Good. In addition, the presentation of the evaluation category by the evaluation category presentation unit 3213 may be continuously performed through steps S1 to S5, or the presentation of the evaluation category may disappear every time each step is completed.

≪Example of FIG. 9≫
In the example of FIG. 9, among the “reference sound”, the “evaluation sound” represented by the evaluation target acoustic signal T ₁ , and the “evaluation sound” represented by the evaluation target acoustic signal T ₂ , a pair of sounds to be compared are randomly selected. Select and output the selected sounds in order.

A specific example of processing is shown below.
First, the reproduction control unit 301 randomly selects a pair to be compared from “reference signal”, “degraded signal D ₁ ”, and “degraded signal D ₂ ”. Examples of sets to be compared are a set of “reference signal” and “degraded signal D ₁ ”, a set of “reference signal” and “degraded signal D ₂ ”, “degraded signal D ₁ ”, and “degraded signal D”. ₂ ”. Of the signals constituting the pair to be compared, a signal output first is called a “first output signal”, and a signal output later is called a “second output signal”. Any of the signals constituting the pair to be compared may be output first. For example, when comparing a set of “reference signal” and “degraded signal D ₁ ”, “reference signal” may be “first output signal” and “degraded signal D ₁ ” may be “second output signal”. Alternatively, the “reference signal” may be the “second output signal” and the “deterioration signal D ₁ ” may be the “first output signal”.

Next, the “reference sound or evaluation sound” corresponding to the “first output signal” is output from the Lch, and the “near-end speaker sound” corresponding to the “first output signal” is output from the Rch (step S21). ). The process of step S21 when the “first output signal” is the “reference signal” is the same as the above-described step S1. Step S21 if the "first output signal" is "degraded signal D _2" except that it does not provide an evaluation instruction S ₁ to the evaluation instruction presentation unit 3212 is the same as step S2 described above. When the “first output signal” is “degraded signal D ₁ ”, the process of step S21 replaces “degraded signal D ₂ ” with “degraded signal D ₁ ” in the process of step S2 described above, This is a process in which the “acoustic signal T ₂ ” is replaced with “evaluation target acoustic signal T ₁ ” and the evaluation instruction S ₁ is not presented to the evaluation instruction presentation unit 3212.

Next, the “reference sound or evaluation sound” corresponding to the “second output signal” is output from the Lch, and the “near-end speaker sound” corresponding to the “second output signal” is output from the Rch (step S22). . Processing in step S22 in case the "second output signal" is "reference signal" is intended to addition to the step S1 described above carries out a process of presenting the evaluation instruction S ₁ to the evaluation instruction presentation unit 3212. The process of step S21 when the “second output signal” is “degraded signal D ₂ ” is the same as step S2 described above. When the “second output signal” is “degraded signal D ₁ ”, the process of step S21 is performed by replacing “degraded signal D ₂ ” with “degraded signal D ₁ ” in the process of step S2 described above. an acoustic signal T ₂ "is a substituted processed" evaluated sound signals T ₁ ".

  Finally, the evaluation value is input and confirmed (step S5).

In addition, as a modified example of steps S21 and S22, it may not indicate whether the sound output from the Lch is a “reference sound” or an “evaluation sound”. That is, instead of the focus content F ₁ and the focus content F ₂ , content indicating an instruction to focus on Lch (eg, “focus on“ female voice (left side) ”) may be presented. In this case, the evaluator 350-n performs the subjective evaluation without being notified whether the presented sound is the “reference sound” or the “evaluation sound”.

≪Example of FIG. 10≫
In the example of FIG. 10, the “reference sound” is output at the first time, and the “evaluated sound” or the evaluation target sound signal T represented by the “hidden reference sound” or the evaluation target sound signal T ₁ at the second time and the third time, respectively. _The “evaluation sound” represented by ₂ is output. Here, when the “hidden reference sound” is output for the second time, the “evaluation sound” represented by the evaluation target acoustic signal T ₁ or the “evaluation sound” represented by the evaluation target acoustic signal T ₂ is output for the third time. (Pattern 1). On the other hand, when the “evaluation sound” represented by the evaluation target acoustic signal T ₁ or the “evaluation sound” represented by the evaluation target acoustic signal T ₂ is output for the second time, the “hidden reference sound” is output for the third time. (Pattern 2). The “hidden reference sound” means a “reference sound” that is output without indicating that it is a “reference sound”. Whether to use pattern 1 or pattern 2 is determined randomly.

  A specific example of processing is shown below.

  First, the “reference sound” corresponding to the “reference signal” is output from the Lch, and the “near-end speaker sound” corresponding to the “reference signal” is output from the Rch (step S31). The process in step S31 is the same as that in step S21 described above.

Next, the playback control unit 301 randomly selects pattern 1 or pattern 2.
When the pattern 1 is selected, first, the “hidden reference sound” corresponding to the “reference signal” is output from the Lch, and the “near-end speaker sound” corresponding to the “reference signal” is output from the Rch ( step S32), then evaluated acoustic signal T ₁ is represented "evaluated sound" or "degraded signal D _2" evaluated sound signal T ₂ represents "evaluated sound" of the "degraded signal D ₁ 'from Lch output Then, “Near-end speaker sound” corresponding to “Deteriorated signal D ₁ ” or “Deteriorated signal D ₂ ” is output from Rch (step S 33).
On the other hand, when the pattern 2 is selected, the “evaluation sound” represented by the evaluation target acoustic signal T ₁ or the “evaluation sound” represented by the evaluation target acoustic signal T ₂ is output from the Lch, and the “deterioration signal D ₁ ” or A “near-end speaker sound” corresponding to the “deterioration signal D ₂ ” is output (step S32), and then a “hidden reference sound” corresponding to the “reference signal” is output from the Lch. A “near-end speaker sound” corresponding to the “reference signal” is output (step S33).

The process of outputting the “hidden reference sound” corresponding to the “reference signal” from the Lch and outputting the “near-end speaker sound” corresponding to the “reference signal” from the Rch is performed instead of the attention content F _2. This is the same as step S1 described above, except that the content F ₁ is presented to the attention content presentation unit 3211 and the evaluation instruction S ₁ is presented to the evaluation instruction presentation unit 3212. Further, “evaluation sound” represented by the evaluation target acoustic signal T ₁ or “evaluation sound” represented by the evaluation target acoustic signal T ₂ is output from the Lch, and corresponds to the “deterioration signal D ₁ ” or “deterioration signal D ₂ ” from the Rch. In the process of outputting the “near-end speaker sound” to be performed, “degraded signal D ₂ ” is replaced with “degraded signal D ₁ ” in the process of step S2 or the process of step S2, and “evaluation target acoustic signal T ₂ ”is the same as the processing in which“ evaluation target acoustic signal T ₁ ”is replaced.

  Finally, the evaluation value is input and confirmed (step S5). However, the evaluator 350-n determines which one of the sounds output in steps S32 and S33 is the evaluation sound, and inputs the evaluation value only for the sound determined to be the evaluation sound. A sound that is not judged as an evaluation sound is automatically regarded as a “hidden reference sound” and is given an evaluation value “5” for the hidden reference sound. Further, the evaluator 350-n may input instructions to the input unit 330-n so that steps S31 to S33 can be executed any number of times in a desired order before step S5.

≪Example of FIG. 11≫
In the example of FIG. 11, the “reference sound” is output at the first time, and the “hidden reference sound” or the evaluation target sound signal T according to the pattern 1 or pattern 2 selected at random for the second time and the third time, respectively. _The “evaluation sound” represented by ₁ or the “evaluation sound” represented by the evaluation target acoustic signal T ₂ is output. However, the evaluation values for the second and third outputs are input (steps S132 and S133), and finally, the final determination value is input (step S105). Note that the evaluator 350-n inputs the evaluation value “5” to the one judged as “hidden reference sound” among the sounds outputted in steps S132 and S133, and judged as “evaluation sound”. Enter your own evaluation value in. Other details are the same as in the example of FIG.

<< Example of FIG. 12 >>
In FIG. 12, the “reference sound” is output for the first time (step S41), and “evaluation sound 1” to “evaluation sound x” for the second to x + 1th time (x is an integer of 3 or more (eg, x is 14 or less)). "Is output (steps S42-1 to S42-x), and an evaluation value is input and confirmed (step S5). The “evaluation sound 1” to “evaluation sound x” are at least one of “evaluation sound” represented by the evaluation target acoustic signal T ₁ and “evaluation sound” represented by the evaluation target acoustic signal T _2. "Reference sound" and one or more "anchor sounds". The “anchor sound” represents a sound that is a reference for bad acoustic quality. When a plurality of anchor sounds are included, a sound quality standard that gradually deteriorates may be used. In step S5, the evaluation values of the sounds output in steps S42-1 to S42-x are input. Further, the output order of “evaluation sound 1” to “evaluation sound x” is determined randomly. However, even if the evaluator 350-n inputs an instruction to the input unit 330-n, the steps S42-1 to S42-x can be executed any number of times in a desired order before step S5. Good. Others are the same as the example of FIG.

[Fourth Embodiment]
The inventor has obtained an evaluation value obtained in the third embodiment (a MOS value based on a five-step evaluation of the difference between the reference sound corresponding to the reference acoustic signal and the evaluation sound corresponding to the evaluation target acoustic signal (exemplified in Table 1). )) And PESQ (the PESQ values corresponding to these reference acoustic signals and evaluation target acoustic signals) can be approximated to a linear relationship. This has not been known in the past (see, for example, “Appendix V-1 / JJ-201.001 <Formulation of relationship between PESQ value and listening MOS value>” in Non-Patent Document 1). In the present embodiment, based on this knowledge, the MOS value can be estimated by a linear calculation with a small amount of calculation without performing complicated subjective evaluation and non-linear calculation with a large amount of calculation from the PESQ by linear calculation. Details will be described below.

  FIG. 14 shows MOS values (DMOS (Degradation MOS) values) obtained by performing an evaluation test using “reference signals” and “degradation signals” as exemplified in the third embodiment, and the corresponding values. It is a graph showing the relationship between the PESQ value obtained from the “reference acoustic signal” and the “evaluation target acoustic signal”. The vertical axis represents the MOS value (DMOS value), and the horizontal axis represents the PESQ value. A small diamond mark represents a measured value obtained by a subjective evaluation test, and a large square mark on the broken line represents an estimated value based on the linear relationship. As shown in this figure, the relationship between the MOS value obtained in the third embodiment and the corresponding PESQ value can be approximated by a linear relationship. Therefore, if a reference function consisting of a prepared reference acoustic signal and an evaluation target acoustic signal is used and a linear function (linear function) representing this linear relationship is formulated, a new reference acoustic signal and an evaluation target The PESQ value is calculated from the acoustic signal, and the MOS value can be calculated by substituting the PESQ value into this linear function.

<Configuration>
As illustrated in FIG. 13, the acoustic quality evaluation device 4 of this embodiment includes a PESQ calculation unit 41 and a linear conversion unit 42. The acoustic quality evaluation apparatus 4 is an apparatus configured by, for example, one or more computers as described above executing a predetermined program. Further, a part or all of the processing units may be configured using an electronic circuit that realizes a processing function independently.

<Pretreatment>
As a pre-processing of the sound quality evaluation process, a set of a reference sound signal and an evaluation target sound signal based on a signal including the reference sound signal is used as a reference signal, and the reference sound signal (second reference sound signal) and the reference sound are used. PESQ value (second PESQ value) corresponding to the evaluation target acoustic signal (second evaluation target acoustic signal) based on the signal including the signal, the reference sound corresponding to the reference acoustic signal, and the evaluation target acoustic signal A linear relationship between the MOS value (second MOS value) based on the five-step evaluation of the difference from the corresponding evaluation sound is obtained in advance. At this time, various combinations of the reference acoustic signal and the evaluation target acoustic signal based on the signal including the reference acoustic signal are performed, and a subjective evaluation test is performed by a plurality of evaluators. And statistically analyze the linear relationship in a manner that reduces dependence on individual evaluator differences. Information obtained as a result of this analysis is a linear relationship between the PESQ value (second PESQ value) and the MOS value (second MOS value) shown in FIG. Information representing such a linear relationship is set in the linear conversion unit 42. Examples of “information representing a linear relationship” are a linear function F representing the linear relationship, a parameter specifying the linear function F, and the like. An example of the linear function F is a function for inputting a PESQ value and outputting a corresponding MOS value, for example, MOS value = α × PESQ value + β. Α and β are parameters.

  The reference sound signal may be a far-end speaker sound signal as exemplified in the first to third embodiments, may be another sound signal, or may be other music or background sound. It may be an acoustic signal. The evaluation target acoustic signal may be anything as long as it is based on a signal including a reference acoustic signal. An example of the evaluation target acoustic signal is a deterioration signal of a signal including a reference acoustic signal, for example, a signal in which at least one of an echo component and a noise component is superimposed on the reference acoustic signal as exemplified in the first to third embodiments. is there.

  The calculation method of the PESQ value in the PESQ calculation unit 41 is well known, and is described in detail in “ITU-T Recommendation P.862”, for example. “Original X (t)” in the description of “ITU-T Recommendation P.862” corresponds to the reference acoustic signal of the present invention, and “degraded signal Y (t)” corresponds to the acoustic signal to be evaluated of the present invention. Note that the normal PESQ value calculation processing includes processing for correcting a time lag between the reference acoustic signal and the evaluation target acoustic signal.

  The MOS value based on the five-step evaluation about the difference between the reference sound corresponding to the reference sound signal and the evaluation sound corresponding to the evaluation target sound signal is, for example, a five-step evaluation on the difference between the received reference sound and the evaluation sound ( This is the average value of subjective evaluation. The five-level evaluation itself is one of five values representing a five-level evaluation category, but the average MOS value is any value in the range of 1 to 5. There is no limitation on the content of “5-level evaluation of the difference between the reference sound and the evaluation sound”. An example of such a five-level evaluation is “for an evaluation category consisting of a combination of whether or not the difference between the reference sound and the evaluation sound is known and the degree of ease of hearing and / or difficulty in hearing the evaluation sound. It is a “5-level evaluation”. In particular, such a five-step evaluation is “a five-step evaluation for an evaluation category consisting of a combination of whether or not the difference between the reference sound and the evaluation sound is known and a four-step degree of difficulty in hearing the evaluation sound”. In this case, a linear relationship with smaller error is established. More preferably, such a five-step evaluation is “in order to understand the difference between the reference sound and the evaluation sound, and the one-step category indicating that the difference between the reference sound and the evaluation sound is unknown”. It is desirable to be a “5-level evaluation for an evaluation category” including a 4-level category consisting of a combination of the 4-level degree of the stiffness. Specific examples of “whether or not the difference between the reference sound and the evaluation sound is known” and “the degree of difficulty in hearing the evaluation sound” are as illustrated in the third embodiment. Specific examples of “degree of ease of hearing of evaluation sound” are “no problem in listening”, “a little easy to hear”, “easy to hear”, “very easy to hear”. Further, it is desirable that the MOS value based on such a five-step evaluation is based on the five-step evaluation obtained by instructing “evaluation of ease of hearing of evaluation sound”. For example, as exemplified in the third embodiment, contents such as “evaluate the ease of hearing of the evaluation sound“ female voice (left side) ”” were presented to the evaluator during the subjective evaluation test. A MOS value based on a five-step evaluation is desirable.

<Sound quality evaluation process>
Based on the above assumptions, the sound quality evaluation process is performed as follows. First, the PESQ calculation unit 41 receives a reference acoustic signal (first reference acoustic signal) and an evaluation target acoustic signal (first evaluation target acoustic signal) based on a signal including the reference acoustic signal, and inputs the reference acoustic signal. A PESQ value (first PESQ value) for the signal and the evaluation target acoustic signal is obtained and output. This PESQ value is input to the linear conversion unit 42. The linear conversion unit 42 linearly converts the input PESQ value based on the linear relationship described above to obtain and output an estimated MOS value (first MOS value). For example, the linear conversion unit 42 outputs the result obtained by substituting the PESQ value into the above-described linear function F as the MOS estimated value.

[Modification 1 of Fourth Embodiment]
As illustrated in FIG. 15, the sound quality evaluation apparatus of this modification includes a PESQ calculation unit 41, a linear conversion unit 42, a far-end speaker sound signal storage unit 102, and a data storage unit 180. The PESQ calculation unit 41 reads the far-end speaker acoustic signal from the far-end speaker acoustic signal storage unit 102 as a reference acoustic signal, and the evaluation target acoustic signal T ₁ corresponding to the far-end speaker acoustic signal from the data storage unit 180. (See FIG. 3). The PESQ calculation unit 41 obtains and outputs PESQ values for these. The subsequent processing is the same as in the fourth embodiment. Note that the PESQ calculation unit 41 may read the reference sound signal from the data storage unit 180 instead of reading the far end speaker sound signal from the far end speaker sound signal storage unit 102 as the reference sound signal.

[Modification 2 of the fourth embodiment]
Fourth may be in the form obtained by substituting evaluated acoustic signal T ₁ of the first modification of the embodiment in the evaluation target sound signal T _2. That is, the PESQ calculation unit 41 reads the far-end speaker acoustic signal as the reference acoustic signal from the far-end speaker acoustic signal storage unit 102, and the evaluation target acoustic signal corresponding to the far-end speaker acoustic signal from the data storage unit 180. read the T _2. The PESQ calculation unit 41 obtains and outputs PESQ values for these. The subsequent processing is the same as in the fourth embodiment.

[Modification 3 of the fourth embodiment]
As illustrated in FIG. 15, the acoustic quality evaluation apparatus of this modification includes a PESQ calculation unit 41, a linear conversion unit 42, a near-end speaker acoustic signal storage unit 101, a far-end speaker acoustic signal storage unit 102, and a data storage. Unit 180 and signal processing unit 621. The signal processing unit 621 is a processing unit that performs some kind of “signal processing”. An example of “signal processing” is processing including at least one of echo cancellation processing and noise cancellation processing. In addition, the “signal processing” may be processing that does not include echo cancellation processing and noise cancellation processing. The PESQ calculation unit 41 reads the far-end speaker sound signal from the far-end speaker sound signal storage unit 102 as a reference sound signal. The signal processing unit 621 reads the evaluated acoustic signals T ₁ corresponding from the data storage unit 180 to the far-end talker's sound signal, near corresponding to the evaluation target sound signal T ₁ from the near end talker sound signal storage unit 101 An end speaker audio signal is read (see FIG. 3). The signal processing unit 621, these evaluated target sound signal T ₁ to signal processing using, sends a signal obtained thereby to the PESQ calculation unit 41 as the evaluation target signal. The PESQ calculation unit 41 obtains and outputs a PESQ value for the input signal. The subsequent processing is the same as in the fourth embodiment.

[Other variations]
The present invention is not limited to the embodiment described above. For example, the reference signal or the deterioration signal may be obtained based on an acoustic signal (music, background sound, etc.) other than voice. Further, the reference signal and the deteriorated signal may not be a time series signal. In addition, the various processes described above are not only executed in time series according to the description, but may be executed in parallel or individually according to the processing capability of the apparatus that executes the processes or as necessary. Needless to say, other modifications are possible without departing from the spirit of the present invention.

  When the above configuration is realized by a computer, the processing contents of the functions that each device should have are described by a program. By executing this program on a computer, the above processing functions are realized on the computer. The program describing the processing contents can be recorded on a computer-readable recording medium. An example of a computer-readable recording medium is a non-transitory recording medium. Examples of such a recording medium are a magnetic recording device, an optical disk, a magneto-optical recording medium, a semiconductor memory, and the like.

  This program is distributed, for example, by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM in which the program is recorded. Furthermore, the program may be distributed by storing the program in a storage device of the server computer and transferring the program from the server computer to another computer via a network.

  A computer that executes such a program first stores, for example, a program recorded on a portable recording medium or a program transferred from a server computer in its own storage device. When executing the process, this computer reads a program stored in its own recording device and executes a process according to the read program. As another execution form of the program, the computer may read the program directly from the portable recording medium and execute processing according to the program, and each time the program is transferred from the server computer to the computer. The processing according to the received program may be executed sequentially. The above-described processing may be executed by a so-called ASP (Application Service Provider) type service that realizes a processing function only by an execution instruction and result acquisition without transferring a program from the server computer to the computer. Good.

  In the above embodiment, the processing functions of the apparatus are realized by executing a predetermined program on a computer. However, at least a part of these processing functions may be realized by hardware.

1, 2 Data generation devices 3-6 Sound quality evaluation device

Claims (7)

A PESQ calculation unit for obtaining a first PESQ value for a first reference acoustic signal and a first evaluation target acoustic signal based on a signal including the first reference acoustic signal;
A second PESQ value corresponding to a second reference acoustic signal based on a second reference acoustic signal and a second evaluation target acoustic signal based on a signal including the second reference acoustic signal; a reference sound corresponding to the second reference acoustic signal; Based on the linear relationship with the second MOS value based on the five-step evaluation of the difference from the evaluation sound corresponding to the second evaluation target acoustic signal, the first PESQ value is linearly converted to obtain the first A linear converter for obtaining a MOS value;
A sound quality evaluation apparatus. The acoustic quality evaluation apparatus according to claim 1,
The second MOS value is for an evaluation category consisting of a combination of whether or not the difference between the reference sound and the evaluation sound is known and the degree of ease and / or difficulty in hearing the evaluation sound. An acoustic quality evaluation device based on a five-step evaluation. The sound quality evaluation apparatus according to claim 1 or 2,
The second MOS value is a five-level evaluation for an evaluation category consisting of a combination of whether or not the difference between the reference sound and the evaluation sound is known, and a four-level degree of difficulty in hearing the evaluation sound. Sound quality evaluation device based on The sound quality evaluation apparatus according to any one of claims 1 to 3,
The second MOS value is a one-step category indicating that the difference between the reference sound and the evaluation sound is not known, the fact that the difference between the reference sound and the evaluation sound is known, and listening to the evaluation sound. An acoustic quality evaluation apparatus based on a five-stage evaluation for an evaluation category including a four-stage category composed of a combination with a four-stage degree for the stiffness. The sound quality evaluation apparatus according to any one of claims 1 to 4,
The second MOS value is an acoustic quality evaluation device based on a five-step evaluation obtained by instructing an evaluation of the ease of hearing of the evaluation sound. A PESQ calculation step for obtaining a first PESQ value for a first evaluation acoustic signal based on a first reference acoustic signal and a signal including the first reference acoustic signal;
A second PESQ value corresponding to a second reference acoustic signal based on a second reference acoustic signal and a second evaluation target acoustic signal based on a signal including the second reference acoustic signal; a reference sound corresponding to the second reference acoustic signal; Based on the linear relationship with the second MOS value based on the five-step evaluation of the difference from the evaluation sound corresponding to the second evaluation target acoustic signal, the first PESQ value is linearly converted to obtain the first A linear conversion step to obtain a MOS value;
A method for evaluating sound quality.   The program for functioning a computer as an acoustic quality evaluation apparatus in any one of Claim 1 to 5.