JP4973376B2 - Apparatus for detecting basic period of speech and apparatus for converting speech speed using the basic period - Google Patents

Description

  The present invention relates to a technique for detecting a fundamental period of speech and performing speech speed conversion using the fundamental period.

In recent years, in order to expand the user base of mobile phones, mobile phones targeting a user base different from the conventional one have started to be distributed. As an example of this type of mobile phone, there is a mobile phone in which the number of operation keys is reduced to the minimum necessary number for elderly people who are not accustomed to key operations. In this type of mobile phone for elderly people, it is convenient that a speech speed conversion function is provided in order to make it easy to hear the voice of a call. Also, in order to prevent mistakes in listening to important call contents even in a general mobile phone, it is convenient to be able to convert the speech speed such as slowing down the speech speed. As an example of a technique for realizing such speech speed conversion, a technique disclosed in Non-Patent Document 1 can be cited. Non-Patent Document 1 discloses an algorithm called PICOLA that realizes speech speed conversion by inserting or deleting waveforms in units of basic periods. As a method for detecting the basic period of speech, the phase shift processing is performed on the speech data while changing the phase shift amount, and the correlation value between the speech force data before the phase shift and the speech data after the phase shift is calculated. In addition, there is an autocorrelation calculation method in which the time length corresponding to the phase shift amount at which the correlation value reaches a peak is detected as the basic period of the voice represented by the voice data. For example, when the audio data is a sampling data sequence of an audio signal, a process of calculating a correlation value between the sampling data sequence and a sampling data sequence shifted by n (natural number) samples is executed while changing the value of n. The basic period is detected by specifying the value of n at which the correlation value peaks.
Naotaka Morita and Fumitada Itakura, “Expansion and compression of speech over time using pointer movement control (PICOLA) and its evaluation”, Proc. Of the Acoustical Society of Japan, p.149-150, October 1986 Moon

  By the way, in a voice call using a mobile phone, in order to effectively use a limited communication band, it is a common practice to compress and transmit voice data representing a call voice. Specifically, in a mobile phone, audio data to be transmitted to the other party is compressed by the audio codec and transmitted, while the compressed audio data received from the other party is decompressed by the same audio codec to restore the audio data. And the process which reproduces | regenerates an audio | voice according to the audio | voice data is performed. In the audio codec, so-called irreversible compression is performed in order to increase the compression rate. For this reason, the audio data before being compressed by the transmitting-side mobile phone cannot be completely restored by the receiving-side mobile phone. For example, audio data compressed and decompressed using an audio codec tends to have higher harmonic components emphasized compared to the audio data before compression. The enhancement of the higher harmonic components has the advantage that the features of the voice stand out more easily. On the other hand, when the fundamental period described above is detected, the period of the emphasized higher harmonics is erroneously detected as the fundamental period. There is a case.

FIG. 6A is a diagram showing a processing result of a basic period detection process by an autocorrelation calculation method for audio data that has not been compressed and decompressed using an audio codec, and FIG. It is a figure which shows the processing result at the time of performing the basic period detection process by an autocorrelation calculation method, after performing compression and decompression | decompression with an audio codec about audio | voice data. 6A and 6B, the horizontal axis is a coordinate axis representing the above-described phase shift amount in units of samples, and the vertical axis is a coordinate axis representing a correlation value (FIGS. 6A and 6B). In b), the smaller the correlation value, the higher the correlation (the degree of similarity is higher)). As is apparent from a comparison between FIG. 6A and FIG. 6B, the latter has a shorter basic period than the former. This is because the higher harmonic component is emphasized in the latter compared to the former. In this way, for audio data that has been compressed and decompressed by an audio codec, it becomes difficult to accurately detect the fundamental period, and noise is caused by the addition or deletion of a waveform in the fundamental period unit that was erroneously detected. May cause problems in speech speed conversion.
The present invention has been made in view of the above problems, and a first object of the present invention is to accurately detect the fundamental period of audio data that has been compressed and decompressed by an audio codec and emphasized higher harmonic components. It is to provide technology that makes it possible. A second object of the present invention is to provide a technique that makes it possible to avoid the occurrence of noise in speech speed conversion for speech data in which high-order harmonic components are emphasized.

  In order to achieve the first object, the present invention provides a filter that receives audio data, removes high-order harmonic components from the audio data, and outputs the same, and a phase shift process for the output data of the filter while changing the phase shift amount. The correlation value between the output data before the phase shift and the output data after the phase shift is calculated, and the time length corresponding to the phase shift amount at which the correlation value reaches the peak is detected as the basic period of the voice represented by the voice data And a fundamental period detecting unit that provides the fundamental period detecting device. According to such a basic period detection device, since the higher-order harmonic components are removed from the audio data, which is the detection target of the basic period, by the filter, the basic period can be accurately detected.

  In a more preferred aspect, the filter is a low-pass filter or a band-pass filter that blocks passage of band components having a frequency higher than a predetermined cutoff frequency.

  In another preferred aspect, the filter of the basic period detection device is a filter having a variable filter characteristic, and the process of emphasizing the high-order harmonic component included in the audio data at the transmission source of the audio data is performed. It is characterized by comprising specifying means for specifying the applied frequency component and a control unit for setting a filter characteristic for suppressing the frequency component specified by the specifying means in the filter.

  In still another preferred aspect, the filter of the fundamental period detection device is a filter having a variable filter characteristic, and is a frequency characteristic data representing the frequency characteristic of the sound by the signal strength in each of the bands divided in octaves. A plurality of frequency characteristic data representing frequency characteristics of voices of different voice qualities, and the frequency characteristic data of the voice represented by the voice data analyzed by analyzing the voice data applied to the filter. Comparing the frequency characteristic data generated by the analyzing means and each of the plurality of frequency characteristic data stored in the storage means to identify the one having the smallest discrepancy, The filter characteristics of the filter are controlled according to the difference between the measured frequency characteristic data and the identified frequency characteristic data. And having a control unit for.

  In order to achieve the second object, the present invention provides the basic period detection device according to any of the above aspects and the received audio data to the basic period detection device to detect the basic period, and receives the received data. There is provided a speech rate conversion apparatus comprising: a speech rate conversion unit that outputs voice data by inserting or deleting a waveform in units of basic periods detected by the fundamental period detection device. According to such a speech rate conversion apparatus, since the waveform is inserted or deleted in an accurate basic period unit detected from the voice data from which the higher harmonics are removed, the occurrence of the problem such as the noise generation described above is caused. Avoided.

Hereinafter, the best mode for carrying out the present invention will be described with reference to the drawings.
(A: 1st Embodiment)
FIG. 1 is a block diagram showing a configuration example of a speech rate conversion apparatus 1A according to the first embodiment of the present invention. The speech speed conversion device 1A is built in, for example, a mobile phone, performs speech speed conversion on audio data supplied from the audio codec 2 of the mobile phone, and includes a D / A converter and a speaker (both not shown). The voice output system 3 is used to output the voice subjected to the speech speed conversion. As shown in FIG. 1, the speech speed conversion apparatus 1A includes a filter 10A, a basic period detection unit 20A, and a speech speed conversion unit 30A.

  The filter 10A is, for example, a low-pass filter (hereinafter referred to as LPF), and removes band components having a frequency higher than a predetermined cutoff frequency from the audio data supplied from the audio codec 2, and outputs the result. The value of the cut-off frequency of the filter 10A may be determined as appropriate depending on how much higher-order harmonic components that affect the detection of the fundamental period are removed. In the present embodiment, the filter 10A is configured with an LPF, but may be configured with a band-pass filter.

  The basic period detection unit 20A detects the basic period of the voice represented by the voice data from the voice data from which the band component having a frequency higher than the cut-off frequency is removed by the filter 10A. This is an electronic circuit that executes a process of supplying (basic cycle data) to the speech speed conversion unit 30A. An existing algorithm can be used as the fundamental period detection algorithm in the fundamental period detection unit 20A, and in this embodiment, the fundamental period detection algorithm based on the autocorrelation calculation method described above is employed. More specifically, the basic period detection unit 20A performs the above-described phase shift processing in units of the number of samples on the audio data supplied from the filter 10A, and the audio power data before the phase shift and the audio after the phase shift. A correlation value with the data is calculated, and a time length corresponding to the phase shift amount at which the correlation value reaches a peak (minimum in this embodiment) is detected as a basic period of the voice represented by the voice data. Whether the time length corresponding to the phase shift amount that minimizes the correlation value is the basic period or whether the time length corresponding to the phase shift amount that maximizes the correlation value is the basic period is the correlation value (That is, the smaller the value, the higher the correlation, or the higher the value, the higher the correlation).

The speech speed conversion unit 30A performs a time axis companding process according to a predetermined speech speed conversion algorithm (for example, the above-described PICOLA) on the speech data given from the speech codec 2, and the speech data that is the processing result Is an electronic circuit for supplying to the audio output system 3. An operation unit (not shown) of a mobile phone is connected to the speech speed conversion unit 30A. When an instruction to slow down the speech speed and an instruction indicating the degree are given via the operation unit, the speech speed conversion unit 30A responds to the voice data given from the voice codec in accordance with the designated degree. When the instruction to indicate that the speech speed is to be increased and its degree is given via the operation unit, the voice is A process is executed in which the number of waveforms corresponding to the instructed degree is deleted and output in units of the basic period from the audio data given from the codec. Here, speaking speed conversion is not performed on the output data of the filter 10A, but speaking speed conversion is performed on the sound data output from the sound codec 2 (that is, sound data in which higher-order harmonic components are emphasized). The reason is that speech speed conversion is performed while maintaining the advantage obtained by emphasizing higher-order harmonic components (that is, the features of speech are emphasized and easy to hear).
The above is the configuration of the speech speed conversion apparatus 1A.

  With the configuration described above, in the speech rate conversion apparatus 1A according to the present embodiment, detection of the fundamental period is performed on audio data from which band components having a frequency higher than the cut-off frequency have been removed. . For this reason, if the cut-off frequency is appropriately determined, it is possible to remove the high-order harmonic components emphasized by the voice codec of the other party's mobile phone and detect an accurate basic period. 2 and 3 are diagrams illustrating the effects of the present embodiment, and are diagrams illustrating processing results when the autocorrelation calculation processing is performed on the same audio data as in the case of FIG. 6B described above. FIG. 2A is a diagram showing the relationship between the autocorrelation value and the phase shift amount when a first-order LPF with a cutoff frequency of 400 Hz is used as the filter 10A, and FIG. 2B is a cut as the filter 10A. It is a figure showing the relationship between the autocorrelation value at the time of using a 2nd order LPF with an off frequency of 400 Hz, and a phase shift amount. As apparent from a comparison between FIG. 6B and FIG. 2A and FIG. 2B, the calculation results shown in FIG. 2A and FIG. Is relaxed, and the fundamental period can be detected almost accurately. Further, in FIG. 2 (b), the influence of the higher order harmonic component is further alleviated compared to FIG. 2 (a), and the influence of the higher order harmonic component is almost eliminated even compared to FIG. 6 (a). I understand. Therefore, when an LPF is used as the filter 10A, it can be said that it is preferable to use a secondary LPF rather than a primary LPF.

  FIG. 3A is a diagram showing the relationship between the autocorrelation value and the phase shift amount when a second-order LPF having a cutoff frequency of 600 Hz is used as the filter 10A, and FIG. 3B is a cut as the filter 10A. It is a figure showing the relationship between the autocorrelation value at the time of using a 2nd order LPF with an off frequency of 800 Hz, and a phase shift amount. In the calculation results shown in FIGS. 3 (a) and 3 (b), the influence of higher harmonic components is almost eliminated. Therefore, in the aspect using the second-order LPF as the filter 10A, the fundamental period can be obtained almost accurately by using the second-order LPF having a cut-off frequency in the range of at least 400 to 800 Hz. It is possible to avoid the occurrence of noise at the time. In this embodiment, an LPF that inhibits the passage of a band component having a frequency higher than a predetermined cutoff frequency is used as the filter 10A. However, according to an instruction input via the operation unit of the mobile phone. The cut-off frequency may be switched, and whether the filter 10A functions as the primary LPF or the secondary LPF may be switched according to the instruction.

(B: Second embodiment)
FIG. 4 is a block diagram showing a configuration example of a speech speed conversion apparatus 1B according to the second embodiment of the present invention. As apparent from a comparison between FIG. 4 and FIG. 1, the speech speed conversion apparatus 1B has a control unit 40B and a speech speed conversion apparatus in that it has a filter 10B instead of the filter 10A. Different from 1A.

  The filter 10B is a filter with variable filter characteristics. The filter characteristics of the filter 10B are controlled by the control unit 40B. The control unit 40B includes, for example, a CPU (Central Processing Unit), a non-volatile memory such as a FlashROM in which a control program for causing the CPU to execute the control process of the filter characteristics is written, and a control program for executing the control program. It includes a RAM, which is a volatile memory used as a work area (both not shown). The CPU operating in accordance with the control program obtains characteristic data indicating which high-order harmonic components are emphasized in the compressed audio data from the source of the compressed audio data decompressed by the audio codec 2 And the process which sets the filter characteristic of the filter 10B according to the characteristic data is performed. More specifically, the CPU sets a filter characteristic in the filter 10B that suppresses higher-order harmonic components indicated by the characteristic data (for example, gives a shelving-type filter characteristic to the filter 10B and cuts off the filter characteristic). Execute processing to adjust frequency and gain.

  As an example of the characteristic data, there is data indicating input / output characteristics of a voice codec built in a mobile phone that is a transmission source of the compressed voice data. If such input / output characteristics are known, it is possible to specify which higher-order harmonic components are emphasized by the transmission-side audio codec, and suppress higher-order harmonic components that are emphasized by the transmission-side audio codec. This is because such filter characteristics can be set in the filter 10B. For transmission / reception of such characteristic data, for example, a control channel of a mobile telephone network may be used. In addition, if the voice communication is performed between mobile phones capable of performing voice communication and packet communication in parallel, the above-described characteristic data is transmitted by packet communication. Characteristic data may be transmitted and received. Thus, according to the speech rate conversion apparatus 1B, the filter characteristic corresponding to the input / output characteristic of the voice codec mounted on the mobile phone of the other party is set in the filter 10B. In recent years, various types of mobile phones have been distributed in the market, and the types of audio codecs installed in these mobile phones are also diverse. In general, the input / output characteristics of the audio codec differ depending on the type of the audio codec. However, according to the speech speed converting apparatus 1B according to the present embodiment, the input / output of the audio codec possessed by the mobile phone of the other party is provided. Since the filter characteristic corresponding to the characteristic is set in the filter 10B, it becomes possible to accurately detect the basic period of the call voice regardless of what kind of voice codec the mobile phone of the other party has. It is possible to avoid the generation of noise in the speech speed conversion process using the basic period.

  In the present embodiment, the characteristic data indicating the input / output characteristics of the voice codec of the mobile phone of the other party is obtained from the mobile phone that is the other party of the call. Therefore, it is not necessary to acquire the characteristic data from the mobile phone of the other party. For example, if the input / output characteristics of the voice codec implemented in a mobile phone for a carrier are different for each carrier (carrier) that provides mobile phone service, it is stored in the phone book table of the mobile phone. Each telephone number is associated with data representing the filter characteristics corresponding to the input / output characteristics of the voice codec installed in the mobile phone for the carrier specified by the telephone number. What is necessary is just to make control part 40B perform the process which specifies the filter characteristic corresponding to a number from the storage content of the said telephone directory table, and sets to the filter 10B. In the second embodiment, the filter characteristics of the filter 10B are set based on the input / output characteristics of the voice codec mounted on the mobile phone of the other party, but the speech speed conversion device 1B is further mounted. The filter characteristics of the filter 10B may be set in consideration of the input / output characteristics of the voice codec installed in the mobile phone (receiver-side mobile phone).

(C: Third embodiment)
FIG. 5 is a block diagram showing a configuration example of a speech rate conversion apparatus 1C according to the third embodiment of the present invention. As is apparent from a comparison between FIG. 5 and FIG. 4, the speech speed conversion apparatus 1C is based on the point that the control unit 40C is provided instead of the control unit 40B and the point that the storage unit 50C is newly provided. It is different from the apparatus 1B. The storage unit 50C is a non-volatile memory such as FlashROM, for example. The storage unit 50C stores frequency characteristic data represented by signal intensity in each frequency band obtained by dividing the frequency characteristic of audio in octave units. Specifically, the storage unit 50C stores first frequency characteristic data representing frequency characteristics of male standard voice and second frequency characteristic data representing frequency characteristics of female standard voice. Yes. Here, the first frequency characteristic data and the second frequency characteristic data represent the frequency characteristics of the sound by the signal intensity in each frequency band divided in octave units, so that the acoustic characteristics of the sound are expressed in detail. It is not possible, but it can represent the general characteristics of male standard voice and female standard voice.

  The control unit 40C has the same hardware configuration as the control unit 40B, but a control program different from that of the control unit 40B is previously written in the nonvolatile memory of the control unit 40C. This control program causes the CPU of the control unit 40C to execute the following two processes. First, by analyzing the voice data received from the voice codec 2 (that is, voice data given to the filter 10B), the frequency characteristic data (the voice frequency characteristic represented by the voice data) of the voice represented by the voice data, (Data represented by signal intensity in each frequency band divided in units of octaves). Second, the frequency characteristic data generated in the first process is compared with each of the two types of frequency characteristic data stored in the storage unit 50C to identify the one with the smaller divergence, and the generated frequency This is a process for setting the filter characteristic (specifically, the filter characteristic for reducing the difference between the two) in the filter 10B according to the difference between the characteristic data and the specified frequency characteristic data.

  For example, when it is determined by the second process that the deviation from the second frequency characteristic data is smaller than the deviation from the first frequency characteristic data, the second frequency characteristic data and the first frequency characteristic data The control unit 40C executes a process of setting a filter characteristic for reducing the difference from the frequency characteristic data generated by the process in the filter 10B. Specifically, the control unit 40C determines that the signal strength of the frequency characteristic data generated in the first process for a certain frequency band among the plurality of frequency bands divided in octave units is the second frequency characteristic data. When the signal intensity is larger than the signal strength in the same frequency band by a predetermined threshold or more, a process for setting the filter characteristic for attenuating the signal in the frequency band to the filter 10B is executed. As described above, the first frequency characteristic data is data representing the frequency characteristic of male standard voice, and the second frequency characteristic data is data representing the frequency characteristic of female standard voice. By determining which of the two types of frequency characteristic data has a small deviation from the frequency characteristic data generated in the first process, it is specified whether the other party is a male or a female. In the audio data received from the audio codec 2, by setting a filter characteristic in the filter 10B that reduces the difference between the frequency characteristic data thus identified and the frequency characteristic data generated in the first process, The band component emphasized compared to the standard male (or female) voice (that is, the higher-order harmonic component emphasized by the voice codec of the other party) is suppressed by the filter processing by the filter 10B to detect the basic period. It is given to the part 20A. That is, according to the speech speed conversion apparatus 1C according to the present embodiment, the filter characteristic suitable for the voice quality of the other party is set in the filter 10B, and the high-order harmonic components that become an obstacle to the detection of the fundamental period are removed. In the present embodiment, the frequency characteristic data corresponding to two types of voice qualities, male standard voice and female standard voice, are stored in the storage unit 50C, but three or more types of different voice qualities are stored. The frequency characteristic data indicating the frequency characteristic of the voice may be stored in the storage unit 50C.

(D: Other embodiments)
As mentioned above, although each embodiment of this invention was described, you may add the deformation | transformation described below to each said embodiment.
(1) In the above-described embodiment, speech speed conversion using the basic period (that is, waveform insertion in the basic period unit) is performed by accurately detecting the basic period of the audio represented by the audio data decompressed by the audio codec. (Or deletion) was avoided. However, speech processing that can use the fundamental period (or fundamental frequency that is the inverse of the fundamental period) is not limited to speech speed conversion, and in other speech processing that uses the fundamental period or fundamental frequency. However, it goes without saying that it is desirable that they are accurately detected. Therefore, a fundamental period detector (or fundamental frequency detector) is configured by combining the above-described filter 10A and fundamental period detector 20A, and the fundamental period detector (basic frequency detector) executes the other voice processing. It may be incorporated in a signal processing device. Similarly, the fundamental period detector may be configured by combining the filter 10B according to the second embodiment, the fundamental period detector 20A, and the controller 40B. The filter 10B, the fundamental period detector 20A according to the third embodiment, The basic period detection device may be configured by combining the control unit 40C and the storage unit 50C.

(2) In the above-described embodiment, the case where the speech speed conversion device according to the present invention is incorporated into a mobile phone has been described. However, the subject of incorporation of the speech speed conversion device according to the present invention is not limited to a mobile phone. For example, a telephone conference terminal may be used. In short, audio data obtained by transmitting compressed audio data in which high-order harmonic components are emphasized by compression by the audio codec via the communication network and decompressing the compressed audio data received via the communication network by the audio codec If the electronic device reproduces voice according to the above, it is possible to perform speech speed conversion with less noise by incorporating the speech speed conversion device according to the present invention.

It is a block diagram which shows the structural example of 1 A of speech-speed converters which concern on 1st Embodiment of this invention. It is a figure which shows the processing result at the time of using the 1st-order LPF whose cutoff frequency is 400 Hz as the filter 10A of the speech speed converter 1, and the case where the 2nd-order LPF is used. It is a figure which shows the processing result at the time of using the secondary LPF whose 800 Hz is the same as the case where the secondary LPF whose cutoff frequency is 600 Hz is used as the filter 10A of the speech speed converting apparatus 1. It is a block diagram which shows the structural example of the speech-speed converter 1B which concerns on 2nd Embodiment of this invention. It is a block diagram which shows the structural example of 1C of speech-speed converters which concern on 3rd Embodiment of this invention. It is a figure for demonstrating the problem of the conventional speech speed conversion in a mobile telephone.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1A, 1B, 1C ... Speech speed conversion apparatus, 2 ... Voice codec, 3 ... Voice output system, 10A, 10B ... Filter, 20A ... Basic period detection part, 30A ... Speech speed conversion part, 40B, 40C ... Control part, 50C ... memory part.

Claims (2)

  A filter that receives audio data, outputs high-order harmonic components from the audio data, and outputs a filter, the filter having a variable filter characteristic;
  Identifying means for identifying a frequency component subjected to processing to be emphasized at a transmission source of the audio data among high-order harmonic components included in the audio data;
  A control unit for setting a filter characteristic for suppressing the frequency component specified by the specifying means in the filter;
  The output data of the filter is subjected to a phase shift process while changing the phase shift amount, a correlation value between the output data before the phase shift and the output data after the phase shift is calculated, and the correlation value reaches the peak corresponding to the phase shift amount A basic period detection unit that detects a time length as a basic period of the voice represented by the voice data;
  A basic period detection device comprising:   A fundamental period detection device according to claim 1;
  The received voice data is supplied to the fundamental period detector to detect the fundamental period, and the received voice data is output by inserting or deleting a waveform in units of the fundamental period detected by the fundamental period detector. A speech speed conversion unit;
  A speech rate conversion device comprising:

Images