JPH098680A - Voice decoding device - Google Patents

Abstract

PURPOSE: To reduce the abnormal change of the false background noise outputted to a silent section at the time when the level of the background noise as the last frame of a sound section is quickly changed by disturbance or the like with respect to the voice decoding device which receives and decodes the intermittent transmission signal transmitted only in sound sections in the adaptive difference PCM encoding communication. CONSTITUTION: A controller 4 outputs control signals (d) and (e) indicating a sound section or a silent section in accordance with a sound detection flag (b) separated from a reception demodulated signal (a) by a demultiplexer 3. An auxiliary prediction coefficient holder 7 is provided, and a prediction coefficient (i) of the last frame in a prediction coefficient holder 6 which extracts, updates, and stores a prediction coefficient (g) of each frame used for decoding of voice encoded data (c) by a decoder 5 is stored and updated plural times, and the average value of these plural stored values is obtained as a prediction coefficient h=f for reproducing of the false background noise in the silent section.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech decoding device in speech coding communication, and more particularly to an adaptive difference PC for a coding system.
The present invention relates to a speech decoding device to which the M (ADPCM) system is applied.

[0002]

2. Description of the Related Art When voice communication is performed, it is said that the time rate at which either one of the callers speaks is about 35%. In recent years, the range of personal communication, which is a communication mainly based on individuals, has been expanding. Here, voice communication using a portable terminal is mainly performed. As a matter required for such a portable terminal, firstly, there is cordlessness. Secondly, it is required to reduce circuit power consumption because a battery is used for convenience of carrying and it is required to endure long-term use.

As a conventional method of reducing the circuit power consumption, there is a method of paying attention to the utterance time rate of a voice and operating the transmission circuit only when uttering a voice, and putting the circuit in a rest state during the other transmission time. In order to realize such a technique, a voice detection function may be provided on the transmission side to add a discontinuous transmission device that transmits only a sound period. In that case, the problem is on the receiving side. That is, the reproduced sound is intermittent on the receiving side, which makes the sound very unpleasant. The cause is that the background noise is superimposed on the voice when the voice is transmitted, but the background noise is not transmitted when there is no voice. That is, it is known that the background noise is modulated and transmitted only when there is a voice signal. As a method for solving such a problem, a method is known in which a pseudo background noise similar to the background noise on the transmitting side is generated in a silent section in which a voice signal is not transmitted on the receiving side.

The present invention is intended for the decoder on the receiving side, but since the decoder on the receiving side has the same constituent parts as the encoder on the transmitting side, the outline of the ADPCM encoder on the transmitting side will be described. Will be explained. FIG. 5 is a block diagram of the transmission side encoder. In the figure, 51 is a μ-law PC of 64 kb / s
Uniform PC for converting M input signal to linear 13-bit PCM
M converter. Reference numeral 52 is a subtracter 52 that subtracts the prediction signal j that is the output of the adaptive predictor 53 from the output of the uniform PCM converter 51 to obtain the difference signal k. This differential signal k is quantized by the adaptive quantizer 54, and 32 kb / s voice data is sent to the transmission line as the output of the ADPCM voice encoder. On the other hand, the adaptive inverse quantizer 56 has a capacity of 32 kb.
The output voice data of / s is adaptively dequantized and the quantized difference signal m is output. The adder 55 adds the quantized difference signal m and the prediction signal j and outputs the reproduction signal n. The adaptive predictor 53 calculates a prediction coefficient and uses it to generate and output a prediction signal j from the quantized difference signal m and the reproduction signal n.

The prediction coefficient calculated by the adaptive predictor 53 to generate the prediction signal j is a coefficient for predicting a sample value at a certain time point by two adjacent sample values in the past, and the value is a tone color. Represents For example, the prediction coefficient has different occurrence distributions in the case of a voice signal and the case of background noise, and even with the same background noise, white noise having a flat frequency characteristic,
The generation distribution is different even for noise having a slope such as the well-known pink noise. As an example, FIG. 6 is a waveform example diagram of the input signal and the prediction coefficient, and shows a1 and a2 of the prediction coefficients calculated by the adaptive predictor 53 and the temporal change of the voice. It can be seen that the value of the prediction coefficient is different between the sections of the sections A, B, and C where the voice is present and the sections D and E of the background noise only. Here, the background noise superimposed on the voice is
For example, it is noise of an actual air conditioner (air conditioner) having a tilted frequency characteristic. The prior art and the present invention both focus on this prediction coefficient as one means for generating pseudo background noise. An example will be described.

FIG. 4 is a block diagram of a conventional decoding device. In FIG. 4, 1 is an antenna. Reference numeral 2 denotes a reception demodulator that receives and demodulates a modulated wave that is obtained by multiplexing a voice detection flag of the voice detector that determines the presence or absence of voice from the encoder on the transmission side and ADPCM coded data.
3 is the ADPCM encoded data c of the received and demodulated signal a
And a voice detection flag b. Reference numeral 4 denotes a controller that receives the voice detection flag b from the demultiplexer 3 and outputs a control signal to the power holder, the ADPCM decoder 5, and the prediction coefficient holder 6 for performing control described later. 5 receives the control signal d from the controller 4,
The ADPCM decoder generates random ADPCM data for pseudo background noise, transmits and receives the prediction coefficient data to and from the prediction coefficient holder 6, and decodes the ADPCM coded data c from the demultiplexer 3. Reference numeral 6 calculates the average value of the prediction coefficient g of the ADPCM decoder 5 for each frame, updates the storage of the average value in the voiced section and stops the update in the silent section by the control signal e from the controller 4. It is a prediction coefficient holder. Reference numeral 8 is a speaker.

The operation will be described with reference to FIGS. 4 and 7. First, in FIG. 4, a modulated wave in which a voice detection flag and ADPCM coded data are multiplexed and modulated is received by an antenna 1, received and demodulated by a reception demodulator 2, and a demodulation signal a is sent to a demultiplexer 3. The voice detection flag mentioned here indicates the result of detection by the voice detector of a portion with voice (voiced section) and a portion without voice (voiceless section) of the input voice of the encoder on the transmission side. The demultiplexer 3 uses the voice detection flags b and A
Separate into DPCM encoded data c. In this case, as an example, the ADPCM coded data c has 5 msec as one frame. The voice detection flag b is sent to the controller 4 every 5 msec. Upon receiving the voice detection flag b, the controller 4 receives a control signal d, which indicates either a voiced section or a silent section.
Output e. The ADPCM encoder 5 internally generates random data within a range that the ADPCM encoded data c can take in order to generate pseudo background noise, because the modulated signal is cut off in the silent section, and the data is held as a prediction coefficient. Decoding is performed using the prediction coefficient f given from the device 6.

The prediction coefficient holder 6 is used for ADPC when transmission is interrupted.
In order to add the spectrum information of the actual noise to the pseudo background noise generated by the M decoder 5, the ADPCM decoder 5
The prediction coefficient g in is extracted, the average value is calculated for each frame, and the updated value is held. When the control signal e changes from the voiced section to the silent section, the average value of the prediction coefficient of the immediately preceding frame, that is, the last frame of the voiced section is held and the update is stopped. The ADPCM decoder 5 cancels this update, decodes the internally generated random data using the prediction coefficient f held, and outputs pseudo background noise.

It is generally said that the speech detector on the encoder side is a hangover in which when a voiced section is changed to a silent section, it is judged as a voiced section for about several 100 msec in order to prevent the ending of conversation. Is equipped with
The prediction coefficient value of the last frame of the voiced section is the value of background noise. With this effect, even if decoding is performed using random data, noise having a tone color similar to the background noise superimposed on the actual voice on the encoder side is generated.

FIG. 7 is a waveform example diagram for explaining the prediction coefficient for the pseudo background noise, showing the case where there is no disturbance. In the figure, (A) shows the input speech signal (speech + background noise) of the encoder, and (B) shows the temporal change of the prediction coefficient a1 as an example among the prediction coefficients for this. If the speech is ideally detected like the speech detection flag (C) in the speech detection processing on the encoder side, the encoder outputs the speech signal (D).
Only the voice data of is transmitted intermittently. (E) is a prediction coefficient a1 for this audio signal (D). In the prior art,
In order to insert the pseudo background noise into the sections a, b, and c like the voice signal (D), the frame immediately before the transition from the voiced section to the silence section, here, t held by the prediction coefficient holder 6, Random data of sections a, b, and c are decoded using the average value (F) of the prediction coefficients a1 of t-1 and t-2, and pseudo background noise is output. That is, the prediction coefficient used by the decoder 5 for decoding is the prediction coefficient a1 (G). In this state, the tone color of the pseudo background noise in the silent sections a, b, and c is the same as the prediction coefficient value of the actual background noise in the sections a, b, and c, as in the prediction coefficient a1 (F). , It becomes a pseudo-background noise that is continuous without any discomfort.

[0011]

Next, problems of the prior art will be described with reference to FIG. FIG. 8 is a waveform example diagram of the prediction coefficient when there is a disturbance. When a sudden disturbance or the like is added to the last background noise frame (t-1) in the voiced section, for example, when an abnormal noise is input to the microphone on the encoder side or due to an error on the transmission path. Etc., or when the received voice detection flag indicates an incorrect determination result, for example, when the voice detector on the encoder side causes a determination error and determines that there is a silent section in the voiced section or transmission When an error occurs in the voice detection flag on the road, the input voice signal (A), the prediction coefficient (B), the voice detection flag (C)
As shown in, the final frame t-1 in the voiced section has a level different from the background noise in both the input speech (A) and the prediction coefficient a1 (B). In the conventional method, only the average value of the prediction coefficient of the last frame in the voiced section is used to generate the pseudo background noise in the subsequent silence section. That is, since the average value of the Z part of the prediction coefficient a1 (E) is used as the prediction coefficient of the section b, the section a and the section c are considerably different like the difference x and y of the prediction coefficient a1 (F). Since different values are used, the tone color of the pseudo background noise in the section b is different, and in the sections a, b, and c, there is a discontinuity and discomfort as shown in (G).

An object of the present invention is to reduce the adverse effect on the pseudo background noise generation due to intermittent reception for a short time, which is a problem of the prior art, and to realize the ADPCM voice coding system.
(EN) Provided is a voice decoding device which prevents discomfort in reproduced pseudo background noise even when there is a disturbance having a large difference in tone color from the last part of a voiced section or immediately before becoming unvoiced. Especially.

[0013]

A speech decoding apparatus of the present invention comprises an adaptive differential PCM coded signal of a speech signal, a utterance period and a background noise period of a fixed time immediately after the utterance period becomes a silent period. A voice detection flag indicating any one of a voiced section including a voiced section up to the next voiced section is multi-modulated, and an intermittent transmission signal is transmitted in which only the voiced section is transmitted and the transmission of the voiceless section is stopped. A reception demodulator for demodulation, a demultiplexer for separating the output of the reception demodulator into an adaptive differential PCM coded signal and a voice detection flag, and a voiced section and a silence according to the voice detection flag output from the demultiplexer. A controller that outputs a control signal indicating any one of the sections, and a prediction coefficient for the adaptive differential PCM coded signal of the sound section output from the demultiplexer when the control signal from the controller indicates the sound section. Using ADPCM for outputting a reproduced voice and outputting a pseudo background noise as reproduced noise in a silent section by decoding a random code generated internally when the control signal indicates a silent section by an average value of a prediction coefficient given from the outside. When the control signal from the decoder and the controller indicates a voiced section, the prediction coefficient of the ADPCM decoder is extracted, an average value for each frame is obtained, stored and updated, and the control signal from the controller is In a speech decoding apparatus provided with a prediction coefficient holder that stops storing and updating when indicating a silent section and gives the prediction coefficient of the immediately preceding background noise period to the ADPCM decoder, the prediction coefficient holder calculates The average value of the prediction coefficient of the background noise period is extracted and stored, and the prediction coefficient obtained by weighting the past value is given to the prediction coefficient holder to give the ADPCM recovery value. It is characterized in that an auxiliary prediction coefficients cage for the prediction coefficient average value of a silent interval for vessels.

Further, the auxiliary prediction coefficient holder extracts the prediction coefficient of the background noise period from the prediction coefficients for which the prediction coefficient holder has calculated the average value for each frame over a plurality of past times, and sequentially updates and stores it. 3. A prediction coefficient memory, and an average value calculator for calculating an average value of stored values of the prediction coefficient memory for a plurality of times and providing a prediction coefficient to the prediction coefficient holder. According to the auxiliary prediction coefficient holder described above, the prediction coefficient of the background noise period of the frame extracted each time the prediction coefficient holder obtains an average value for each frame is multiplied by a first positive value less than 0.5. A first multiplier that multiplies the input value from the first multiplier, an adder that adds the input value from the first multiplier and the other input value to obtain a prediction coefficient to be given to the prediction coefficient holder, and Update and store output and output And a second coefficient multiplier for multiplying the output value of the prediction coefficient memory by subtracting the first multiplication value from 1 as a second multiplication value and setting the other input value of the adder. It is characterized by having and.

[0015]

FIG. 1 is a block diagram showing an embodiment of the present invention. In FIG. 1, 1 is an antenna, 2 is a voice detection flag and ADP
A reception demodulator for receiving and demodulating a modulated wave in which CM coded data is subjected to multiplex modulation, a demultiplexer 3 for separating a reception demodulated signal a into ADPCM coded data c and a voice detection flag b,
Reference numeral 4 is a controller that receives the voice detection flag b and outputs control signals d and e. Reference numeral 5 is an ADPCM decoder. The above is the same as the conventional circuit. 6 predictive coefficient holder, the predictive coefficient g which is an internal variable of the ADPCM decoder 5 is extracted in the voiced section, the average value is calculated for each frame, the average value is stored and updated, and the update is stopped in the silent section. To do. Reference numeral 7 is an auxiliary prediction coefficient holder added in the present invention, and 8 is a speaker.

[Operation] The operation is the same as the prior art except for the operations of the prediction coefficient holder 6 and the auxiliary prediction coefficient holder 7, and the differences from the prior art will be described in detail. Prediction coefficient holder 6
Is the ADPCM decoder 5 for the voiced section according to the control signal e.
The prediction coefficient g calculated in the above is extracted, the average value is calculated for each frame, and the update is held, and when the sound section changes to the silent section, the update is stopped and the value i of the last frame is held as the auxiliary prediction coefficient. Output to the container 7. The auxiliary prediction coefficient holder 7 is
Every time the value i of the final frame calculated by the prediction coefficient holder 6 is input, the pseudo background noise is generated by using the prediction coefficient value immediately before the change from the past sound section to the silent section. A process for reducing the influence of the problem of the conventional technique on the prediction coefficient value is performed. For example, weighting the data of the last frame in the past to suppress the influence of the current data, or performing the process of weighting the past data by averaging the data of the past several frames to perform the prediction coefficient storage. The processed value h is output to 6. The prediction coefficient holder 6 outputs this value h as f in the silent section, and AD
The PCM decoder 5 uses this value f to generate pseudo background noise.

FIG. 2 is a block diagram showing a first embodiment of the auxiliary prediction coefficient holder 7 which forms the essential part of the present invention. In FIG. 2, 21 is an average value calculator, and 22 is a prediction coefficient memory. The average value calculator 21 calculates the average value of the prediction coefficients of the last frame of the memory 22 a plurality of times, and the prediction coefficient holder 6
, The average value h is output. Prediction coefficient memory 22
Sequentially stores and updates the past prediction coefficient values for a plurality of final frames, for example, 5 frames from t to t-4.

[Operation] When the change from the voiced section to the silent section is detected, the auxiliary prediction coefficient holder 7 extracts the prediction coefficient average value i of the final frame from the prediction coefficient holder 6. The memory 22 updates and stores the value in the memory every time the average value of the prediction coefficients of the final frame is input. Where t
Is the value of i that is currently input, t-1 is the value that was input one time ago, and a total of 5 values up to t-4 are stored. Each time a new value i is input, Each value is sequentially updated in the past memory. Next, the average value calculator 21 calculates the average value of the five values output from the memory 22, and supplies the average value h to the prediction coefficient holder 6. The prediction coefficient holder 6 gives the value h to the decoder 5 as the value f. By this processing, the prediction coefficient of the region b in FIG. 8F becomes as shown by the broken line l, and in the region c, it smoothly approaches the normal value as shown by the broken line k, and is affected by Z. Also, the conventional difference x becomes m and y becomes n without any extreme difference between the regions a, b, and c, and the discomfort is greatly reduced.

Next, FIG. 3 is a block diagram showing a second embodiment of the auxiliary prediction coefficient holder 7 forming the essential part of the present invention.
In FIG. 3, 31 is an adder, 32 and 33 are multipliers, 3
4 is a prediction coefficient memory. The multiplier 32 has 0 <α <
A multiplication value α of 0.5 is set to lighten the weighting of the currently input prediction coefficient average value i of the last frame. A multiplication value β that sets 1−α = β is set in one of the multipliers 33, and the previous prediction coefficient average value h output from the adder 31 is set.
Acts to increase the weighting of. For example, α =
When a multiplication value of 0.05 and β = 1-α = 0.95 is set, the adder 31 adds 5% of the current input value i and 95% of the previous output value h from the adder 31. Will be output. The prediction coefficient memory 34 uses the output value h of the adder 31.
Are sequentially stored and updated, and the stored value p is given to the multiplier 33 and multiplied by the multiplication value β to be one input of the adder 31 for the next time.

When the voiced section is changed to the silent section, the auxiliary prediction coefficient holder 7 extracts the prediction coefficient value i of the final frame from the prediction coefficient holder 6. Adder 31, multiplier 32,
At 33, the following equation (1) is calculated.

## EQU00001 ## h = p.times.0.95 + i.times.0.05 (1) That is, h is the prediction coefficient value that includes the value of the previous i,
It does not greatly depend on the value of i currently input. This means that even if the value of i at a certain point changes abruptly, it will not affect the output value of h so much and will settle to an appropriate value. Prediction coefficient memory 3
4 updates and stores the new h. The initial value is set to 0 here (it settles to an appropriate value in a short time.) By this processing, it is said that the rapid change of the present value is suppressed by the past value as in the first embodiment of FIG. The effect is obtained.

[0021]

As described in detail above, the pseudo background noise is generated by using the prediction coefficient of the last frame in the past, so that even when a signal or the like which changes abruptly in a short time comes in. Since it is possible to generate pseudo background noise having almost the same spectrum information as the background noise superimposed on the voice without being affected by the signal as it is,
The feeling of strangeness given to the listener is reduced.

[Brief description of drawings]

FIG. 1 is a configuration example diagram of a speech decoding apparatus showing an embodiment of the present invention.

FIG. 2 is a configuration diagram showing a first embodiment of the auxiliary prediction coefficient holder shown in FIG. 1 of the present invention.

FIG. 3 is a configuration diagram showing a second embodiment of the auxiliary prediction coefficient holder shown in FIG. 1 of the present invention.

FIG. 4 is a diagram showing a configuration example of a conventional speech decoding device.

FIG. 5 is a block diagram of a transmission side encoder of the ADPCM encoding system.

FIG. 6 is a waveform example diagram of an input signal and a prediction signal.

FIG. 7 is a waveform example diagram illustrating a prediction coefficient for pseudo background noise.

FIG. 8 is a waveform example diagram illustrating a prediction coefficient for pseudo background noise when there is a disturbance.

[Explanation of symbols]

 1 antenna 2 reception demodulator 3 demultiplexer 4 controller 5 ADPCM decoder 6 prediction coefficient holder 7 auxiliary prediction coefficient holder 8 speaker a demodulation signal b voice detection flag c ADPCM coded signal e control signal f pseudo spurious noise Prediction coefficient value given to g g Prediction coefficient 21 Average value calculator 22 Prediction coefficient memory 31 Adder 32 Multiplier 33 Multiplier 34 Prediction coefficient memory 51 Uniform PCM converter 52 Subtractor 53 Adaptive predictor 54 Adaptive quantizer 55 Adder 56 Adaptive inverse quantizer j Prediction signal m Quantization difference signal n Reproduction signal

Claims (3)

[Claims] 1. A voiced section including an adaptive differential PCM coded signal of a voice signal, a voiced period and a background noise period of a fixed time immediately after the voiced period is changed to a voiceless period and a next voiced section. A reception demodulator for receiving and demodulating an intermittent transmission signal in which a voice detection flag indicating any one of the silent sections is subjected to multiple modulation, and only the voiced section is transmitted and transmission in the silent section is stopped; A demultiplexer that separates an output into an adaptive differential PCM coded signal and a voice detection flag, and control that outputs a control signal indicating either a voiced section or a silent section according to the voice detection flag output from the demultiplexer And an adaptive difference PC of a voiced section output from the demultiplexer
When the control signal from the controller indicates a voiced section, the M coded signal is decoded using a prediction coefficient and is output as reproduced voice, and when the control signal indicates a silent section, a random code generated internally is output. An ADPCM decoder which decodes according to an average value of prediction coefficients given from the outside and outputs pseudo background noise as reproduction noise in a silent section; and when the control signal from the controller indicates a sound section, the ADPCM decoder
The prediction coefficient of the PCM decoder is extracted, the average value for each frame is obtained and stored and updated, and when the control signal from the controller indicates a silent section, the storage and update are stopped and the immediately preceding background noise period In a speech decoding device provided with a prediction coefficient holder for giving a prediction coefficient to the ADPCM decoder, the prediction coefficient average value of the background noise period obtained by the prediction coefficient holder is extracted and stored, and a past value is weighted. To the ADP to give the prediction coefficient obtained by
A speech decoding apparatus, characterized in that an auxiliary prediction coefficient holder for setting a prediction coefficient average value in a silent section for a CM decoder is provided. 2. The auxiliary prediction coefficient holder retains the prediction coefficient of the background noise period among the prediction coefficients for which the prediction coefficient holder has calculated the average value for each frame over a plurality of past times, and sequentially updates and stores it. The prediction coefficient memory and an average value calculator that calculates an average value of a plurality of values stored in the prediction coefficient memory and uses the prediction coefficient as a prediction coefficient to be given to the prediction coefficient holder. Voice decoding device. 3. The auxiliary prediction coefficient holder according to claim 1,
A first multiplier that multiplies the prediction coefficient of the background noise period of the frame extracted each time the prediction coefficient holder calculates the average value for each frame, by using a positive number less than 0.5 as a first multiplication value; , An adder for adding the input value from the first multiplier and the other input value to obtain a prediction coefficient given to the prediction coefficient holder, and a prediction for updating and storing the output of the adder A coefficient memory and a second multiplier for multiplying an output value of the prediction coefficient memory by subtracting the first multiplication value from 1 as a second multiplication value and setting the other input value of the adder. The speech decoding apparatus according to claim 1, further comprising: