JPH0683399A - Silence processing method for speech coding - Google Patents

Abstract

(57) [Abstract] [Purpose] Concerning the silence processing method in speech coding such as vector quantization or differential coding, the speech level is continuously attenuated by using the speech level information in the decoder. The purpose of this is to realize a sound-deadening process. [Structure] In speech coding in which a code vector selected from a codebook 101 is input to a speech synthesis filter 103 to synthesize speech, a vector to be input to a speech synthesis filter in a decoder is selected from the codevectors of the codebook as silence processing. It is configured to switch to zero or a vector of smaller magnitude.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a silence processing method in speech coding such as vector quantization or differential coding.

Generally, in voice encoding, when the input voice is silent, the silence is not transmitted, and other data (for example, facsimile data) is transmitted during the silent period to improve the use efficiency of the transmission path. There is. In this case, if the transmission signal transmission is suddenly stopped when the sound changes from voice to silence, click noise occurs in the reproduced sound on the decoder side. Become. In this silence processing, for example, as shown in FIG. 10, the output audio level is not lowered sharply but is smoothly lowered over a certain decay time.

[0003]

2. Description of the Related Art FIG. 9 shows a general structure of a voice encoding system having a silence processing function. In FIG. 9, reference numeral 30 is a voice detector for detecting whether the input voice is voiced or silent, and 10 is a coder for voice-coding the input voice by vector quantization, differential encoding or the like to generate the coded information, 20 Is a decoder for decoding the coded information received via the transmission line, and 40 is an attenuating / silencing circuit for silencing.

The attenuating / silencing circuit 40 operates so as to smoothly reduce the level of the output voice in the manner as shown in FIG. 10 so that noise is not generated at the change point from voiced to silence.

FIG. 11 shows a concrete example of the configuration on the decoder side of the voice encoding system having this silence processing function. This decoder uses LD-CELP as a voice encoding method.
(Low Delay Code Excited Linear Prediction) system is used. As shown in the figure, the decoder 20 includes a codebook 21 that accumulates code vectors, a gain prediction filter 21 that applies a gain to the code vector read from the codebook 21, and a synthesis filter that synthesizes speech based on the gained code vectors. 23 and the like are included. The gain prediction filter 22 and the synthesis filter 23 are adapted to adaptively update their coefficients by backward prediction by an adaptive predictor (not shown).

[0006]

In order to perform the silence processing for continuously attenuating the voice level as shown in FIG. 10, it is necessary to know the voice level at the start of the attenuation, so that the decoder It is necessary to calculate and store the level of the output voice in the latter stage of 20 or the like. This function is generally provided in the attenuation / silencer circuit 40 provided in the subsequent stage of the decoder 20, but as a result, there is a problem that the configuration of the silencer circuit becomes complicated.

The present invention has been made in view of the above problems, and it is an object of the present invention to realize a muting process for continuously attenuating a voice level by utilizing the information of the voice level in a decoder. is there.

[0008]

FIG. 1 is a diagram illustrating the principle of the present invention. According to the silence processing method of the present invention, a code vector selected from the codebook 101 is input to the voice synthesis filter 103 to synthesize a voice. It is configured to switch from the code vector of the code book to zero or a vector of smaller size.

Further, the silence processing method according to the present invention is such that, in speech coding in which a code vector selected from a codebook is input to a speech synthesis filter to synthesize speech, as a silence processing, the coefficient of the speech synthesis filter is used in a decoder. It is configured to be gradually smaller.

Further, the silence processing method according to the present invention is such that, in the speech coding in which the code vector selected from the codebook is inputted to the speech synthesis filter to synthesize the speech, the gain applied to the code vector in the decoder is treated as the silence processing. The configuration is such that a gain adjustment that gradually attenuates is performed.

Further, the silence processing method according to the present invention is such that, in the speech coding in which the code vector selected from the codebook is inputted to the speech synthesis filter to synthesize the speech, the coding information inputted to the decoder as the silence processing. In the decoder so that a code vector having a smaller size is selected from the code book.

Further, the silence processing method according to the present invention is a speech encoding in a form having a filter for accumulating information of past output speech in the decoder, and the output speech accumulated in the decoder is used as the silence processing. The information is gradually attenuated.

[0013]

In the first embodiment of the silence processing method, as the silence processing, the vector input to the speech synthesis filter in the decoder is switched from the code vector of the codebook to zero or a small vector, thereby Make the output voice of the voice synthesis filter attenuate smoothly.

In the second embodiment of the silence processing method, as the silence processing, the coefficient of the voice synthesis filter is gradually reduced in the decoder, whereby the output voice of the voice synthesis filter is smoothly attenuated. To do so.

Further, in the third embodiment of the silence processing method, as the silence processing, the gain adjustment for gradually attenuating the gain applied to the code vector in the decoder is performed, whereby the output voice of the voice synthesis filter is smoothed. To be attenuated.

Further, in the silence processing method of the fourth mode, as the silence processing, the coding information input to the decoder is changed so that a code vector having a smaller size is selected from the codebook in the decoder. Therefore, the output voice of the voice synthesis filter is smoothly attenuated.

Further, in the fifth embodiment of the silence processing method, as the silence processing, the output voice information accumulated in the filter in the decoder is gradually attenuated so that the output voice is smoothly attenuated. To

[0018]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 2 shows a voice encoding system using a silence processing system as an embodiment of the present invention. In this embodiment, the LD-CELP method is used as the audio encoding method.

In FIG. 2, reference numeral 3 denotes a voice detector for detecting whether the input voice is voiced or silent. The voiced / non-voiced determination information detected by the voice detector 3 is transmitted to the decoder side via a transmission line. Be transmitted to.

Reference numeral 1 is an encoder, which includes a codebook 11, a gain prediction filter 12, a synthesis filter 13, and a subtractor 1.
4, a minimization circuit 15 and the like are included. This encoder 1
In the above, the code vector searched from the codebook 11 is multiplied by the gain in the gain prediction filter 12 and input to the synthesizing filter 13.
The speech is synthesized in step S1, the difference between the synthesized speech and the input speech is calculated by the subtractor 14, and the minimization circuit 15 searches for the code vector of the codebook 11 so that the power of the difference is minimized. Then, the codebook index is transmitted to the decoder side as encoded information.

The synthesis filter 13 is composed of an IIR type filter as shown in FIG. 3, for example. That is, the output signal Y _i is sequentially passed through the delay element Z ⁻¹ , each delay output is multiplied by the coefficient a _k , the total thereof is obtained, and the total value Σ
Is added to the input signal X _i to obtain the output signal Y _i .

A decoder 2 includes a codebook 21, a gain prediction filter 22, a synthesis filter 23, and a switch 2.
It is configured to include 4 etc. In this decoder 2, a code vector is selected from the code book 21 based on the code book index received from the encoder side, the code vector is multiplied by the gain in the gain prediction filter 22, and the result is input to the synthesis filter 23. The voice is synthesized by the filter 23 to produce a reproduced sound. The synthesizing filter 23 also has the same configuration as that shown in FIG.

Reference numeral 4 is a silencer circuit. This silencer circuit 4
Does not have the function of smoothly attenuating the voice level as described above, but simply turns on / off the output voice from the decoder 2 based on the voiced / non-voiced determination information.

In the encoder 1 and the decoder 2 described above, the gain prediction filters 12, 22 and the synthesis filter 1 are used.
The adaptive predictors 3 and 23 are not shown.

The operation of this embodiment will be described below. In the decoder 2, when the voiced / non-voiced determination information sent from the encoder side changes from voiced to silence, the switch 24
Is switched from the codebook 21 side so that the zero vector is always selected and input to the synthesis filter 23. At this time, the synthesizing filter 23 stores the voice signal before the point of change from the voice to the silence, that is, the voice level information in each delay element Z ^−1. The audio signals stored in the delay elements of the synthesizing filter 23 are gradually attenuated, so that a signal whose audio level is continuously attenuated from the value at the time of switching can be output.

In this embodiment, the zero vector is input to the synthesis filter 23 as the silence processing,
Of course, a vector having a small size may be input instead of the zero vector.

As another embodiment, instead of inputting a zero vector to the synthesis filter 23 as the silence processing, when silence is determined, the coefficients a ₁ to
and the value of a _n to go to attenuate gradually, may be caused to continuously attenuate the output sound.

FIG. 4 shows the configuration of a decoder as another embodiment of the present invention. The codebook 21, the gain prediction filter 22, the synthesis filter 23, and the like in the figure are the same as those described above, but in this embodiment, the gain of the gain prediction filter 22 is adjusted by operating based on the voiced / non-voiced determination information. The gain adjusting circuit 25 is provided.

The operation of this embodiment will be described. When silence is determined, the gain adjusting circuit 25 continuously attenuates the magnitude of the gain of the gain predicting filter 22.
As a result, the magnitude of the excitation vector input to the synthesis filter 23 is gradually attenuated, and thus the output voice is also continuously and smoothly attenuated from the voice level at the time of switching.

FIG. 5 shows the configuration of a decoder as another embodiment of the present invention. In the figure, the configuration of the decoder 2 is the same as the conventional one. The difference is that an encoding information changing circuit 5 is provided in the preceding stage of the decoder, and the encoding information and the voice / silence determination information are input to the encoding information changing circuit 5, and when the voice is present, the encoding is performed. The information is passed as it is, but when it is determined that there is no sound, the coded information is changed so that the code vector having the smallest magnitude is extracted from the codebook. The encoding information changing circuit 5 may be provided on the encoder side or the decoder side.

If the encoding information changing circuit 5 is provided, the code vector having the smallest size is always selected from the code book 21 on the decoder side and input to the synthesis filter 23 when there is no sound. So the synthesis filter 2
The output voice No. 3 is continuously and smoothly attenuated from the voice level at the time of switching.

In each of the above-mentioned embodiments, the adaptive update of the synthesis filter of the encoder and the decoder and the gain prediction filter is stopped while the output voice level is attenuated by the silence processing. Of course, by doing so, it becomes possible to allocate the processing capacity for performing the adaptive update processing to other processing.

FIG. 6 shows another embodiment of the present invention. In each of the above-mentioned embodiments, when the input voice changes from the voice to the silence, the silence processing is performed to continuously attenuate the output voice on the decoder side, but the application of the present invention is not limited to this. However, since noise is generated on the decoder side even when the encoder and the decoder are out of synchronization, it is effective to apply the silence processing of the present invention in such a case.
This embodiment relates to the silencing process when the synchronization is lost.

The basic structure of the encoder and decoder in this embodiment is the same as that of the embodiment shown in FIG. In FIG. 6, the adaptive predictor 1 not shown in the above-described embodiments is shown.
6, 17, 26, 27 are shown. Reference numeral 6 denotes an out-of-sync check circuit provided on the decoder side, which detects that the encoder and the decoder are out of synchronization. This out-of-sync detection is
This can be performed by using the synchronization bit included in the transmission information or the clock extracted from the transmission information.

When the out-of-sync check circuit 6 detects out-of-sync between the encoder and the decoder, the switch 24 is switched to input the zero vector to the synthesis filter 23. As a result, the output voice of the synthesis filter 23 is continuously attenuated, and it is possible to suppress the generation of noise due to the loss of synchronization.

Various modifications can be made in implementing the present invention. For example, although the embodiments described above apply the present invention to speech encoding of the LD-CELP method, the present invention is not limited to this, and speech encoding methods by other vector quantization, for example, normal CELP method. Or VSELP (Vector Sum Excited Linear Prediction
Not only can it be applied to a speech coding method such as a vector addition excitation linear prediction) method, but also to a differential coding method as shown in FIG.

As is well known, the differential encoding system subtracts the output signal (predicted value) X _n-1 'one sample before created by the predictor D from the input signal X _n to obtain the difference e _n , The quantized value e _n 'of the difference e _n is transmitted to the decoder. In the decoder, the output signal (predicted value) X one sample before generated by the predictor D is added to the difference e _n 'transmitted from the encoder.
_n-1 'is added to obtain the output signal X _n '.

FIG. 8 shows the configuration of a decoder as another embodiment of the present invention which uses this differential encoding method.
In the figure, 51 is an adder that adds the received difference e _n 'and the predicted value X _n-1 ', 52 is a predictor that generates the predicted value X _n-1 'one sample before from the output signal X _n ', 53 is the output signal X
A multiplier for multiplying _n ′ by a coefficient α, a switch 54 for inputting zero instead of the difference e _n ′ to the adder 51 at the time of silence determination, 55 an output signal of the adder 51 at the time of silence determination for the multiplier 53 It is a switch that switches to input to.

According to such a configuration, when the input voice is determined to be silent, the signal input to the adder 51 is changed to the difference e.
_{From n} ′ to zero, the output signal X _n ′ is simultaneously multiplied by the coefficient α (where | α | <1). As a result, the subsequent output signal is attenuated from X _n 'by a power of α.

[0040]

As described above, according to the present invention, the audio level is continuously stored by utilizing the audio level information in the decoder without providing a circuit for newly storing the audio level. It is possible to realize the silencing process that attenuates the sound.

[Brief description of drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is a diagram showing a configuration of a voice encoding system by a silence processing as an embodiment of the present invention.

FIG. 3 is a diagram showing a configuration example of a synthesis filter in the embodiment system.

FIG. 4 is a diagram showing the configuration of a decoder that performs silence processing as another embodiment of the present invention.

FIG. 5 is a diagram showing a configuration of a decoder by a silence processing as another embodiment of the present invention.

[Fig. 6] Fig. 6 is a diagram showing the structure of a voice encoding system by a silence processing as another embodiment of the present invention.

[Fig. 7] Fig. 7 is a diagram illustrating a configuration example of a differential encoding method.

[Fig. 8] Fig. 8 is a diagram showing the structure of a decoder that performs silence processing as another embodiment of the present invention.

FIG. 9 is a diagram showing a configuration example of a conventional speech encoding system.

FIG. 10 is a diagram illustrating a sound level attenuation mode by the silence processing.

FIG. 11 is a diagram showing a configuration example of a conventional decoder.

[Explanation of symbols]

 1, 10 Encoder 2, 20 Decoder 3, 30 Speech detector 4 Silencer circuit 5 Encoding information change circuit 6 Out of sync check circuit 11, 21 Codebook 12, 22 Gain prediction filter 13, 23 Synthesis filter 14 Subtractor 15 minimization circuit 16, 17, 26, 27 adaptive predictor 24 switch 25 gain adjustment circuit 51 adder 52 predictor 53 multiplier 54, 55 switch

Front page continued (72) Inventor Shoji Fujino 1015 Kamiodanaka, Nakahara-ku, Kawasaki-shi, Kanagawa, Fujitsu Limited (72) Inventor Mitsuru Tsuboi 1015, Kamedota, Nakahara-ku, Kawasaki, Kanagawa (72) Invented Toshiaki Nobumoto 3-22-8 Hakata Ekimae, Hakata-ku, Fukuoka City, Fukuoka Prefecture, Fujitsu Kyushu Digital Technology Co., Ltd.

Claims (5)

[Claims] 1. In speech coding for inputting a code vector selected from a codebook (101) to a speech synthesis filter (103) to synthesize speech, a vector to be inputted to the speech synthesis filter in a decoder is used as silence processing. A silence processing method in which the code vector of the codebook is switched to zero or a vector of small size. 2. In a voice encoding in which a code vector selected from a codebook is input to a voice synthesis filter to synthesize a voice, as a silence process, a decoder is configured to gradually reduce the coefficient of the voice synthesis filter. Processing method. 3. In speech coding in which a code vector selected from a codebook is input to a speech synthesis filter to synthesize speech, a gain adjustment for gradually attenuating a gain applied to the code vector in a decoder is performed as a silence processing. Silence processing method. 4. In speech encoding for synthesizing speech by inputting a code vector selected from a codebook to a speech synthesis filter, coding information inputted to a decoder as silence processing is
A silence processing method in which a code vector having a smaller size is selected in a decoder so as to be selected from a code book. 5. In a voice encoding in a form including a filter for accumulating information of past output speech in a decoder, as output of silence processing, output speech information accumulated in the decoder is gradually attenuated. Silence processing method.