JPH08137495A - Voice encoding device - Google Patents

Abstract

PURPOSE: To enable voice encoding by changing a code book used for encoding in accordance with quantity of errors in code book search, in a voice encoding device using a CELP system. CONSTITUTION: Making a code vector in an adaptive code book 3 as a calender signal, adaptive code book search is performed by each synthesized voice processed by a linear prediction filter 5 using a coefficient of a linear prediction coefficient extractor 1 and an input voice, and it is decided using the minimum error whether the adaptive code book is used or not in an error discriminator 16. Encoding is performed classifying it into encoding by a probability code book 4 + the adaptive code book 3 or encoding by only the adaptive code book 3, by a changeover switch 15.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice coding device used for digital mobile communication and the like.

[0002]

2. Description of the Related Art FIG. 4 shows a CELP (Code Excited Linear Pre) which divides speech into a certain frame length and encodes a speech excitation signal with a code vector for each frame.
FIG. 3 is a block diagram of a speech coding apparatus according to the dictive Coding method.

In FIG. 4, this speech encoding apparatus is a linear prediction coefficient extractor 1 for obtaining linear prediction coefficients from input speech.
, A linear predictive coefficient quantizer 2 for quantizing the obtained linear predictive coefficient, an adaptive codebook 3 consisting of several adaptive code vectors of frame length, and a probability consisting of some random code vectors of frame length Codebook 4
, Linear prediction filters 5 and 6 based on the linear prediction coefficient extracted by the linear prediction coefficient extractor 1, amplifiers 7 and 8, and gain quantizers 9 and 10 for quantizing the gains of the amplifiers 7 and 8.
And an adder 11 for adding the outputs of the amplifiers 7 and 8, an error evaluator 12 for finding a combination that minimizes the error between the added signals and the input voice, and a coded excitation signal. Excitation signal decoding device 13 and delay device 14
And have.

Next, the operation of the above conventional example will be described. The linear prediction coefficient extractor 1 obtains a linear prediction coefficient from the input speech corresponding to the frame to be encoded, the linear prediction coefficient quantizer 2 quantizes it, and transmits it to the excitation signal decoding device 13. In addition, the code vector in the adaptive codebook 3 is used as an excitation signal, and the synthesis amplified by the amplifier 7 by the amplification factor adjusted so that the error between each synthesized speech processed by the linear prediction filter 5 and the input speech is minimized. The error between the speech and the input speech is examined by the error evaluator 12, and the number of the adaptive code vector having the smallest error in the adaptive codebook 3 is determined (hereinafter referred to as adaptive codebook search).

Next, the code vector in the stochastic codebook 4 is used as an excitation signal, and the amplifier 8 is controlled by the amplification factor adjusted so that the error between each synthesized speech processed by the linear prediction filter 6 and the input speech is minimized. The error between the synthesized speech and the input speech obtained by adding the amplified synthesized speech and the synthesized speech based on the adaptive code vector obtained by the adaptive codebook search (hereinafter referred to as the stochastic codebook search) is examined. The code vector number that minimizes the error in the code, the gain of the amplifier 8 quantized by the gain quantizer 10 and the gain of the amplifier 7 quantized by the gain quantizer 9 corresponding to this code vector are encoded. Transmit as a result. Next, the excitation signal decoding device 13 decodes the encoded excitation signal using the above encoding result, and this decoded excitation signal passes through the delay device 14 and becomes an adaptive codebook in the next frame.

[0006]

However, in the conventional speech coding apparatus described above, sufficient speech quality can be obtained by coding only the adaptive codebook 3 in the steady state of speech, and coding by the stochastic codebook 4 is performed. Probability codebook 4 even if voice quality is hardly improved even if added
However, there is a problem in that useless information is transmitted by performing encoding by.

Further, when the adaptive codebook 3 and the excitation signal of the voice are considerably different from each other at the rising portion of the voice, effective coding using the adaptive codebook 3 cannot be performed and the voice quality is deteriorated. Had a problem.

The present invention solves such a conventional problem, and provides a speech coding apparatus capable of realizing high-quality speech coding by eliminating unnecessary transmission of information in speech coding. With the goal.

[0009]

In order to achieve the above object, the present invention examines an error between a reproduced voice and a voice signal only by an adaptive codebook, and when the error is small, a code by only the adaptive codebook is used. Further, when the error is large, an error determining means is provided for changing the encoding to only the probability codebook.

[0010]

According to the present invention, when the error between the reproduced voice and the voice signal due to only the adaptive codebook is small, the present invention provides
By encoding with only the adaptive codebook, the amount of information required for encoding can be reduced in some frames. In addition, when the error between the reproduced voice and the voice signal due to the adaptive codebook only is small, the coding information used for the probabilistic codebook encoding is assigned to the adaptive codebook or gain quantization to obtain a high-quality voice. Encoding can be realized.

Further, when the error between the reproduced voice and the voice signal due to only the adaptive codebook is large, the coding information used for the adaptive codebook coding is allocated to the stochastic codebook or the gain quantization to improve the coding. Quality voice coding can be achieved.

[0012]

1 shows the structure of a first embodiment of the present invention, in which elements similar to those of the conventional example shown in FIG. In FIG. 1, this speech coding apparatus includes a linear prediction coefficient extractor 1 that obtains a linear prediction coefficient from input speech, a linear prediction coefficient quantizer 2 that quantizes the obtained linear prediction coefficient, and the number of frame lengths. Adaptive codebook 3 consisting of these adaptive code vectors, probability codebook 4 consisting of several probability code vectors of frame lengths, and linear prediction filters 5 and 6 based on the linear prediction coefficients extracted by the linear prediction coefficient extractor 1. And amplifiers 7 and 8
, Gain quantizers 9 and 10 for quantizing the gains of the amplifiers 7 and 8, an adder 11 for selectively adding the outputs of the amplifiers 7 and 8 by a changeover switch 15, and inputs to the added signals. An error evaluator 12 that finds a combination that minimizes the error with speech, an excitation signal decoding device 13 that decodes an encoded excitation signal, a delay device 14, and an amplifier 8
, And the output of the gain quantizer 10, and the changeover switch 15 that determines the use of the probability codebook 4 from the results of the error evaluator 12 using only the adaptive codebook 3 And an error determiner 16 for controlling

Next, the operation of this embodiment will be described. The linear prediction coefficient extractor 1 obtains a linear prediction coefficient from the input speech corresponding to the frame to be encoded, the linear prediction coefficient quantizer 2 quantizes it, and transmits it to the excitation signal decoding device 13. Also, an adaptive codebook search is performed by using the code vector in the adaptive codebook 3 as an excitation signal and each synthetic speech processed by the linear prediction filter 5 and the input speech. Next, the error determiner 16 normalizes the minimum error value obtained by the adaptive codebook search with the input voice power, and if this value is less than a certain threshold value, the changeover switch 15 is turned off to change the adaptive codebook. Only the adaptive code vector number obtained by the search and the gain of the amplifier 7 quantized by the gain quantizer 9 are transmitted as the coding result. If the normalized error is greater than or equal to the threshold value, the changeover switch 15 is turned on and the probability code book 4
The code vector in the probability codebook 4 is used as an excitation signal to perform a probability codebook search using each synthesized speech processed by the linear prediction filter 6 and the adaptive codebook search result. And the gain of the amplifier 8 quantized by the gain quantizer 10 corresponding to this code vector and the gain of the amplifier 7 quantized by the gain quantizer 9 are transmitted as the coding result. Next, the excitation signal decoding device 13 decodes the encoded excitation signal using the above encoding result, and this decoded excitation signal passes through the delay device 14 and becomes an adaptive codebook in the next frame.

In the present embodiment, when the error due to the adaptive codebook search result becomes less than a certain threshold value, a means for stopping the coding by the stochastic codebook is provided so that the error due to the adaptive codebook is small. In addition, it is possible to code the voice only by the adaptive codebook, and it is possible to reduce the coding information necessary for the steady state of the voice (when the error of the adaptive codebook is small).

(Second Embodiment) FIG. 2 shows the structure of a second embodiment of the present invention. The same elements as those in the first embodiment are designated by the same reference numerals. This embodiment is different from the first embodiment in that an adaptive codebook 3 is connected to an interpolating device 17, another adaptive codebook 18 is connected to the interpolating device 17, and the adaptive codebook 18 is switched by a changeover switch 19. By connecting to the linear prediction filter 5 via the amplifier 7, and further connecting the gain quantizer 20 to the amplifier 7 so that the outputs of the two gain quantizers 9 and 20 are selected by the changeover switch 21. is there. Changeover switch 1
9 and 21 are linked with the changeover switch 15 and the error determiner 16
Controlled by.

Next, the operation of this embodiment will be described. The linear prediction coefficient extractor 1 obtains a linear prediction coefficient from the input speech corresponding to the frame to be encoded, the linear prediction coefficient quantizer 2 quantizes it, and transmits it to the excitation signal decoding device 13. Also, an adaptive codebook search is performed by using the code vector in the adaptive codebook 3 as an excitation signal and each synthetic speech processed by the linear prediction filter 5 and the input speech. Next, the error determiner 16 normalizes the minimum error value obtained by the adaptive codebook search with the input voice power. If this value is less than a certain threshold value, the changeover switches 15 and 19 are turned off, and the changeover switch is changed. 21 is switched to the a side, and a code vector in the adaptive codebook 15 created by interpolation from the adaptive codebook 3 is used as an excitation signal, and an adaptive codebook search is performed by the synthetic speech processed by the linear prediction filter 5 and the input speech. And the gain of the amplifier 7 quantized by the gain quantizer 20 having a larger quantization number than the gain quantizer 9 is coded. It is transmitted as the conversion result. If the normalized error is greater than or equal to the threshold value, the changeover switches 15 and 19 are turned on, the changeover switch 21 is changed over to the b side, and the code vector in the stochastic codebook 4 is used as the excitation signal for linear prediction. A stochastic codebook search is performed using each synthesized speech processed by the filter 6 and the adaptive codebook search result, and the code vector number that minimizes the error in the stochastic codebook 4 and the gain quantization corresponding to this code vector. The gain of the amplifier 8 quantized by the device 10 and the gain of the amplifier 7 quantized by the gain quantizer 9 are transmitted as the encoding result. Next, the excitation signal decoding device 13 decodes the excitation signal encoded using the above encoding result, and this decoded excitation signal passes through the delay device 14,
It becomes the adaptive codebook in the next frame.

In the present embodiment, when the error due to the adaptive codebook search result becomes less than a certain threshold value, the coding by the stochastic codebook is stopped and the number of quantizations is made by using a plurality of adaptive codebooks. It is possible to realize more optimal adaptive codebook encoding by quantizing with a large number of gain quantizers, and improve the voice quality in the steady state of the voice (when the error of the adaptive codebook is small). .

(Third Embodiment) FIG. 3 shows the structure of a third embodiment of the present invention. The same elements as those in the first embodiment are designated by the same reference numerals. The present embodiment is different from the first embodiment in that the changeover switch 15 'is an amplifier 7 and a gain quantizer 9.
Is provided on the output side of the linear prediction filter 6 via the changeover switch 23.
That is, another gain quantizer 24 is selectively provided by the gain quantizer 10 and the changeover switch 25. Changeover switches 15 ', 23, 2
5 is controlled by the error determiner 16.

Next, the operation of this embodiment will be described. The linear prediction coefficient extractor 1 obtains a linear prediction coefficient from the input speech corresponding to the frame to be encoded, the linear prediction coefficient quantizer 2 quantizes it, and transmits it to the excitation signal decoding device 13. Also, an adaptive codebook search is performed by using the code vector in the adaptive codebook 3 as an excitation signal and each synthetic speech processed by the linear prediction filter 5 and the input speech. Next, the error determiner 16 normalizes the minimum error value obtained by the adaptive codebook search with the input voice power, and if this value is above a certain threshold value, the changeover switch 15 'is turned off, and the changeover switch 23 is turned on. Is turned on, and the changeover switch 25 is changed to the b side, and the probability code book 2
A codebook search is performed using 2 and the probability codebook 4, and the number of the probability code vector that minimizes the error and the gain quantizer 24 having a larger number of quantizations than the gain quantizer 10
The gain of the amplifier 8 quantized in (1) is transmitted as a coding result. If the normalized error is less than the threshold value, the changeover switch 15 ′ is turned on, the changeover switch 23 is turned off, and the changeover switch 25 is changed over to the a side, so that the code vector in the probability codebook 4 is set. As an excitation signal, a probabilistic codebook search is performed using each synthesized speech processed by the linear prediction filter 6 and the adaptive codebook search result, and the code vector number that minimizes the error in the probabilistic codebook 4 and this code vector The gain of the amplifier 8 quantized by the gain quantizer 10 and the gain of the amplifier 7 quantized by the gain quantizer 9 are transmitted as the coding result. Next, the excitation signal decoding device 13 decodes the excitation signal encoded using the above encoding result, and this decoded excitation signal passes through the delay device 14,
It becomes the adaptive codebook in the next frame.

In this embodiment, when the error due to the adaptive codebook search result exceeds a certain threshold value, the coding by the adaptive codebook is stopped and the number of quantizations is made by using a plurality of adaptive codebooks. It is possible to prevent unnecessary coding by the adaptive codebook by quantizing with a large number of gain quantizers, and select a more effective probability codebook at the start of speech (when the error of the adaptive codebook is large). can do.

[0021]

As is apparent from the above embodiment, the present invention is provided with means for stopping the coding by the stochastic codebook when the error due to the adaptive codebook search result becomes less than a certain threshold value. As a result, when the error due to the adaptive codebook is small, the speech can be coded only by the adaptive codebook, and the coding information required for the steady state of the speech (when the error of the adaptive codebook is small) is reduced. be able to.

Further, according to the present invention, when the error due to the adaptive codebook search result becomes less than a certain threshold value, the coding by the stochastic codebook is stopped and the number of quantizations is made by using a plurality of adaptive codebooks. It is possible to realize more optimal adaptive codebook encoding by quantizing with a large number of gain quantizers, and improve the voice quality in the steady state of the voice (when the error of the adaptive codebook is small). be able to.

Further, according to the present invention, when the error due to the adaptive codebook search result exceeds a certain threshold, the coding by the adaptive codebook is stopped and the quantization is performed by using a plurality of adaptive codebooks. By quantizing with a large number of gain quantizers, unnecessary coding by the adaptive codebook can be prevented, and a more effective stochastic codebook can be used at the start of speech (when the error in the adaptive codebook is large). Can be selected.

[Brief description of drawings]

FIG. 1 is a block diagram of a CELP audio encoding apparatus according to a first embodiment of the present invention.

FIG. 2 is a block diagram of a CELP audio encoding apparatus according to a second embodiment of the present invention.

FIG. 3 is a block diagram of a CELP audio encoding apparatus according to a third embodiment of the present invention.

FIG. 4 is a block diagram of a CELP-based speech encoding apparatus in a conventional example.

[Explanation of symbols]

1 Linear Prediction Coefficient Extractor 2 Linear Prediction Coefficient Quantizer 3, 18 Adaptive Codebook 4, 22 Stochastic Codebook 5, 6 Linear Prediction Filter 7, 8 Amplifier 9, 10, 20, 24 Gain Quantizer 11 Adder 12 Error evaluator 13 Excitation signal decoding device 14 Delay device 15, 15 ', 19, 21, 23, 25 Change switch 16 Error determiner 17 Interpolation device

Claims (3)

[Claims] 1. A means for extracting a linear prediction coefficient for a speech signal of each frame of input speech cut out at a fixed frame interval, an adaptive codebook for storing an adaptive code vector, and a stochastic codebook for storing a probability code vector. A linear prediction filter for generating a synthetic speech signal from each of the code vectors for the input speech cut out at frame intervals, and an amplifier for adjusting a gain so that an error between the synthetic speech signal and the speech signal is minimized. A gain quantizer that quantizes a gain that minimizes the error; an error evaluator that evaluates an error between the synthesized speech signal and the speech signal; a synthesized speech signal based on the adaptive code vector; and the speech signal. And a means for stopping the encoding process by the stochastic codebook when the error between . 2. A means for extracting a linear prediction coefficient for a speech signal of each frame of input speech cut out at a fixed frame interval, a plurality of adaptive codebooks for storing adaptive code vectors, and a probability of storing a probability code vector. A codebook, a linear prediction filter that generates a synthesized speech signal from each of the code vectors with respect to the input speech cut out at frame intervals, and a gain is adjusted so that an error between the synthesized speech signal and the speech signal is minimized. Amplifier, a plurality of gain quantizers having different quantization numbers corresponding to the adaptive codebook that quantizes the gain that minimizes the error, and the probability codebook that quantizes the gain that minimizes the error. A gain quantizer corresponding to, an error evaluator for evaluating an error between the synthesized speech signal and the speech signal, When the error between the synthesized speech signal by the code vector and the speech signal is small, the coding process by the stochastic codebook is stopped, and the gain quantizer with a large number of quantization is performed by using the plurality of adaptive codebooks. A speech coding apparatus provided with a switching unit for quantizing. 3. A means for extracting a linear prediction coefficient for a speech signal of each frame of input speech cut out at fixed frame intervals, an adaptive codebook for storing an adaptive code vector, and a plurality of probabilities for storing a probability code vector. A codebook, a linear prediction filter that generates a synthesized speech signal from each of the code vectors with respect to the input speech cut out at frame intervals, and a gain is adjusted so that an error between the synthesized speech signal and the speech signal is minimized. An amplifier, a gain quantizer corresponding to the adaptive codebook that quantizes the gain that minimizes the error, and a quantization number corresponding to the probability codebook that quantizes the gain that minimizes the error. A plurality of different gain quantizers, an error evaluator that evaluates an error between the synthesized speech signal and the speech signal, and the adaptive coprocessor In a gain quantizer with a large number of quantizations using the plurality of stochastic codebooks, the coding process by the stochastic codebook is stopped when the error between the synthesized speech signal by the code vector and the speech signal is large. A speech coding apparatus provided with a switching unit for quantizing.