JPH09185396A - Speech encoding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To excellently encode a speech signal even in noisy environment and to obtain a speech signal of high sound quality by providing a gain control means which increases and decreases the gain of both or one of a probability code book and an adaptive code book according to the decision result of a speech decision means and the analytic result of a pitch periodicity analyzing means. SOLUTION: On the basis of spectrum parameters obtained by analyzing an input signal by an LPC analyzer 16, a composing filter 8 puts together signals outputted from the probability code book 4 and the adaptive code book 1 and an error evaluation unit 14 outputs codes corresponding to the delay of the adaptive code book 1 minimizing the distortion of the composite signal, the indexes of the probability code book 4, and respective gains. According to the decision result of a speech decision unit 11 and the analytic result of the pitch periodicity analyzer 10, the gains of the probability code book 4 and adaptive code book 1 are increased and decreased by a gain controller 15.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coder.

[0002]

2. Description of the Related Art As a widely used means for efficiently compressing a speech signal, the speech signal is coded using a linear prediction parameter representing a spectrum envelope and a sound source parameter corresponding to the linear prediction residual signal. There is a method to make it. A speech coding method using such a linear prediction method can obtain a relatively high quality synthesized speech with a small transmission capacity. Therefore, in combination with recent advances in hardware technology, various application methods have been widely used. Researched and developed. Among them, CELP (Code Excited Linear Predictive Coding), which uses an adaptive codebook obtained by repeating past sound source signals, is used as a method for obtaining good sound quality.
The method is well known. For the CELP method, for example, “Improved speech quality and
efficientvector quantization in SELP ”(ICASP '88
s4.4, pp. 155-158, 1988).

FIG. 6 is a block diagram of a conventional code-driven linear prediction type speech encoding apparatus having an adaptive codebook. In the figure, the adaptive codebook 51 is connected to a first input terminal of an adder 53 via a multiplier 52. The probability code book 54 includes a multiplier 55 and a switch 56.
Is connected to the second input terminal of the adder 53 via.

The output terminal of the adder 53 is connected to the adaptive codebook 51 via the delay circuit 57, and is also connected to the first input terminal of the synthesis filter 58.

Further, the buffer memory 59 connected to the input terminal 66 to which the digital audio signal is input is the LPC.
It is connected to the second input terminal of the synthesis filter 58 via the analyzer 60, and is also connected to the first input terminal of the subtracter 62 via the subframe divider 61. The second input terminal of the subtractor 62 is connected to the output terminal of the synthesis filter 58, and the output terminal is connected to the error evaluator 64 via the perceptual weighting filter 63. The error evaluator 64 is connected to the adaptive codebook 51, the probability codebook 54, and the multipliers 52 and 55.

Further, the LPC analyzer 60 and the error evaluator 64 are connected to a multiplexer 65.

In the above structure, the original audio signal sampled at 8 kHz, for example, is input from the input terminal 66, and a predetermined frame interval (for example, 20 m) is input.
The audio signal of s, that is, 160 samples) is stored in the buffer memory 59. The buffer memory 59 sends the original audio signal to the LPC analyzer 60 in frame units. LP
The C analyzer 60 uses a linear prediction (LP
C) Analysis is performed to extract the linear prediction parameter α representing the spectral characteristic, and the linear prediction parameter α is sent to the synthesis filter 58 and the multiplexer 65. The subframe divider 61 divides the original audio signal of a frame into predetermined subframe intervals (for example, 5 ms, that is, 40 samples). That is, subframe signals from the first subframe to the fourth subframe are created from the original audio signal of the frame.

The delay L and the gain β of the adaptive codebook 51 are determined by the following processing.

First, the delay circuit 57 gives a delay corresponding to the pitch period to the input signal of the synthesizing filter 58 in the preceding subframe, that is, the driving sound source signal, and creates it as an adaptive code vector. For example, if the assumed pitch period is 40 to 167 samples, 128 types of signals with a delay of 40 to 167 samples are created as adaptive code vectors and stored in the adaptive codebook 51. At this time, the switch 56 is in an open state. Therefore, each adaptive code vector from the adaptive codebook 51 is multiplied by a variable gain value in the multiplier 52, then passes through the adder 53 and is input to the synthesis filter 58 as it is. The synthesizing filter 58 performs the synthesizing process using the linear prediction parameter α, and sends the synthesized vector to the subtractor 62. The subtractor 62 subtracts the original speech vector and the synthesized vector and sends the obtained error vector to the perceptual weighting filter 63. The perceptual weighting filter 63 performs a weighting process on the error vector in consideration of the perceptual characteristic, and sends it to the error evaluator 64. The error evaluator 64 calculates the mean square of the error vector, searches for the optimum adaptive code vector having the smallest mean square value, and determines the delay L thereof.
And the gain β are sent to the multiplexer 65. In this way, the delay L and the gain β of the adaptive codebook 51 are determined.

Next, the index i and the gain γ of the probability code book 54 are determined by the following processing.

In the probability code book 54, for example, 512 kinds of stochastic signal vectors corresponding to the subframe length (that is, 40 dimensions) are stored in advance, and an index is assigned to each. At this time, the switch 56 is in a closed state.

First, the multiplier 52 selects the optimum gain β for the optimum adaptive code vector determined by the above process.
And then sends it to the adder 53.

Next, each probability code vector from the probability code book 54 is multiplied by a variable gain value in the multiplier 55, and then input to the adder 53. The adder 53 adds the optimum adaptive code vector multiplied by the optimum gain β and each probability code vector, and inputs the addition result to the synthesis filter 58. The subsequent process is performed in the same manner as the process of determining the parameters (delay L and gain β) of the adaptive codebook 51 described above.

That is, the synthesizing filter 58 performs the synthesizing process using the linear prediction parameter α and sends the synthesized vector to the subtractor 62. The subtractor 62 subtracts the original speech vector and the synthesized vector and sends the obtained error vector to the perceptual weighting filter 63. The perceptual weighting filter 63 performs a weighting process on the error vector in consideration of the perceptual characteristic, and sends it to the error evaluator 64. The error evaluator 64 calculates the mean square of the error vector, searches for the probability code vector having the smallest mean square value, and sends the index i and the gain γ to the multiplexer 65. In this way, the index i and the gain γ of the probability codebook 54 are determined.

The multiplexer 65 quantizes the linear prediction parameter α, the delay L and the gain β of the adaptive codebook 51, and the index i of the probability codebook 54.
And gain γ are each multiplexed.

Further, in such a speech coding apparatus, as described above, the pitch periodicity for voiced sound is created by delaying the past excitation signal. However, since past sound source signals are originally made from noise sequences,
It becomes difficult to create a pulse sequence corresponding to a voiced sound source. Due to this influence, particularly in voiced sound, the reproduced sound contains a lot of high frequency noise, and the sound quality is deteriorated.
Various proposals have been made to solve this problem, one example of which is “DETAILS TO ASSIST INIMPLEMENTATI
ON OF FEDERAL STANDARD 1016 CELP ”(NATIONAL COMM
UNICATIONSSYSTEM, TECHNICAL INFORMATION BULLETIN 92
-1, PP.10-11, 1992).

FIG. 7 is a block diagram of a decoder corresponding to the speech coding apparatus of the code driven linear prediction system shown in FIG. In the figure, the adaptive codebook 70 is a multiplier 7
It is connected to the first input terminal of the adder 72 via 1. The probability code book 73 includes a multiplier 74 and a switch 75.
Is connected to the second input terminal of the adder 72 via. The output terminal of the adder 72 is connected to the adaptive codebook 70 via the delay circuit 76, and the output terminal 79
Is connected to the first input terminal of the synthesis filter 77.

The demultiplexer 78 also includes an adaptive codebook 70, a probability codebook 73, and a multiplier 7.
1, 74 and the second input terminal of the synthesis filter 77.

The composition of the synthesizing filter 77 is assumed to be the same as the composition of the synthesizing filter 58 shown in FIG.

In the above configuration, the demultiplexer 78 decomposes the received signal into the linear prediction parameter α, the delay L and the gain β of the adaptive codebook 70, the index i and the gain γ of the probability codebook 73,
A synthesis filter 77 for the decomposed linear prediction parameter α
Then, the delay L and the gain β are output to the adaptive codebook 70 and the multiplier 71, respectively, and the index i and the gain γ are output to the probability codebook 73 and the multiplier 74, respectively.

An adaptive code vector of the adaptive codebook 70 is selected based on the delay L of the adaptive codebook 70 output from the demultiplexer 78. Here, the adaptive codebook 70 is the adaptive codebook 5 in the encoding device.
It has the same contents as the contents of 1. That is, the past driving sound source signal is input to the adaptive codebook 70 via the delay circuit 76. The multiplier 71 amplifies the input adaptive code vector based on the received gain β, and the adder 7
Send to 2.

Next, the probability code vector of the probability code book 73 is selected based on the index i of the probability code book 73 output from the demultiplexer 78.
Here, the probability code book 73 has the same contents as the contents of the probability code book 54 in the encoding device. Multiplier 7
4 amplifies the input probability code vector based on the received gain γ and sends it to the adder 72.

The adder 72 adds the amplified probability code vector and the amplified adaptive code vector and sends them to the synthesis filter 77 and the delay circuit 76. The synthesis filter 77 performs synthesis processing using the received linear prediction parameter α as a coefficient, and outputs a synthesized speech signal from the output terminal 79.

[0024]

However, although the speech coding apparatus using the linear prediction analysis as described above can obtain high-quality coding performance at a relatively low bit rate, this apparatus can be used for non-speech. In the environment where signals, that is, background noises inevitably exist, for example, when used as a mobile telephone or a voice recording device, the background noises are mixed in the voice signals, and the quality of the encoded signals is greatly deteriorated. There was a problem.

The speech coding apparatus of the present invention has been made in view of such a problem, and an object of the speech coding apparatus is to adequately code a speech signal even in a noisy environment to obtain a speech signal of high sound quality. It is to provide a speech coding apparatus capable of obtaining

[0026]

In order to achieve the above object, the speech coding apparatus according to the first aspect of the invention has a frame-based input signal divided into predetermined frame intervals as a speech signal or a non-speech signal. A voice discriminating means for discriminating whether the signal is a voice signal, a linear predictive analyzing means for analyzing the input signal and outputting a spectrum parameter thereof, and a subframe dividing means for further dividing the frame interval of the input signal into predetermined subframe intervals. An adaptive codebook in which a plurality of signals created by delaying past sound source signals are stored in advance, a probability codebook in which a plurality of noise signal waveforms at the subframe intervals are stored, and the probability codebook, A driving sound source signal generating means for generating a driving sound source signal based on a signal output from both or one of the adaptive codebooks; A synthesis filter for synthesizing speech by using a signal output from the stochastic codebook and the adaptive codebook as a driving sound source signal based on the spectrum parameter, and a delay of the adaptive codebook that minimizes distortion of the synthesized signal with respect to the input signal. , A probability codebook index, and an error minimization means for outputting a code corresponding to each gain, a pitch periodicity analysis means for analyzing the pitch periodicity of an input signal, a discrimination result of the voice discrimination means, and the pitch. Gain adjusting means for increasing / decreasing the gain of either or both of the probability codebook and the adaptive codebook according to the analysis result of the periodicity analysis means is provided.

The speech encoding apparatus according to a second aspect of the present invention is the speech encoding apparatus according to the first aspect of the present invention, wherein in the drive excitation signal generation means, the speech discrimination means discriminates the input signal as a speech signal. Occasionally, a driving sound source signal is generated from the signals output from the stochastic codebook and the adaptive codebook, and when the sound discriminating means discriminates the input signal as a non-speech signal, the signal output only from the stochastic codebook. Generate a driving sound source signal from.

A speech coding apparatus according to a third aspect of the present invention is the speech coding apparatus according to the first invention, wherein the gain adjusting means is operable when the speech discrimination means discriminates the input signal as a non-speech signal. , The gain of the stochastic codebook and / or the adaptive codebook is attenuated at a predetermined ratio with respect to the case of discriminating the speech signal.

The speech coding apparatus according to a fourth aspect of the present invention is the speech coding apparatus according to the first or second aspect, wherein the speech discrimination means determines the speech depending on the energy level of the input signal for each frame. / Non-speech is discriminated, and a threshold determination means for determining a discrimination threshold according to the frame energy at the start of encoding is provided, and the difference between the current frame energy and the frame energy at the start of encoding is If it is larger than the discrimination threshold value determined by the threshold value determination means, it is determined as voice, and if it is less than it is determined as non-voice.

That is, the speech coder according to the first aspect of the invention discriminates by the speech discriminating means whether the input signal in frame units divided into a predetermined frame interval is a speech signal or a non-speech signal, and at the same time, it is linear. The predictive analysis unit analyzes the input signal and outputs the spectrum parameter thereof, and further, the frame interval of the input signal is divided into predetermined subframe intervals by the subframe dividing unit. Also, it is output from both or one of the adaptive codebook in which a plurality of signals created by delaying the past sound source signal are stored in advance and the probability codebook in which a plurality of noise signal waveforms at the subframe intervals are stored. The drive sound source signal generating means generates a drive sound source signal based on the signal. Then, based on the spectrum parameter, while synthesizing the speech by using a signal output from the stochastic codebook and the adaptive codebook as a driving sound source signal by a synthesizing filter, the error minimization means converts the synthesized signal to the input signal. The adaptive codebook delay that minimizes the distortion, the probability codebook index, and the code corresponding to each gain are output. Then, the pitch periodicity of the input signal is analyzed by the pitch periodicity analysis means, and both the probability codebook and the adaptive codebook are selected according to the discrimination result of the voice discrimination means and the analysis result of the pitch periodicity analysis means. Alternatively, one of the gains is increased or decreased by the gain adjusting means.

A speech coding apparatus according to a second invention is the speech coding apparatus according to the first invention, wherein when the speech discrimination means discriminates the input signal as a speech signal,
When the driving sound source signal generating means generates a driving sound source signal from the signals output from the stochastic codebook and the adaptive codebook, and when the sound discriminating means discriminates the input signal as a non-speech signal, only the stochastic codebook is generated. The drive sound source signal is generated from the signal output from the.

A speech coding apparatus according to a third aspect of the present invention is the speech coding apparatus according to the first invention, wherein when the speech discrimination means discriminates an input signal as a non-speech signal, the gain adjustment means is used. , The gain of the stochastic codebook and / or the adaptive codebook is attenuated at a predetermined ratio with respect to the case where it is discriminated as a voice signal.

A speech coding apparatus according to a fourth aspect of the invention is the speech coding apparatus according to the first, second or third invention, wherein the speech discrimination means is used to detect the input signal for each frame. Voice / non-voice is discriminated according to the amount of energy, and the discrimination threshold is determined by the threshold determination means according to the frame energy at the start of encoding.
Then, if the difference between the current frame energy and the frame energy at the start of encoding is larger than the discrimination threshold value determined by the threshold value determination means, it is determined as voice, and if it is smaller, it is determined as non-voice.

[0034]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of a speech coder to which the present invention is applied.

In the figure, the adaptive codebook 1 is connected to the first input terminal of the adder 3 as the driving sound source signal generating means via the multiplier 2 and the switch 19, and the stochastic codebook 4 is multiplied by the multiplier 5. And adder 3 via switch 6
Is connected to the second input terminal of. The output terminal of the adder 3 is connected to the first input terminal of the subtracter 12 via the synthesis filter 8 and also connected to the adaptive codebook 1 via the delay circuit 7.

Further, the buffer memory 9 connected to the input terminal 19 is the LPC analyzer 1 as the linear prediction analysis means.
6 is connected to the synthesizing filter 8 via 6, and the subtracter 12 is connected to the subtracter 12 via the subframe divider 17 as a subframe dividing means, and the voice discriminator 1 as a voice discriminating means.
1 to the gain adjuster 15 and further to the gain adjuster 15 as the gain adjusting means via the pitch periodicity analyzer 10 as the pitch periodicity analyzing means. The gain adjuster 15 is connected to the multiplier 5. The output terminal of the subtractor 12 is the perceptual weighting filter 13
Is connected to the input terminal of the error evaluator 14 as an error minimizing means. The output terminal of the error evaluator 14 is connected to the adaptive codebook 1, the probability codebook 4, and the multipliers 2 and 5.

Further, the multiplexer 18 is connected to the voice discriminator 11, the LPC analyzer 16 and the error evaluator 14.

FIG. 2 shows the structure of the voice discriminator 11 as the voice discriminating means shown in FIG. In the figure, the frame energy analysis circuit 120 is an adder 121.
Connected to the first input terminal of. The initial frame energy analysis circuit 122 is connected to a threshold value determining circuit 124 as a threshold value determining means and a second input terminal of the adder 121. Output terminal of adder 121 and threshold value determination circuit 1
24 is connected to the discrimination circuit 123.

In the above structure, from the input terminal 9
For example, an original audio signal sampled at 8 kHz is input, and a predetermined frame interval (for example, 20 m
The audio signal of s, that is, 160 samples) is stored in the buffer memory 9. The buffer memory 9 receives the input signal on a frame-by-frame basis from the LPC analyzer 16 and the sub-frame divider 1.
7, the voice discriminator 11, and the pitch periodicity analyzer 10. The LPC analyzer 16 performs a linear prediction (LPC) analysis on the input signal, extracts a linear prediction parameter α representing a spectral characteristic, and sends it to the synthesis filter 8 and the multiplexer 18. Subframe divider 17
Divides the input signal of a frame into predetermined subframe intervals (for example, 5 ms, that is, 40 samples). Here, sub-frame signals from the first sub-frame to the fourth sub-frame are created from the input signal of the frame.

The voice discriminator 11 discriminates whether the input signal of the frame is voice or non-voice by the following method. That is,
In the configuration shown in FIG. 2, the frame energy analysis circuit 120 calculates the frame energy E _f [dB] of the input frame input signal by the following formula.

[0042]

[Equation 1] However, s (n) is the input signal in sample n, and N is the frame length.

Also, the initial frame energy analysis circuit 1
22 is the frame energy E _b [dB] at the start of encoding
Is similarly calculated by the above formula.

The threshold determining circuit 124 determines a threshold according to the magnitude of the background noise energy based on the relationship between the background noise energy [dB] and the threshold [dB] as shown in FIG. To 123. Also, adder 1
At 21, the initial frame energy E _b [dB] is subtracted from the frame energy E _f [dB], and the subtraction result is sent to the discrimination circuit 123. Then, the determination circuit 123 compares the input subtraction result with a threshold value, and if the subtraction result is larger than the threshold value, determines that the frame input signal is a voice signal, and otherwise determines that it is a non-voice signal.

Returning to FIG. 1, when the voice discriminator 11 discriminates that the input signal is a voice signal, the switch 19 is closed. When it is determined that the input signal is a non-voice signal, the switch 19 is opened. By such a control operation, the driving sound source signal is generated from the signals output from the adaptive codebook 11 and the probability codebook 4 in the voice section, and the driving sound source signal is generated only from the probability codebook 4 in the non-voice section. This is because the adaptive codebook 1 only functions as another stochastic codebook in the non-speech section, and thus it hardly contributes to the improvement of sound quality. In the voice section, the delay L and the gain β of the adaptive codebook 1 are determined in the same manner as in the conventional example described above. Further, the index i and the gain γ of the probability codebook 4 are also determined in the same manner as the above-mentioned conventional example.

The pitch periodicity analyzer 10 analyzes the pitch periodicity of the frame input signal. In the present embodiment, for example, the cross-correlation between a signal (pitch prediction signal) obtained by multiplying the selected adaptive code vector by β and the input signal is calculated. That is, if the value of this cross-correlation is high, it can be said that the voiced sound has higher periodicity, and conversely, if the value of the cross-correlation is low, it can be said that it is unvoiced or non-voiced.

In calculating the cross-correlation, the generalized cross-correlation R between the input signal and the pitch prediction signal expressed by the following equation is used here.

[0048]

[Equation 2] However, s (n) is the input signal in sample n as described above, and p (n) is the pitch prediction signal in sample n.

The gain adjuster 15 is the pitch periodicity analyzer 1
The value of the generalized cross-correlation R as the analysis result of 0 and the discrimination result v / uv (v is voice (voice)) of the voice discriminator 11.
, And uv means increase or decrease the gain of both or one of the probability codebook 4 and the adaptive codebook 1 in accordance with unvoice (non-voice). In the present embodiment, when it is determined that the signal is a non-voice signal, FIG.
Based on the relationship between the generalized cross-correlation R [dB] and the gain multiplication factor of the probability codebook as shown in, the gain γ of the voice probability codebook 4 is set at a predetermined ratio with respect to when it is determined to be a voice signal. Try to attenuate it.

By such processing, the normal pitch enhancement processing is performed in the voice section, and the gain γ of the probability codebook 4 is attenuated in the non-voice section as shown in FIG. 4, so that the background noise is suppressed. You can

The multiplexer 18 receives the quantized linear prediction parameter α, the delay L and the gain β of the adaptive codebook 1, the index i and the gain γ of the probability codebook 4, and the speech discrimination information v / uv. It is multiplexed and transmitted.

Next, the decoding operation of the speech decoding apparatus corresponding to the above speech coding apparatus will be described in detail with reference to the drawings.

FIG. 5 is a block diagram of a speech decoding apparatus corresponding to the speech encoding apparatus of FIG. In the figure, the adaptive codebook 30 is connected to a first input terminal of an adder 33 via a multiplier 31 and a switch 32. The probability code book 36 is connected to the second input terminal of the adder 33 via the multiplier 37 and the switch 38. The output terminal of the adder 33 is connected to the adaptive codebook 30 via the delay circuit 40, and is also connected to the first input terminal of the synthesis filter 34 having the output terminal 39.

The demultiplexer 35 also includes an adaptive codebook 30, a probability codebook 36, and a multiplier 3
1, 37 and the second input terminal of the synthesis filter 34.

In the above configuration, the demultiplexer 35 processes the received signal by using the linear prediction parameter α, the delay L and the gain β of the adaptive codebook 30, the index i and the gain γ of the probability codebook 36, and the voice discrimination information v. / Uv and decomposed linear prediction parameter α into synthesis filter 34, delay L and gain β into adaptive codebook 30 and multiplier 31, index i and gain γ into probability codebook 36, respectively. In the multiplier 37,
The voice discrimination information v / uv is output to the switch 32.

Then, the opening / closing operation of the switch 32 is controlled based on the voice discrimination information v / uv output from the demultiplexer 35. That is, if the voice discrimination information v / uv indicates that it is a voice signal, the switch 32 is closed and the information from the adaptive codebook 30 is used. on the other hand,
If the voice discrimination information v / uv indicates that it is a non-voice signal, the switch 32 is opened to make the adaptive codebook 30 unused.

The adaptive code vector of the adaptive codebook 30 is selected based on the delay L of the adaptive codebook 30 output from the demultiplexer 35. Here, adaptive codebook 30 has the same content as adaptive codebook 1 in the speech coding apparatus shown in FIG. That is, the past driving sound source signal is input to the adaptive codebook 30 via the delay circuit 40. The multiplier 31 amplifies the input adaptive code vector by the received gain β and sends it to the adder 33.

The probability code vector of the probability code book 36 is selected based on the index i of the probability code book 36 output from the demultiplexer 35. Here, the probability code book 36 has the same contents as the contents of the probability code book 4 in the speech coding apparatus shown in FIG. The multiplier 37 amplifies the input probability code vector by the received gain γ and sends it to the adder 33.

The adder 33 adds the amplified adaptive code vector and the amplified probability code vector and sends them to the synthesis filter 34 and the delay circuit 40. The synthesis filter 34 performs a synthesis process using the received linear prediction parameter α as a coefficient, and outputs a synthesized speech signal.

As described above, in this embodiment, the ratio of gain adjustment in the pitch enhancement processing is switched between the voice section and the non-voice section. That is, when it is determined that the input signal is a voice signal, normal pitch enhancement processing is performed, but when it is determined that the input signal is a non-voice signal, the gain of the probability codebook is attenuated. Therefore, even in a noisy environment, the audio signal can be properly encoded to obtain a high-quality audio signal. Further, since the threshold decision circuit 124 in the voice discriminator 11 decides the discrimination threshold according to the frame energy at the start of encoding,
The threshold value is determined according to the size of the background noise, and the voice / non-voice can be discriminated more accurately.

[0061]

According to the first, second and third aspects of the present invention, it is possible to obtain a high-quality audio signal by satisfactorily encoding the audio signal even in a noisy environment.

According to the invention described in claim 4, in addition to the effect of the invention described in claim 1, there is an effect that the voice / non-voice can be discriminated more accurately.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a speech coding apparatus to which the present invention is applied.

FIG. 2 is a diagram showing a configuration of a voice discriminator shown in FIG.

FIG. 3 is a diagram showing a relationship between background noise energy and a threshold value.

FIG. 4 is a diagram showing a relationship between a generalized cross-correlation and a gain factor of a probability codebook.

5 is a diagram showing a configuration of a speech decoding apparatus corresponding to the speech encoding apparatus shown in FIG.

FIG. 6 is a block diagram of a conventional audio encoding device.

7 is a diagram showing a configuration of a speech decoding apparatus corresponding to the speech encoding apparatus shown in FIG.

[Explanation of symbols]

1 ... Adaptive codebook, 2, 5 ... Multiplier, 3 ... Adder,
4 ... Probability code book, 6, 19 ... Switch, 7 ... Delay circuit, 8 ... Synthesis filter, 9 ... Buffer memory, 10 ...
Pitch periodicity analyzer, 11 ... Speech discriminator, 12 ... Subtractor, 13 ... Perceptual weighting filter, 14 ... Error evaluator,
15 ... Gain adjuster, 16 ... LPC analyzer, 17 ... Subframe divider, 18 ... Multiplexer.

Claims (4)

[Claims] 1. A voice discriminating means for discriminating whether a frame unit input signal divided into predetermined frame intervals is a voice signal or a non-voice signal, and a linear which analyzes the input signal and outputs a spectrum parameter thereof. Prediction analysis means, subframe division means for further dividing the frame interval of the input signal into predetermined subframe intervals, and an adaptive codebook in which a plurality of signals created by delaying past sound source signals are stored in advance. , A stochastic codebook storing a plurality of noise signal waveforms at the subframe intervals, and a driving sound source signal generation for generating a driving sound source signal based on a signal output from either or both of the stochastic codebook and the adaptive codebook. Means, and the stochastic codebook and the adaptive codebook based on the spectral parameters. A synthesis filter for synthesizing speech with the signal output from as the driving sound source signal, an adaptive codebook delay that minimizes distortion of the synthesized signal with respect to the input signal, a probability codebook index, and a code corresponding to each gain. Error minimizing means for outputting, pitch periodicity analyzing means for analyzing pitch periodicity of the input signal, the probability codebook according to the discrimination result of the voice discriminating means and the analysis result of the pitch periodicity analyzing means. A speech coding apparatus comprising: a gain adjusting unit that increases or decreases the gain of both or one of the adaptive codebooks. 2. The driving sound source signal generating means generates a driving sound source signal from the signals output from the probability code book and the adaptive code book when the sound judging means judges the input signal as a sound signal, 2. The speech encoding apparatus according to claim 1, wherein when the speech discriminating means discriminates the input signal as a non-speech signal, a driving excitation signal is generated from a signal output from only the probability codebook. 3. The gain adjusting means, when the voice discriminating means discriminates an input signal as a non-voice signal, both the probability codebook and the adaptive codebook at a predetermined ratio with respect to the case of discriminating the input signal as a voice signal, or The speech coding apparatus according to claim 1, wherein one of the gains is attenuated. 4. The speech discrimination means discriminates speech / non-speech according to the energy level of an input signal for each frame, and a threshold for determining a discrimination threshold according to the frame energy at the start of encoding. A deciding means is provided, wherein if the difference between the current frame energy and the frame energy at the start of coding is larger than the discrimination threshold decided by the threshold deciding means, it is voice, and if it is smaller, it is non-voice. The speech coding apparatus according to Item 1, 2 or 3.