JPH0969000A - Voice parameter quantizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce a quantization error when a characteristic parameter of a sound is quantized. SOLUTION: A voice/voiceles decision part 1 deciding whether an input voice signal a to be coded is a voice or voiceless is provided, and when the voiceless is changed into the voice, the characteristic parameter c is quantized as it is using a single code book 4 in a multiplex mode quantizer 3, and when the sound is continued, the quantization error is reduced by quantizing a difference between the quantization characteristic parameter of the last time and the characteristic parameter c to be quantized this time. Further, a threshold value decision part deciding whether or not the quantization characteristic parameter outputted from the multiplex mode quantizer 3 exists within the certain range is provided in the constitution, and when it is decided that the quantization characteristic parameter doesn't exist in the threshold values, the quantization error is reduced by quantizing with the multiplex quantizer 3 by using the single code book 4.

Description

Detailed Description of the Invention

[0001]

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

[0002]

2. Description of the Related Art In recent years, a voice coding technique has been put into practical use, in which a voice signal is converted into a digital signal, and further data compression is performed to record and reproduce a voice coded signal converted into a small amount of data. Among these techniques, there is one that quantizes a parameter extracted from a feature of a voice signal using a codebook.

FIG. 4 is a block diagram showing the structure of a conventional speech parameter quantizer as described above. In FIG. 4, 101 is a feature parameter extraction unit, 102 is a quantizer, and 103 is a single codebook. Further, a is an input voice signal, c is a characteristic parameter, d is a single codebook data, and f is a transmission parameter.

The operation of the speech parameter quantizer constructed as above will be described. First, the feature parameter extraction unit 101 extracts a feature of the input voice signal a as a voice signal, and transmits it as a feature parameter c to the quantizer 102. The quantizer 102 quantizes the characteristic parameter c using the single codebook 103 and outputs it as the transmission parameter f. When decoding the voice, the decoded voice is created using the transmission parameter f and the single codebook 103.

[0005]

However, in the above-mentioned conventional speech parameter quantizer, the codebook used for quantizing the characteristic parameter needs to substantially satisfy the distribution range of the characteristic parameter. The codebook created so as to satisfy the condition of 1 has a problem that the quantization error is large. In general, it is known that a speech signal has a high correlation with a nearby speech signal. This can be said also about a characteristic parameter, and a difference between a certain characteristic parameter and a characteristic parameter calculated immediately before that is quantized. If the transmission parameter is used, quantization with less quantization error is possible. However, in this case, since the content of the codebook used is a very small data sequence as compared with the conventional codebook, the quantization error may increase when the characteristics of the audio signal change. For example, in the case of quantizing the power of a voice signal as a characteristic parameter, when there is a change from silence to voice, there is a problem that the difference between the characteristic parameters becomes large and the quantization error becomes large.

The present invention solves the above-mentioned conventional problems, and provides a speech parameter quantizer capable of reducing a quantization error in quantizing a speech feature parameter in a speech parameter quantizer. The purpose is to provide.

[0007]

In order to achieve the above object, a first structure of the present invention is to provide a voice / sound determination unit for determining whether an input voice signal is voiced or not, and a voice from the input voice signal. A characteristic parameter extraction unit for extracting the characteristic parameter of
When extracting characteristic parameters, a multimode quantizer that switches the quantization method based on the information from the voice / sound determination unit, a single codebook created in consideration of the distribution of the data to be quantized, and the quantization It is provided with a difference codebook created in consideration of the difference between the data and the previous data of the data.

A second structure of the present invention is, in addition to the above structure,
It is provided with means for changing the quantization method of the multimode quantizer according to the data calculated in the speech parameter quantizer.

[0009]

The present invention can realize the speech parameter quantizing device having the above-mentioned first configuration, which can reduce the quantization error of the characteristic parameter generated in the speech parameter quantizing device, particularly in the voiced section.

Further, according to the second aspect of the present invention, in the speech parameter quantizer, the speech parameter quantizer capable of reducing the quantization error of the characteristic parameter generated in the speech parameter quantizer, especially when the power fluctuation is large in the voiced section. Device can be realized.

[0011]

【Example】

(First Embodiment) A first embodiment of the present invention will be described below with reference to the drawings. FIG. 1 shows the configuration of a speech parameter quantizer according to the first embodiment of the present invention. In FIG. 1, reference numeral 1 denotes a sound / sound determination unit that determines whether the input sound signal is sounded or not, based on the power of the input sound signal.
Reference numeral 2 is a feature parameter extraction unit that extracts a voice feature parameter from an input voice signal, and 3 is a multi-mode quantizer that can change the quantization method according to an instruction from the voiced / unvoiced determination unit 1 when the feature parameter is quantized. 4 is a single codebook created in consideration of the distribution of data to be quantized, 5 is a difference optimized for quantizing the difference between the characteristic parameter to be quantized and the characteristic parameter calculated immediately before it. It is a codebook. Further, a is an input voice signal, b is a voiced / unvoiced flag, b
2 is the old voiced / unvoiced flag calculated immediately before the voiced / unvoiced flag b, c is the characteristic parameter, d is the independent codebook data, and e is
Is differential codebook data, and f is a transmission parameter.

The operation of the speech parameter quantizer configured as described above will be described below with reference to FIG. First, the voiced / unvoiced determination unit 1 determines whether the input audio signal a is voiced or not, and transmits the determination result to the multimode quantizer 3 as a voiced / unvoiced flag b. In addition, the same input voice signal a is analyzed in the characteristic parameter extraction unit 2,
The characteristic parameter c is calculated. Next, the multi-mode quantizer 3 quantizes the characteristic parameters. At this time, the voiced / unvoiced flag b transmitted to the multimode quantizer 3 is transmitted.
And the value of the old voiced / unvoiced flag b2 determine the quantization method. In the case where the old voiced / unvoiced flag b2 is silent and the voiced / unvoiced flag b indicates voiced, the multimode quantizer 3
Quantizes the characteristic parameter c as it is using the single codebook 4. Further, when the old voiced / non-voiced flag b2 indicates voiced and the voiced / non-voiced flag b indicates voiced, the multimode quantizer 3 quantizes the characteristic parameter c from the previously quantized value. The value obtained by subtracting the value of is quantized using the differential codebook 5. In addition, regardless of the value of the old sound / silence flag b2, when the sound / silence flag b indicates no sound, the multimode quantizer 3 uses the single codebook 4 to quantize the feature parameter c as it is. The multimode quantizer 3 outputs the codebook number selected at the time of quantization as the transmission parameter f. When decoding the voice, it is judged from the voiced / unvoiced flag b and the old voiced / unvoiced flag b2 which method the multimode quantizer 3 quantized, and based on this, the single codebook 3 or the differential code. A decoded voice is created using the book 4, the transmission parameter f, and the characteristic parameter decoded immediately before.

As described above, according to the present embodiment, it is possible to reduce the quantization error of the characteristic parameter that occurs in the voice parameter quantization device during voice encoding, particularly in the voiced section.

(Embodiment 2) Next, a second embodiment of the present invention will be described with reference to the drawings. FIG. 2 shows the configuration of a speech parameter quantizer according to the second embodiment of the present invention. In FIG. 2, the presence / absence determination unit 1, the characteristic parameter extraction unit 2, the multimode quantizer 3,
The individual codebook 4, the differential codebook 5, the input voice signal a, the sound / silence flag b, the old sound / silence flag b2, the characteristic parameter c, the single codebook data d, the differential codebook data e, and the transmission parameter f are implemented. Same as Example 1. Reference numeral 6 denotes a threshold value determination unit that determines whether or not the value of the quantized characteristic parameter exceeds a certain threshold value. Further, g is a quantization feature parameter quantized by the multimode quantizer 3, and h is a mode switching flag.

The operation of the speech parameter quantizer configured as described above will be described below with reference to FIG. First, the voiced / non-voiced sound determination unit 1 determines whether the input audio signal a is voiced or non-voiced, and transmits the determination result to the multimode quantizer 3 as a voiced / non-voiced flag b. Further, the same input voice signal a is analyzed by the characteristic parameter extraction unit 2 to calculate the characteristic parameter c. Next, the multimode quantizer 3
At, the feature parameters are quantized. At this time, the voiced / unvoiced flag b transmitted to the multimode quantizer 3 is transmitted.
Then, the quantization method is determined by the values of the old voiced / unvoiced flag b2 and the mode switching flag h output from the threshold value determination unit 6. When the old voiced silence flag b2 indicates silence and the voiced silence flag b indicates voiced, the multimode quantizer 3 uses the independent codebook 4 to quantize the feature parameter as it is. Further, when the old voiced / non-voiced flag b2 indicates voiced and the voiced / non-voiced flag b indicates voiced, the multi-mode quantizer 3 determines the characteristic parameter to be quantized this time from the previously quantized value of the characteristic parameter. The subtracted value is quantized using the differential codebook 5. In addition, the sound / silence flag b
If the mode switching flag h indicates to use the single codebook 4, the multimode quantizer 3 uses the single codebook 4 regardless of the value of the old voiced silence flag b2. The feature parameter is quantized as it is. The multimode quantizer 3 outputs the codebook number selected at the time of quantization as the transmission parameter f, and the quantized feature parameter value as the quantized feature parameter g. The threshold determination unit 6 determines whether the value of the quantization feature parameter g is within a certain range,
If it is determined that it is out of the range, the mode switching flag h
Is converted into a signal instructing the use of the single codebook 4 and transmitted to the multimode quantizer 3. When decoding voice, the multi-mode quantizer 3 uses the method for quantizing the voiced / unvoiced flag b, the old voiced / unvoiced flag b2, and whether the previously decoded feature parameter exceeds a threshold value. It is determined whether or not the single codebook 4 or the differential codebook 5 is used,
Decoded speech is created using the transmission parameter f and the characteristic parameter decoded immediately before.

Next, a case where the gain of the adaptive codebook is quantized will be described with reference to FIG. 3 as an example of use of this embodiment. FIG. 3 shows the state of gain of an adaptive codebook used for CELP and the like. The gain has a value that the adaptive codebook gain is considerably larger than 1 when the voice signal to be encoded changes from silence to voice, and is close to 1 when the voice signal to be encoded is in a steady state. It is generally known to take. It is assumed that the speech parameter quantizer given as an example performs silence compression, and does not encode the speech signal when there is silence.

In state 1 of FIG. 3, the audio signal to be encoded is silent. At this time, since the voiced / unvoiced flag b indicates no sound, the gain is not quantized at this time. In the state 2, the voice signal to be encoded changes from silence to voice. At this time, since the voiced / unvoiced flag b indicates voiced and the old voiced / unvoiced flag b2 indicates no voice, the multimode quantizer 3 uses the independent codebook 4 to quantize the gain as it is. In state 3, the voice signal to be encoded remains voiced. At this time, since the voiced / unvoiced flag b and the old voiced / unvoiced flag b2 indicate voice, the multimode quantizer 3 uses the differential codebook 5 to quantize the gain. In state 4, the audio signal to be encoded remains voiced. At this time, the voiced / non-voiced flag b and the old voiced / non-voiced flag b2 indicate voice, but since the quantization characteristic parameter g when quantized immediately before exceeds the threshold value, the mode switching flag h is Single codebook 4
To use the multimode quantizer 3
Quantizes the gain using the single codebook 4. In state 5, the audio signal to be encoded remains voiced. At this time, since the voiced / unvoiced flag b and the old voiced / unvoiced flag b2 indicate voice, the multimode quantizer 3 uses the differential codebook 5 to quantize the gain.

As described above, according to the present embodiment, a voice that takes a constant value at the time of voice coding, particularly when the voice power is in a steady state, such as the gain of an adaptive codebook for CELP coding. In the speech parameter quantizer for quantizing the parameters, it is possible to reduce the quantization error of the characteristic parameter that occurs when the power fluctuation is large in the voiced section.

[0019]

As described above, according to the present invention, by changing the quantizing method of the quantizer in consideration of the result of the voiced / unvoiced judgment, the quantization when the parameter extracting the feature of the voice signal is quantized. The conversion error can be reduced.

Further, according to the present invention, the feature of the voice signal is extracted by changing the quantization method of the quantizer in consideration of the result of the voiced / unvoiced determination and the value of the parameter for extracting the feature of the quantized voice. It is possible to reduce the quantization error when quantizing these parameters.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a speech parameter quantizer in a first embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a speech parameter quantization device according to a second embodiment of the present invention.

FIG. 3 is a characteristic diagram showing a usage example of a second embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a speech parameter quantization device in a conventional example.

[Explanation of symbols]

 1 voiced / unvoiced determination unit 2 feature parameter extraction unit 3 multi-mode quantizer 4 individual codebook 5 differential codebook 6 threshold determination unit 7 quantizer a input voice signal b voiced silence flag b2 old voiced silence flag c feature parameter d single codebook data e difference codebook data f transmission parameter g quantization feature parameter h mode switching flag

Claims (2)

[Claims] 1. A voice parameter quantizer for converting a voice signal into a digital signal and quantizing the voice signal, comprising a voiced / non-voiced determination unit for determining whether the input voice signal is voiced or not, and a voice from the input voice signal. A characteristic parameter extraction unit for extracting a characteristic parameter, and when quantizing the characteristic parameter,
A multi-mode quantizer that switches the quantization method according to the information from the voiced / unvoiced determination unit, a single codebook created in consideration of the distribution of the data to be quantized, the data to be quantized, and the data thereof. A speech parameter quantization apparatus comprising: a difference codebook created in consideration of a difference between previous data. 2. The speech parameter quantizer according to claim 1, further comprising means for changing a quantization method of the multimode quantizer according to at least data calculated in the speech parameter quantizer.