JPS6117197A - Voice analysis/synthesization method and apparatus - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a speech analysis method and apparatus for analyzing speech, extracting feature parameters, and synthesizing speech based on the feature parameters.

[Technical background of the invention]

Speech analysis and synthesis apparatuses using analysis and synthesis methods such as LPG, PARCOR, l-, SP, and formant systems compress and store speech information in order to analyze speech and extract feature parameters. The original speech is then synthesized from these feature parameters. Furthermore, in order to compress the amount of information, the extracted feature parameters are quantized and the quantized parameters are stored.

During this quantization, linear transformation is often performed in which the feature parameter and the quantization parameter have a linear relationship. If N bits are allocated to the quantization parameter, the value of the quantization parameter is expressed as a positive number and ranges from 0 to 2N-1. However, due to the statistical nature of the distribution of feature parameters, there are cases where there are very few melons that take values around 0 and around 2N-1. At this time, if the number of bits N of the quantization parameter is decreased, 2N-1 for the new number of bits N
Since the feature parameters are distributed in a range exceeding , it is impossible to simply reduce the number of bits N, resulting in a useless range and requiring an extra amount of information to store and transmit the feature parameters.

Taking this into consideration, some or all of the feature parameters are non-linearly converted and converted into descendants, and the feature parameters are created using quantization parameters with the minimum number of bits necessary. However, in the case of nonlinear quantization in this manner, it was necessary to perform complex inverse nonlinear transformation when restoring the quantization parameters to feature parameters. Normally, in an LSI-based speech analysis and synthesis device, this complicated inverse conversion is implemented using a ROM, which poses a problem of increasing the circuit scale.

[Object of the Invention] The present invention was made in consideration of the above circumstances, and it is possible to perform quantization with less distortion using a small amount of information, and to obtain high-quality synthesized sound without increasing the circuit scale. It is an object of the present invention to provide a speech analysis and synthesis method A5 and a device that can perform the following.

(Summary of the Invention) In order to achieve the above object, a speech analysis method according to the present invention detects whether some parameters in quantization parameters of speech feature parameters are a specific pattern, and detects whether or not a specific pattern is detected. Based on whether or not the minimum necessary parameter part of the quantization parameter is converted, the conversion parameter or the mother childization parameter is switched, and the switched parameter is used as the synthesis parameter to synthesize speech. It is said that

Further, the speech synthesis device according to the present invention includes a pattern detection means for detecting whether some parameters in the quantization parameters are a specific pattern, and converting other parameters in the quantization parameters. A conversion means for outputting a parameter, a switching means for switching between a mothering parameter and a conversion parameter based on the detection result of the pattern detection means, and a synthesis means for synthesizing speech using the parameter switched by the switching means as a synthesis parameter. It is characterized by:

(Embodiment of 5R, Bright) The present invention will be explained below based on the illustrated embodiment.
The figure shows a speech analysis and synthesis device according to an embodiment of the present invention.

The audio signal input to the analysis circuit 1o is LPG, PAR
Analysis such as COR, LSP, and Ltz method is performed, and feature parameters 11 are output. This characteristic parameter 11 has M bits and is input to a quantization circuit 12 and a comparison circuit 13. The quantization circuit 12 quantizes the input feature parameters 11 of M pits and outputs N-bit quantization parameters 15. The comparison circuit 13 converts the control signal 14 based on the value of the feature parameter 11 into the quantization circuit 12.
The quantization circuit 12 is switched.

The quantization parameters 15 obtained by further compressing the information of the feature parameters 11 in this way are stored in the memory 2o.

Speech is synthesized from the maternalization parameters stored in the memory 20 as follows. The upper N' bits of the quantization parameter 21 from the memory 2o are input to the pattern detection circuit 31, and the lower N-N' bits are input to the conversion ROM 32. Pattern detection circuit 31 is maternalization parameter 2
It is detected whether the upper N' bits of 1 are in a specific pattern or not, and the detection signal 27 is outputted to the switching circuit 34. On the other hand, the conversion RO'M32 is the lower N-N' of the maternalization parameter.
A conversion parameter 24 of M bits is output. The bit expansion circuit 33 outputs a predetermined additional parameter 25 of the M-N pin I-.
is added to the quantization parameter 21, and the M-bit parameter 26 is input to the switching circuit 34. The switching circuit switches between the parameter 26 and the conversion parameter 24 and outputs it as an M-bit composite parameter 27.

When a specific pattern is detected by the pattern detection circuit 31, the parameter 26 is selected by the detection signal 27 and output as a synthesis parameter 27. Conversely, if a specific pattern is not detected, the conversion parameter 24 is selected and output as the synthesis parameter 27. The synthesis circuit 35 synthesizes speech based on the synthesis parameters 27.

Next, a more detailed explanation will be given using specific examples shown in FIGS. 2(a), 2(b), and 2(c). Here, information is compressed from a 4-bit feature parameter to a 3-bit quantization parameter. In this specific example, each parameter is treated as an unsigned integer.

The characteristic parameters shown in FIG. 2 <a> take values from 2 to 9. The quantization circuit 12 quantizes the quantization parameters from the feature parameters. In this specific example, the quantization is 4
This is done by deleting the most significant bit of the bit characteristic parameter (FIG. 2(b)). For speech synthesis, the quantization parameter 21 stored in the memory 2o
The upper two bits of the data are input to the pattern detection circuit 31. The pattern detection circuit 31 detects whether the top two hits are the pattern rooJ or not, and outputs a different detection signal 27 to the switching circuit 35 based on the detection result. - The force conversion ROM 32 inputs the lower 1 bit of the quantization parameter 21 and outputs a 4-pit conversion parameter 24 as shown in FIG. 2(C). Further, the bit expansion circuit 33 is “0”.
This additional parameter 25 is added to the quantization parameter 21 to become a 4-bit parameter 26.

The switching circuit 34 switches between the parameter 26 and the conversion parameter 26 based on the detection signal 27. That is, when the pattern roOJ is detected, the conversion parameter is switched to 24, and when the pattern "00" is not detected,
Switch to parameter 26. As a result, a synthesis parameter 27 shown in FIG. 2(a) is generated from the quantization parameter 21. In this way, a 4-bit synthesis parameter can be obtained from a 3-bit Φ-conversion parameter.

Next, this embodiment will be described in more detail based on other specific examples shown in FIGS. 3(a), 3(b), and 3(c). This specific example is from 2 to
Information is compressed from a 5-bit feature parameter that takes a value of 16 to a 3-bit quantization parameter. This quantization is performed by a quantization circuit 12 and a comparison circuit 13. The comparison circuit 13 compares the value of the feature parameter 11 with "17", and
If the value is less than 7, the upper two bits of the 5-bit feature parameter 11 are deleted, and if it is greater than 7, roo is deleted as shown in FIG. 3(b).
Output OJ and roolJ. For speech synthesis,
Top 2 of quantization parameters 21 stored in memory 2o
The bits are input to the pattern detection circuit 31. The pattern detection circuit 31 detects whether the upper two bits are the pattern "oO" or not, and outputs a different detection signal 27 to the switching circuit 35 based on the detection result. On the other hand, i-conversion ROM
32 inputs the lower 1 bit of the quantization parameter 21,
5-bit conversion parameter 2 as shown in Figure 3(C)
Outputs 4. Further, the bit expansion circuit 33 outputs a 2-bit additional parameter 25 of "00", and this additional parameter 25 is added to the quantization parameter 21 to become a 5-bit parameter 26.

The switching circuit 34 switches between the parameter 26 and the conversion parameter 24 based on the detection signal 27. That is, when the pattern "00J" is detected, the conversion parameter is switched to 24, and when the pattern "OO" is not detected, the conversion parameter is switched to the parameter 26. As a result, a synthesis parameter 27 shown in FIG. 2(a) is generated from the quantization parameter 21.

In this way, it is possible to obtain a 5-bit composite parameter makeup that has been nonlinearly transformed from 3-bit quantization parameters.

As described above, according to this embodiment, it is possible to generate synthesis parameters from quantization parameters using an extremely simple circuit, which is effective. In particular, a G conversion ROM may be used for the nonlinear portion, and nonlinear conversion can be performed with a small capacity conversion ROM. Furthermore, quantization can be performed by arbitrarily combining linear and nonlinear methods.

For example, when analyzing and synthesizing speech using the RA RCOR method, conventionally pitch data is linearly quantized using 7 bits. In this case, t! from 0 to 127! The surrounding pitch data can be quantized. However, when actually analyzing human voices, pitch data below 15 is never obtained, and the range from 0 to 15 is wasted. On the other hand, a low-pitched male voice may have pitch data of 130 or more, and it has not been possible to correctly quantize this in the past. However, according to the above embodiment, even when quantizing with the same 7 bits, 16 to 1
Pitch data in a range of 43 can be quantized, and a high-quality synthesized sound can be obtained without increasing the amount of information of quantization parameters and with almost no increase in circuit scale.

For example, a feature parameter that takes a value of 16 to 160 is set to 7.
To quantize in bits, do as follows. The feature parameters 16 to 127 are quantized as they are, and the values 128 to 160 are logarithmically compressed to 0 to 15.
Convert it to the value of and use it as the quantization parameter. When generating synthesis parameters, the pattern detection device 31 detects the quantization parameters from 0 to 15, and the conversion ROM 32 performs exponential expansion conversion to obtain synthesis parameters from 16 to 160.

〔Effect of the invention〕

As described above, according to the present invention, quantization with little distortion can be achieved with a small amount of information without increasing the circuit scale.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a speech analysis and synthesis device according to an embodiment of the present invention, and FIGS. 2(a), (b), and (C) show characteristic parameters, synthesis parameters, quantization parameters, and
A diagram showing a specific example of a conversion ROM, FIG. 3 (a>, (b),
(C) is a diagram showing another specific example of characteristic parameters, synthesis parameters, quantization parameters, and conversion ROM of the speech analysis and synthesis device. 10...Analysis circuit, 11...Characteristic parameter, 12
Quantization circuit, 13 Comparison circuit, 15 Quantization parameter, 20 Memory, 21 Quantization parameter, 27 Synthesis parameter, 31 Pattern detection circuit , 32... Conversion ROM, 33... Bit expansion circuit, 34... Switching circuit, 35... Synthesizing circuit. Figure 1

Claims (1)

[Claims] 1. The feature parameters are quantized so that some of the parameters have a specific pattern according to the values of the feature parameters obtained by analyzing the voice, and some of the quantized parameters are detecting whether or not it is a specific pattern, and switching between a conversion parameter obtained by converting at least a necessary minimum parameter part of the quantization parameter and the quantization parameter based on whether or not the specific pattern is detected; , A speech analysis and synthesis method characterized in that speech is synthesized using the switched parameters as synthesis parameters. 2. analysis means for analyzing voice and extracting feature parameters; quantizing the feature parameters so that some of the parameters have a specific pattern according to the value of the feature parameters extracted from the analysis means; quantization means for generating quantization parameters; pattern detection means for detecting whether some of the parameters in the quantization parameters are a specific pattern; and inputting other parameters in the quantization parameters. a conversion means for outputting a conversion parameter according to the other parameters; a switching means for switching between the quantization parameter and the conversion parameter according to the detection result of the pattern detection means; A speech synthesis analysis device comprising: synthesis means for synthesizing speech using parameters obtained as synthesis parameters. 3. The speech analysis and synthesis device according to claim 2, wherein the upper bits in the quantization parameter are used as the part of the parameter. 4. The speech analysis and synthesis device according to claim 2 or 3, wherein the number of bits of the synthesis parameter is larger than the number of bits of the quantization parameter. 5. analysis means for analyzing voice and extracting feature parameters; quantizing the feature parameters so that some of the parameters have a specific pattern according to the value of the feature parameters extracted from the analysis means; quantization means for generating a quantization parameter; pattern detection means for detecting whether or not some of the parameters in the quantization parameter are a specific pattern; and a pattern detection means for adding a predetermined parameter to the quantization parameter. , a bit extension means for outputting an extended parameter obtained by extending the quantization parameter; a conversion means for inputting another parameter in the quantization parameter and outputting a transformation parameter according to the other parameter; and the pattern detection. A speech synthesis and analysis device comprising: switching means for switching between the extended parameter and the conversion parameter according to a detection result of the switching means; and a synthesis means for synthesizing speech using the parameters switched by the switching means as synthesis parameters. 6. The speech analysis and synthesis device according to claim 5, wherein the upper bits in the quantization parameter are used as the part of the parameter. 7. The speech analysis and synthesis device according to claim 5 or 6, wherein the number of bits of the synthesis parameter is greater than the number of bits of the quantization parameter.