JPH03245198A - Voice analyzing and synthesizing device - Google Patents

Abstract

PURPOSE:To reproduce a high-quality voice which is stable and has high distinctness by transmitting the whole basic analysis frame of a transmission frame by a variable length frame expressing the same with plural pieces of representative analysis frames. CONSTITUTION:A representative frame deciding means 109 decides the representative analysis frame candidate forming the series array of the interpolation LPC (Linear Prediction Coding) having the optimum min. distortion quantity to be set by taking the synthesized voice and data transmission quantity into account from m-pieces of min. distortion quantities varying from each other obtd. by min. distortion quantity determining means 105 to 107 as the representative analysis frame and expresses n-pieces of the analysis frames as the variable length frame with the quantity and number of the representative analysis frames. The deterioration in the quality of the synthesized voice is suppressed in this way and the high-quality synthesized voice is obtd.

Description

[Detailed Description of the Invention] [Industrial Field of Application] The present invention relates to a speech analysis and synthesis device, and in particular, it converts LPC coefficients, which are characteristic parameters of input speech, to be supplied from the analysis side to the synthesis side together with sound source information into variable length frames. The present invention relates to a speech analysis and synthesis device for transmitting data.

[Conventional technology]

Conventionally, in this type of speech synthesis analysis device, on the analysis side that analyzes input speech, the speech signal is converted into, for example, lQ m SEC.
LPC analysis is performed for each basic analysis frame, and the LPC coefficients obtained as a result and sound source information such as pitch period and power information are transmitted to the synthesis side to synthesize input speech. In this case, in order to compress the amount of audio information to be transmitted as much as possible, the LPC coefficients obtained for each basic analysis frame are not transmitted for each basic analysis frame, but, for example, n basic analysis frames are combined into one transmission. As a frame,
In particular, P representative frames are selected from each transmission frame, and within the range of the transmission frame consisting of these n basic analysis frames, the spectral envelope of the input speech is expressed by the LPC coefficient of each representative frame. In other words, by expressing n basic analysis frames as P frames (n>P), without transmitting all n basic analysis frames, the LPC coefficients of the aforementioned P representative frames and their repetitions can be transmitted. The amount of audio information to be transmitted has been compressed using a variable length frame representation that transmits the number of times Li (i=1 to Pj, n=tLi>).

This is because the macroscopic structure of speech is time-varying and its rate of change is also temporally non-uniform. This is based on the fact that it is more efficient to analyze data by processing it as a variable length frame.

[Invention or problem to be solved]

In the above-mentioned conventional speech analysis and synthesis apparatus using variable length frames, the number of representative frames P is fixedly determined in order to keep the amount of transmitted information constant (for example, 1200 bps).
Therefore, speech that changes rapidly and has large time distortion can follow the changes.If only P representative frames are used, a transmission frame with n basic analysis frames can be expressed accurately, but this will cause deterioration in the quality of synthesized speech. , which has the disadvantage of producing unstable and unclear synthesized speech.

SUMMARY OF THE INVENTION An object of the present invention is to provide a speech analysis and synthesis device that eliminates the above-mentioned drawbacks, significantly suppresses deterioration in the quality of synthesized speech, and obtains high-quality synthesized speech.

[Means to solve the problem]

The speech analysis and synthesis apparatus of the present invention includes a feature parameter extraction means for analyzing an input speech signal in units of transmission frames and every predetermined n analysis frames to extract LPC coefficients of a predetermined order and sound source information; representative frame candidate selection means for selecting a combination of m different numbers of analysis frames from n analysis frames as a representative analysis frame candidate representing the n analysis frames; Interpolation and extrapolation are performed on the LPC coefficients in the representative analysis frame candidates selected by the candidate selection means, and m
In addition to obtaining interpolated LPC coefficient sequences, the difference between the m interpolated LPC coefficient sequences and the LPC coefficient sequences of the n analysis frames extracted by the feature parameter extraction means is measured as a distortion amount, and the distortion A minimum distortion amount measuring means that obtains the m minimum distortion amounts, and a minimum distortion amount of 01 obtained by the minimum distortion amount measuring means, and is set by taking into account the synthesized voice and the amount of data transmission. Interpolation LP with optimal minimum distortion amount
representative frame determining means for determining a representative analysis frame candidate forming a C coefficient sequence as a representative analysis frame, and expressing the n analysis frames as variable length frames using the number and number of the representative analysis frames; and a voice synthesis means for synthesizing input voice signals based on the voice analysis information. be done.

C Example] Next, the present invention will be explained with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention.

In the embodiment shown in FIG.
It consists of a receiving side) 2 and a transmitting side 3.

The analysis side 1 includes an LPG analyzer 1011 that receives an input audio signal, limits the band to a predetermined value, and then performs LPC analysis on the quantized audio signal, and a representative frame candidate selector fl) 102 that selects a representative frame. .

Representative frame candidate selector (21103, representative frame candidate selector f31104, minimum distortion amount measuring device fl that measures the amount of distortion) 105, minimum distortion amount measuring device (21106, minimum distortion amount measuring device +31107, pitch extractor 108.f)
i-suitable distortion amount determiner 109. LPC parameter editor 110
The combining side 2 includes a multiplexer 111 and a demultiplexer 201 . LPC parameter decoder 202 pitch generator 3. Switcher 4. LPC parameter interpolator 205. Noise generator 206. LPC synthesis filter 207. It is configured with a variable gain amplifier 208.

Next, the operation of the embodiment shown in FIG. 1 will be explained.

The input audio signal is passed through the LPC analyzer 101 through a BPF (Ban
After removing unnecessary signals through a Pa5s Filter), each basic analysis frame is sampled by an A-D converter at a predetermined sampling frequency, e.g., 8 KHz, and quantized at a predetermined pitch/number, e.g., 12 hits. It is considered to be an audio signal. After this, the quantized audio signal is analyzed by a well-known analysis method to determine the parameters, etc.
PC coefficients are extracted with a predetermined order. The basic analysis frame is extracted by n LPC coefficients per transmission frame, and then the three representative frame candidate selectors (1) 102 to
The representative frame candidate selector (31104 sequentially selects LPC coefficients in combinations of 4, 5, and 6, respectively, and supplies them to the minimum distortion amount measuring device (1) 105 to the minimum distortion amount measuring device (31107) be done.

In this embodiment, the representative frame candidate selector fi+ 102
~Representative frame candidate selector (31104 are 4
．． Whether the 5 and 6 basic analysis frames and their LPC coefficients are selected, and how many types and numbers of different frames m to use, respectively, are determined in advance based on the operational purpose of the device, past audio materials, etc. Ru.

In this way, the representative frame candidate selector 11) 102 selects 4 out of 20 basic analysis frames per transmission frame.
representative frames containing LPC coefficients are selected. Further, a representative frame candidate selector +2) 103 selects five representative frames from one transmission frame, and a representative frame candidate selector f3) 104 selects six representative frames from one transmission frame.

The minimum distortion amount measuring device fil 105 expresses the amount of distortion due to the four representative frame candidates selected one after another by the representative frame candidate selector ill 105. Similarly, the minimum distortion amount measuring device +21106 is the representative frame candidate selector +2
The minimum distortion amount measuring device (31107 represents the amount of distortion due to the five representative frame candidates selected by the representative frame candidate selector+31104).

This amount of distortion is measured as follows.

For example, the minimum distortion amount measuring device fll 105 uses the LPC coefficients of the four representative frame candidates provided from the five representative frame candidate selector fll 102, and 204=1
LPC coefficients for six (6) basic analysis frames are determined by interpolation processing of interpolation and extrapolation to obtain an interpolated LPC coefficient sequence.

On the other hand, the minimum strain measuring device (1) 105 is the LPC analyzer 1
0 to 20 LPC coefficient sequences of the previous basic analysis frame per transmission frame are input, and the difference between this and the interpolated LPC coefficient sequence is calculated as the amount of distortion. Distortion amount measuring device (1) 105 measures the amount of distortion for the interpolated LPC coefficient sequences obtained for all four representative frame candidates, and supplies the minimum amount of distortion to the optimum amount of distortion determining unit 109.

Similarly, the minimum distortion amount measuring device +21106 and the minimum distortion amount measuring device 107 calculate the interpolated LPC coefficient sequence obtained based on five and six representative frame candidates, and the LPG coefficient sequence obtained from all basic analysis frames, respectively. The minimum amount of distortion is measured from the difference between the two, and is supplied to the optimum amount of distortion determiner 109.

The optimum distortion amount determiner 109 compares the minimum distortion amounts provided by the three minimum distortion amount measuring devices, and determines the optimum distortion amount based on the conditions of the minimum distortion amount and the amount of transmitted data. The number of frame candidates and the frame number are determined, this is set as a representative frame, and the representative frame number and representative frame number are supplied to the multiplexer 111.

The number of the representative frame determined in this way is supplied to the LPC parameter editor 110, and the LPC parameter editor 110 converts the LPC coefficient of the representative frame number into
It is read out from the PC analyzer 101 and supplied to the multiplexer 111 as an LPC coefficient.

Now, the power extractor 112 and the pitch extractor 108 each extract short-term average audio power and pitch information for each basic analysis frame from the quantized audio signal using a known method, and these audio power and pitch information are also used in the multiplexer. 111.

The multiplexer 111 quantizes the number of representative frames and frame numbers provided in this way, the LPC coefficients, and the data between the audio power and pitch information, and combines them as appropriate to synthesize multiplexed data via the transmission line 3. Transmit to side 2.

In this way, variable length frame synthesis is performed in which 20 basic analysis frames per transmission frame are expressed as 4 representative frames, 5 representative frames, or 6 representative frames and sent to the synthesizing side 2.

Now, on the combining side 2, the multiplexed information received via the transmission path 3 is demultiplexed by the demultiplexer 201, and the number of representative frames and LPC coefficients are determined by the LPC parameter decoder 202,
The representative frame number is determined by the LPC parameter interpolator 205.
, audio power is supplied to a variable gain amplifier 208, and pitch information is supplied to a pitch generator 203, respectively.

The LPC parameter decoder 202 determines the number of pins in decoding the LPC coefficients based on the number of representative frames, correctly decodes the LPC parameters, and supplies the decoded LPC parameters to the LPC parameter interpolator 205 .

The LPC parameter interpolator 205 obtains LPC parameters other than the representative frame number by linear interpolation based on the representative frame number and its decoded LPC parameters, and supplies this as a filter coefficient to the LPC synthesis filter 207 having a digital filter configuration.

The pitch generator 203 provided with the pitch information generates a repetitive pulse train corresponding to the pitch information, and supplies this to the variable gain amplifier 2.8 through the switch 205.

The variable gain amplifier 208 amplifies the output of the pitch generator 3 with a variable gain corresponding to the audio power, and outputs the LP as sound source information.
The LPC synthesis filter 207 is driven as an input to the C synthesis filter 207 to obtain digital audio, which is converted to analog using the built-in D- and A converters, and then subjected to a predetermined band restriction using the BPF, and then output as output audio. Send.

The noise generator 206 is supplied with the output of the pitch generator 203, and when the output is zero, it determines that there is no sound, and operates the switch 204 to output a random noise output instead of pitch information.

In this way, it is possible to perform audio analysis and synthesis that can significantly suppress the amount of transmission and also significantly suppress the deterioration of reproduced audio.

(Effects of the Invention) As explained above, the present invention significantly suppresses the amount of transmission and changes This has the effect of being able to reproduce stable, clear, and high-quality audio even for audio with high speed and large amount of time distortion.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention. ]... Analysis side, 2... Synthesis side, 3... Transmission line, 10
1... LPC analyzer, 102... Representative frame candidate selector (1), 103... Representative frame candidate selector (2
), 104...Representative frame candidate selector (3),
1.05... Minimum strain amount measuring device (11, 106... Minimum strain amount measuring device (2), 107... Minimum strain amount measuring device B (3
), 108...Pitch extractor, 109...Optimum distortion amount determiner, 110...LPC parameter editor, 1.1
1... Multiplexer, 201... Multiplex demultiplexer, 202...
・LPC parameter decoder, 203...Pitch generator, 204...Switcher, 205...LPC parameter interpolator, 2 6...Noise generator, 07...LPC synthesis filter, O8・- Variable gain amplifier.

Claims (1)

[Claims] An input audio signal is analyzed in units of transmission frames and every predetermined n analysis frames to obtain LPC (Lin) of a predetermined order.
earPrediction Coding (linear prediction analysis) coefficients and sound source information; representative frame candidate selection means for selecting a representative analysis frame candidate to be expressed; and LP in the representative analysis frame candidate selected by the representative frame candidate selection means;
Interpolation by interpolation and extrapolation is performed on the C coefficient, and n
m interpolated LPC coefficient sequences having m analysis frames are obtained, and L of the m interpolated LPC coefficient sequences and n analysis frames extracted by the feature parameter extraction means.
Minimum distortion amount measuring means for measuring the difference from a PC coefficient sequence as a distortion amount to obtain m minimum distortion amounts that minimize the distortion amount; and m minimum distortion amounts obtained by the minimum distortion amount measuring means. A representative analysis frame candidate forming an interpolated LPC coefficient sequence having an optimal minimum distortion amount set by taking into account the synthesized speech and data transmission amount is determined as a representative analysis frame, and the number and number of this representative analysis frame are used as the representative analysis frame. representative frame determining means for expressing n analysis frames as variable length frames; analysis information sending means for sending the number and number of the representative analysis frames together with the sound source information from the analysis side to the synthesis side as audio analysis information;
A speech analysis and synthesis device comprising: speech synthesis means for synthesizing input speech signals based on the speech analysis information.