JP3617503B2 - Speech decoding method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a decoding method for decoding a voice signal compressed and encoded into a digital signal, and more particularly to a voice decoding method for reproducing a voice with little quality deterioration due to a transmission path error when applied to communication.
[0002]
[Prior art]
Conventionally, as a high-efficiency speech coding method for communication, techniques such as code-excited linear prediction coding (CELP) and multi-band excitation coding (MBE) are representative. It is. About each technique, "Code-excited linear prediction (CELP): High-quality speech at 8 kbps" (M.R. And “A real-time implementation of the improved MBE speech coder” (MS Brandstein, PA Monta, J. C. Hardwick and J. S. Lim, ICAS SP'90, pp. 8). 1990).
[0003]
Here, CELP speech decoding will be described. In CELP speech decoding, about 5 to 50 ms is defined as one frame, and the speech signal is decoded by inputting the encoded speech by dividing the speech of the frame into spectrum information and sound source information. Hereinafter, CELP speech decoding will be described with reference to FIG.
S2 is a code input terminal for the encoded speech signal, and inputs an adaptive code, a noise code, a linear prediction parameter code, and a gain code. Reference numeral 2 denotes a decoding unit, 4 denotes separation means, and S3 denotes an output terminal for a reproduced audio signal. The decoding unit 2 includes a linear prediction parameter decoding unit 15, a synthesis filter 16, an adaptive codebook 17, a noise codebook 18, and a gain decoding unit 19.
The adaptive codebook 17 stores past drive excitation vectors, and outputs a time series vector in which past drive excitation vectors are periodically repeated corresponding to the adaptive code.
The noise codebook 18 stores a plurality of time series vectors generated from random noise, for example, and outputs a time series vector corresponding to the noise code.
[0004]
In the decoding unit 2, the speech code input to the input terminal 2 is separated by the separating unit 4 into an adaptive code, a noise code, a linear prediction parameter code, and a gain code. The linear prediction parameter decoding means 15 decodes the linear prediction parameter from the code of the linear prediction parameter and sets it as a coefficient of the synthesis filter 16. Next, adaptive codebook 17 outputs a time series vector in which past drive excitation vectors are periodically repeated corresponding to the adaptive code, and noise codebook 18 outputs a time series vector corresponding to the noise code. To do. These time series vectors are weighted and added according to each gain decoded from the gain code by the gain decoding means 19, and the addition result is supplied to the synthesis filter 16 as a drive excitation vector, and the output sound S3 is output. can get.
[0005]
Here, in a speech decoding method applied to an application field where a code error occurs in practice, such as mobile communication, the quality of encoded speech due to the code error is suppressed by using an error correction coding technique. Further, when the error cannot be corrected, the waveform is repaired to reduce the influence of the code error.
[0006]
There are the following two types of repair methods. That is, the first restoration method is “Channel coding for digital transmission transmission in the Japan” (MJ McLaughlin, IEICE Radiocommunication System Research Group RCS90-4 ), When the current frame is a frame with a code error, the reproduced sound is generated by repeatedly using the parameters of the past frame, and the power of the reproduced sound is gradually suppressed.
[0007]
Further, as disclosed in Japanese Patent Laid-Open No. 6-12095, the second restoration method uses the code error detection information of each of the past frame, the current frame, and the future frame, and sets each frame error detection state. In response, the audio of the current frame is reproduced and repaired. In this case, since interpolation is performed also using information on future frames, it is possible to perform interpolation with less distortion than when only information on past frames is used.
[0008]
Japanese Patent Laid-Open No. 7-36496 discloses a transmission error compensation method for maintaining audio quality in the sense of hearing even when waveform restoration cannot be performed sufficiently. This does not output reproduced sound due to a transmission error state, but outputs a noise signal instead. Thereby, even when appropriate waveform restoration cannot be performed, it is possible to avoid noise.
[0009]
[Problems to be solved by the invention]
In the conventional waveform restoration method when a code error occurs, the parameters of the past frame are repeatedly used, but there is a problem that the quality of the reproduced speech is still low. For example, since the pitch periodicity is non-stationary in a voiced rising frame, the pitch information parameter suitable for the subsequent voiced stationary part is not always transmitted, and when a code error occurs in the subsequent frame, Even if the parameters of the pitch information were used repeatedly, good reproduction sound could not be obtained.
[0010]
In addition, there is a locally periodic frame even in the unvoiced part. If a code error occurs in the following unvoiced frame, the signal is played back as a buzzer sound because the signal is repeated for a certain period due to repeated parameters. The voice quality was greatly degraded. In addition, in a frame in which a code error has occurred, if the spectrum generated by repeating the parameters of the previous frame is inappropriate, the reproduced voice quality is greatly degraded. In particular, when the power in the high frequency range is large, such as a sharp formant peak in the high frequency range, the sense of noise in the high frequency range becomes prominent and the auditory degradation is significant.
[0011]
Further, the conventional waveform restoration method in which interpolation is performed also using information of a future frame has a problem that a delay necessary for reproduction of the current frame is increased for restoration. If the delay required for audio reproduction becomes large, when applied to communication, natural conversation cannot be performed and the call is disturbed. Therefore, it is desirable that the delay be as small as possible.
In addition, even when a conventional code error occurs and waveform restoration cannot be performed sufficiently, the transmission error compensation method that maintains audible voice quality switches between a reproduced voice and a noise signal for output. However, since there is a discontinuity in the transition from voice to noise or from noise to voice, there is a problem that this leads to deterioration of the reproduced voice quality in terms of audibility. Furthermore, since a noise signal is always output when a transmission error occurs, noise is output even in a portion that is originally silent, which also leads to deterioration in hearing.
[0012]
[Means for Solving the Problems]
When the speech decoding method of the present invention detects that the code has not been correctly decoded in the speech decoding method for decoding the spectrum information and sound source information of the encoded speech in units of frames and reproducing the speech. In addition, the state of the reproduced sound of the previous frame is determined, and the sound of the current frame is reproduced and repaired according to the determination.
[0013]
In the speech decoding method of the next invention, the sound source information includes pitch information, and when the reproduced speech of the previous frame is voiced, the pitch information for the current frame is obtained from the information of the past frame, The audio of the current frame is played back and repaired based on the pitch information.
[0014]
Furthermore, in the audio decoding method of the next invention, when the reproduced audio of the previous frame is unvoiced, the audio of the current frame is reproduced and repaired so as to suppress the pitch periodicity.
[0015]
Furthermore, the speech decoding method of the next invention detects that the code was not correctly decoded in the speech decoding method in which the spectrum information and sound source information of the encoded speech are decoded in units of frames and the speech is reproduced. In this case, the low-band formant structure of the spectrum of the previous frame is maintained, and the sound of the current frame is reproduced and repaired so that the high-band formant peak is suppressed.
[0016]
[0017]
The speech decoding method of the next invention is a speech decoding method that decodes encoded speech information and reproduces speech, and detects that the code has not been decoded correctly and reproduces it according to the detected state. Added noise to voice.
[0018]
Furthermore, the speech decoding method of the next invention detects that the code has not been correctly decoded, and controls the power of noise superimposed on the reproduced speech in accordance with the detected state.
[0019]
Furthermore, the speech decoding method according to the next invention further estimates the state of the speech to be reproduced, and superimposes noise on the reproduced speech according to the estimation result.
[0020]
In the speech decoding method of the next invention, the state of the speech to be reproduced is estimated, and the power of noise superimposed on the reproduced speech is controlled according to the estimation result.
[0021]
DETAILED DESCRIPTION OF THE INVENTION
Embodiments of the present invention will be described below with reference to the drawings.
[0022]
Embodiment 1 FIG.
FIG. 1 in which the same reference numerals are assigned to the corresponding parts to FIG. 5 shows the configuration of the first embodiment of the speech decoding method according to the present invention, and 30 in the figure is a linear prediction parameter when a code transmission error is detected. Linear prediction parameter repair means 31 for repairing, rising judgment means 31 for judging whether the previous frame is a voiced rise by analyzing the reproduced voice, 32 for pitch extraction means for analyzing the reproduced voice and obtaining the pitch, 33 for code This is adaptive code repair means for repairing the adaptive code when a transmission error is detected. Further, 34 is a silent determination means for analyzing whether or not the previous frame is silent by analyzing the reproduced speech, 35 is a gain repair means for repairing a gain when a code transmission error is detected, and 36 is a code transmission error detection. Reproduction audio power control means for controlling the reproduction audio power according to the situation. Further, 37 is a voice state estimating means for estimating the voice state of the current frame when a code transmission error is detected, 38 is a noise generating means for generating a noise signal, and 39 is a noise according to the code transmission error detection situation. This is noise power control means for controlling the power of the signal.
[0023]
In the decoding unit 2, the linear prediction parameter decoding unit 15 decodes the linear prediction parameter from the code of the linear prediction parameter. The linear prediction parameter restoration means 30 calculates a low frequency band from a linear prediction parameter decoded in a frame in which no transmission error of the code is detected, and from a linear prediction parameter decoded in a previous frame in a frame in which a transmission error is detected. A linear prediction parameter that maintains a formant structure of the above and suppresses a high-frequency formant peak, for example, when the linear prediction parameter is expressed as a line spectrum pair parameter, is determined as using only the low-frequency parameter, Set as a coefficient of the synthesis filter 16. FIG. 2 shows an example of a spectrum envelope represented by the linear prediction parameter of the previous frame and a spectrum envelope in which the low frequency structure is maintained and the high frequency band is suppressed by the linear prediction parameter repair processing.
[0024]
The rise determination unit 31 analyzes the reproduced voice of the previous frame in the frame in which the transmission error of the code is detected, determines whether or not the previous frame is a voiced rise, and the determination result is used as the pitch analysis unit 32. The data is output to the code repair means 33. Here, the determination of whether or not a voiced rise is made is, for example, a voiced rise when the power of the voice in the latter half of the frame is greater than a threshold by comparison with that in the first half of the frame. The pitch extraction means 32 analyzes the past reproduced voice and extracts at least one pitch period candidate when the previous frame is determined to be voiced rise in the frame in which the transmission error of the code is detected, The data is output to the adaptive code repair means 33.
[0025]
The adaptive code restoration means 33 uses the adaptive code input from the separation means 4 in a frame in which no code transmission error is detected, and if the previous frame is not a voiced rise even if a code error is detected, If a code error is detected and the previous frame has a voiced rise, for example, the appearance of an adaptive code correctly transmitted in the past from among the pitch period candidates input from the pitch extraction means 32 Compared with the frequency distribution, the one closest to the pitch period corresponding to the code having a high appearance frequency is selected, and the adaptive code corresponding to the selected pitch period is output to the adaptive codebook 17. The adaptive codebook 17 outputs time series vectors in which past drive excitation vectors are periodically repeated corresponding to the adaptive codes. FIG. 3A shows an example of the adaptive code repair process in the eleventh embodiment, and FIG. 3B shows an example of the repair process that repeatedly uses the conventional adaptive code of the previous frame. As shown in the figure, even if the pitch period corresponding to the adaptive code of the previous frame is unsuitable for the current frame, good waveform restoration processing is possible by obtaining the pitch period suitable for the current frame from the past frame information It turns out that it is.
[0026]
The noise codebook 18 outputs a time series vector corresponding to the noise code. The gain decoding means 19 decodes the gain for the time series vector output from the adaptive codebook 17 and the noise codebook 18 from the gain code. The silent determination unit 34 analyzes the reproduced sound of the previous frame in the frame in which the code transmission error is detected, determines whether the previous frame is unvoiced, and outputs the determination result to the gain restoration unit 35. . The gain restoration means 35 obtains the gain input from the gain decoding means 19 for a frame in which no code transmission error is detected, and the gain of the previous frame if a code error is detected and the previous frame is not silent. When a code error is detected and the previous frame is unvoiced, the gain for the time series vector from the adaptive codebook is α times the gain of the previous frame, and the gain for the time series vector from the noise codebook is β Double the output. Here, α and β are, for example, 0 ≦ α <β ≦ 1.
[0027]
The time series vectors from the adaptive codebook 17 and the noise codebook 18 are weighted and added according to the respective gains output from the gain restoration means 35, and the addition result is supplied to the synthesis filter 16 as a drive excitation vector. Playback audio can be obtained. The playback sound power control means 36 suppresses the power of the playback sound, for example, gradually increasing the suppression amount as errors continue in accordance with the code transmission error detection status. When a code transmission error is detected, the speech state estimation unit 37 determines the presence / absence of the current frame from the past reproduced speech and the code of the linear prediction parameter of the current frame, and the determination result is used as the noise generation unit. 38.
[0028]
The noise generation unit 38 detects noise in a frame in which, for example, an error is detected and it is estimated to be sound according to the transmission error detection status of the code and the sound / silence determination result input from the sound state determination unit 37. Generate. The noise power control means 39 controls the noise power, for example, gradually increasing the power at the beginning of noise superposition and gradually decreasing the power at the end of superposition. FIG. 4 shows an example of the power control processing of reproduced voice and noise signal. The reproduced sound output from the reproduced sound power control means 36 and the noise output from the noise power control means 39 are added to obtain an output sound S3.
[0029]
According to the first embodiment, when it is detected that the code has not been correctly decoded, the state of the reproduced sound of the previous frame is determined, and the sound of the current frame is reproduced and restored according to this determination. As a result, it is possible to reproduce audio with little quality degradation due to transmission path errors without increasing the delay. In addition, when a code error occurs, the audio quality on hearing can be maintained by superimposing noise on the reproduced sound according to the state of the reproduced sound.
[0030]
Embodiment 2. FIG.
In the first embodiment described above, in the linear prediction parameter restoration means 30, when a transmission error occurs, a linear prediction parameter is obtained so that a spectrum in which a high-frequency formant peak is suppressed is always obtained from the linear prediction parameter of the previous frame. Although the coefficients of the synthesis filter 16 are used, instead of this, the suppression process is performed only when there is a sharp formant peak in the high band of the spectrum represented by the linear prediction parameter of the previous frame. In other cases, the linear prediction parameter of the previous frame is used. Alternatively, the processing may be selectively performed depending on the state.
[0031]
According to the second embodiment, when the linear prediction parameter repair process is performed, the suppression process is performed only when there is a sharp formant peak in the high band and there is a high possibility that it leads to quality degradation in the sense of hearing. In this case, since the linear prediction parameter of the previous frame is used repeatedly, the continuity of the synthesized speech is improved without generating abnormal noise, and the audio quality in terms of hearing can be improved.
[0032]
Embodiment 3 FIG.
In Embodiment 1 described above, when the previous frame at the time of transmission error detection is silent, the gain restoration means 35 adjusts the gain to suppress the pitch periodicity of the driving sound source vector. For example, the pitch periodicity may be suppressed using another means such as not using a time-series vector from the adaptive codebook. According to the third embodiment, it is only necessary to switch whether or not to use the time-series vector from the adaptive codebook for the suppression of pitch periodicity, as compared with the case where the pitch periodicity is suppressed by adjusting the gain. It can be realized easily.
[0033]
Embodiment 4 FIG.
In the first embodiment described above, the speech state estimation unit 37 determines the presence / absence of the current frame from the past reproduced speech and the code of the linear prediction parameter of the current frame. The sign of gain may also be used as a parameter for determination. According to the fourth embodiment, since the sound / silence determination is performed using not only the past reproduced speech and the code of the linear prediction parameter but also other information, determination with higher accuracy is possible.
[0034]
Embodiment 5 FIG.
In the above-described first embodiment, repair processing at the time of transmission error and noise superimposition processing are applied to CELP speech coding, but instead, this is applied to other speech coding schemes including MBE speech coding. Even in the same way, it is possible to improve the sound quality of the reproduced sound at the time of transmission error.
[0035]
【The invention's effect】
As detailed above,
[0036]
According to the present invention, when it is detected by the decoding method that the code has not been correctly decoded, the state of the reproduced sound of the previous frame is determined, and the sound of the current frame is reproduced and restored according to this determination. As a result, it is possible to perform an effective reproduction and repair process without increasing the delay, and it is possible to reproduce audio with less quality degradation due to transmission path errors.
[0037]
According to the present invention, when the reproduced sound of the previous frame is voiced, the pitch information for the current frame is obtained from the information of the past frame, and the sound of the current frame is reproduced and repaired based on the pitch information. Therefore, even if the pitch information transmitted in the previous frame is unsuitable for the current frame, it is possible to generate a playback sound by obtaining a pitch period suitable for the current frame, and quality degradation due to transmission errors. Less audio can be played.
[0038]
Further, according to the present invention, when the reproduced sound of the previous frame is unvoiced, the sound of the current frame is reproduced and repaired so as to suppress the pitch periodicity. Can be avoided, and voice with little quality degradation due to transmission errors can be reproduced.
[0039]
Further, according to the present invention, when it is detected that the code has not been correctly decoded, the current frame is maintained so that the low-band formant structure of the previous frame spectrum is maintained and the high-band formant peak is suppressed. By reproducing and restoring the sound, it is possible to reduce high-frequency noise while maintaining the continuity of the reproduced sound, and it is possible to reproduce the sound with less quality deterioration due to transmission path errors.
[0040]
Further, according to the present invention, the speech decoding method detects that the code has not been correctly decoded, and superimposes noise on the reproduced speech in accordance with the detection state, thereby smoothing the transition between the reproduced speech and noise. In addition, noise is output only to a portion that is originally sounded, and noise is not output in a silent portion, so that it is possible to maintain audible voice quality.
[Brief description of the drawings]
1 is a block diagram showing a configuration of a first embodiment of a speech decoding method according to the present invention;
FIG. 2 is a schematic diagram for explaining the operation of spectrum restoration processing in the first embodiment;
FIG. 3 is a signal waveform diagram for explaining an operation of adaptive code restoration processing in the first embodiment;
FIG. 4 is a schematic diagram for explaining an example of power control processing of reproduced speech and noise signals in the first embodiment.
FIG. 5 is a block diagram showing an overall configuration of a conventional CELP speech coding / decoding method.
[Explanation of symbols]
2 Decoding unit 4 Separation unit 16 Synthesis filter 17 Adaptive codebook 18 Noise codebook 15 Linear prediction parameter decoding unit 19 Gain decoding unit 30 Linear prediction parameter restoration unit 31 Rising judgment unit 32 Pitch extraction unit 33 Adaptive code restoration unit 34 Silence determination means 35 Gain restoration means 36 Reproduction voice power control means 37 Voice state estimation means 38 Noise generation means 39 Noise power control means

Claims (4)

In the audio decoding method for decoding the encoded audio spectrum information and sound source information in units of frames and reproducing the audio, when it is detected that the code has not been correctly decoded, the reproduced audio of the previous frame is reproduced. When the state is determined and the playback sound of the previous frame is voiced , the current frame is selected from the information of the previous frame by selecting the pitch information corresponding to the pitch cycle with a high frequency of correct transmission in the past. A speech decoding method characterized by obtaining the pitch information of the current frame and reproducing and restoring the speech of the current frame based on the pitch information. In the speech decoding method for decoding the encoded speech spectrum information and sound source information in units of frames and reproducing the speech, if it is detected that the code cannot be decoded correctly, the spectrum of the previous frame is reduced. A speech decoding method comprising: reproducing and restoring speech of a current frame so as to obtain a spectrum in which a formant structure in a region is maintained and a formant peak in a high region is suppressed. In the audio decoding method for decoding encoded audio information and reproducing the audio, when it is detected that the code has not been correctly decoded, the state of the audio to be reproduced is estimated, and according to the estimation result A speech decoding method characterized by superimposing noise on a reproduced speech and controlling the noise power so that the power is gradually reduced at a silent portion at the end of the superposition. When it is detected that the code could not be decoded correctly, the state of the audio to be reproduced is estimated, noise is superimposed on the reproduced audio according to the estimation result, and the power is gradually reduced at the silent part at the end of the superimposition. 4. The speech decoding method according to claim 3, wherein no noise is output in a silent part by controlling the power of the noise so as to cause the noise to be generated.

Images