JPH11249697A - Method and device for extracting pitch position - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reproduce a voice reducing the deterioration of quality due to a transmission error at the time of applying a voice coding/decoding method by limiting the polarity of periodic pulses. SOLUTION: A time series vector outputted from a periodic pulse generation means 101 is weighted in accordance with gain applied from a pulse polarity limit means 105 and supplied to a synthetic file 103. The means 105 limits the polarity of a periodic pulse in the time series vector weighted in order to generate a synthetic voice by outputting the value of gain inputted from an optimum gain calculation means 102 when the sign of the gain is positive or outputting zero when the sign is negative. After executing the processing mentioned above for all time series vectors generated by the means 101, a time series vector minimizing a distance between an input voice and a synthetic voice is searched and the leading pulse position of the pulse string is outputted as a pitch position S6.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pitch position extracting method and a pitch position extracting apparatus for extracting a characteristic point of a voice signal or an acoustic signal for each pitch period, and more particularly to a pitch applied to a voice coding / decoding method used for communication. The present invention relates to a position extracting method and a pitch position extracting device.

[0002]

2. Description of the Related Art A conventional speech encoding / decoding method using a pitch position extracting method is disclosed in Japanese Patent Laid-Open No. 6-202699. FIG. 5 shows an example of the overall configuration of a conventional CELP-based speech encoding / decoding method using the pitch position extraction method. In FIG.
Denotes a decoding unit, 3 denotes a multiplexing unit, and 4 denotes a separating unit. The encoding unit 1 includes a linear prediction parameter analysis unit 5, a linear prediction parameter encoding unit 6, a synthesis filter 7, an adaptive codebook 8, a pitch position extraction unit 9, a noise codebook 10, a pitch synchronization unit 11, a gain encoding unit. 12. Distance calculation means 1
3. Further, the decoding unit 2 includes a linear prediction parameter decoding unit 14, a synthesis filter 15, an adaptive codebook 16, a pitch position decoding unit 17, a noise codebook 18,
It comprises pitch synchronization means 19 and gain decoding means 20.

[0003] In CELP-based speech coding, about 5 to 50 ms is regarded as one frame, and speech of the frame is divided into spectrum information and sound source information and encoded. Hereinafter, the operation of the CELP-based speech encoding / decoding method having the configuration shown in FIG. 5 will be described. First, in the encoding unit 1, the linear prediction parameter analysis unit 5 analyzes the input speech S1 and extracts a linear prediction parameter which is spectrum information of the speech. The linear prediction parameter encoding unit 6 encodes the linear prediction parameter, and sets the encoded linear prediction parameter as a coefficient of the synthesis filter 7.

Next, the encoding of the sound source information will be described.
The adaptive codebook 8 stores past driving excitation vectors, and outputs a time-series vector obtained by periodically repeating the past driving excitation vectors corresponding to the adaptive codes. The pitch position extracting means 9 creates a pulse train having a cycle corresponding to the adaptive code, and generates a synthesized speech using the linear prediction parameter input from the linear prediction parameter analyzing means 5 as a sound source. Search for a pulse train that minimizes the distance to. Then, the pulse position is output to the pitch synchronization means 11 as a pitch position.

FIG. 6 shows an example of the internal configuration of the pitch position extracting means 9. In FIG.
2 is an optimum gain calculating means, 103 is a synthesis filter, 10
4 is a distance calculation means. The operation of the pitch position extracting method having such a configuration will be described. Periodic pulse generation means 1
01 sequentially generates a time-series vector in which the leading pulse position of the pulse train of the cycle corresponding to the adaptive code S4 is shifted by one sample from the leading position of the frame (FIG. 7). The time series vector output from the periodic pulse generating means 101 is weighted according to the gain given from the optimum gain calculating means 102 and is supplied to the synthesis filter 103.
The synthesis filter 103 obtains the weighted synthesized speech of the time-series vector using the coefficient corresponding to the linear prediction parameter S5. The distance calculation means 104 calculates the distance between the synthesized voice and the input voice S1, and the optimum gain calculation means calculates and outputs the gain that minimizes the distance. After performing the above processing on all the time-series vectors generated by the periodic pulse generation means 101, a time-series vector that minimizes the distance between the input voice and the synthesized voice is searched, and the top pulse position of the pulse train is determined. Output as pitch position S6.

[0006] The random codebook 10 stores a plurality of time-series vectors generated from, for example, random noise.
Output a code vector corresponding to the noise code. A pitch synchronization position is set for each code vector, and the pitch synchronization unit 11 cuts out the code vector and sets the pitch period length to match the pitch position extracted by the pitch position extraction unit 9 with the pitch synchronization position. , To generate a time-series vector that is periodically repeated. The time series vectors from the adaptive codebook 8 and the pitch synchronization unit 11 are weighted and added according to the respective gains provided from the gain encoding unit 12, and the result of the addition is supplied to the synthesis filter 7 as a driving excitation vector. To obtain coded speech. The distance calculation means 13 finds the distance between the coded speech and the input speech S1, and searches for an adaptive code, a noise code, and a gain that minimize the distance. After this coding is completed, a code of the linear prediction parameter, an adaptive code for minimizing distortion between the input voice and the coded voice, a noise code, a gain code, and a pitch position code are output as coding results.

On the other hand, in the decoding unit 2, the linear prediction parameter decoding means 14 decodes the linear prediction parameter from the code of the linear prediction parameter, and sets it as a coefficient of the synthesis filter 15. Next, adaptive codebook 16 outputs a time-series vector obtained by periodically repeating a past excitation vector corresponding to the adaptive code. Pitch position decoding means 1
7 outputs the pitch position decoded from the code of the pitch position to the pitch synchronization means 19. The random codebook 18 outputs a code vector corresponding to the random code. The pitch synchronization unit 19 calculates a time series vector from the code vector input from the noise codebook 18 and the pitch position input from the pitch position decoding unit 17 in the same manner as the pitch synchronization unit 11 of the encoding unit 1. Generate Each time-series vector from the adaptive codebook 16 and the pitch synchronization unit 19 is weighted and added according to each gain decoded from the gain code by the gain decoding unit 20, and the addition result is a driving excitation vector Is supplied to the synthesis filter 15 to obtain an output sound S3.

Japanese Patent Application Laid-Open No. 6-202699 also discloses a speech encoding / decoding method using a conventional pitch position extraction method which does not require transmission of a code of a pitch position. FIG. 8 in which the same reference numerals are assigned to corresponding parts to FIG.
The configuration of this conventional speech encoding / decoding method is shown in FIG.
Reference numerals 1 and 22 denote pitch position extracting means. The operation of the encoding / decoding method having such a configuration will be described. First, in the encoding unit 1, the pitch position extracting means 20 extracts, for example, a periodic point having the maximum amplitude of the time series vector as a pitch position from the periodically repeated time series vector output from the adaptive codebook 8. I do. This is equivalent to searching for a pulse train that minimizes the distance from the adaptive codebook to a time-series vector when a pulse train having a cycle corresponding to the adaptive code is created.

The noise codebook 10 stores a plurality of time-series vectors generated from, for example, random noise.
Output a code vector corresponding to the noise code. A pitch synchronization position is set for each code vector, and the pitch synchronization unit 11 cuts out the code vector and sets the pitch vector to the pitch period length so that the pitch synchronization position matches the pitch position extracted by the pitch position extraction unit 21. , To generate a time-series vector that is periodically repeated. Then, an encoded speech is generated using the adaptive codebook 8 and each time-series vector from the pitch synchronization unit 11, and an adaptive code, a noise code, and a gain that minimize the distance between the encoded speech and the input speech S1. Are selected and encoded.

Next, in the decoding unit 2, the pitch position extracting unit 22 uses the periodically repeated time series vector output from the adaptive codebook 16 in the same manner as the pitch position extracting unit 21 of the encoding unit 1. Extract the pitch position. The random codebook 18 outputs a code vector corresponding to the random code. The pitch synchronization unit 19 converts the time series vector from the code vector input from the noise codebook 18 and the pitch position input from the pitch position extraction unit 22 in the same manner as the pitch synchronization unit 11 of the encoding unit 1. Generate. Then, an output speech S3 is obtained using each time-series vector from the adaptive codebook 16 and the pitch synchronization unit 19. With this configuration, the pitch position when the code vector is periodicized is adaptively obtained from the time series vector from the adaptive codebook, so that transmission of the code at the pitch position is unnecessary,
The amount of transmission information can be reduced.

[0011]

In the above-described conventional pitch position extraction method, the distance between a speech signal and a synthesized speech generated by using a periodic pulse as a sound source or the distance between a sound source signal and a periodic pulse is minimum in frame units. And a pulse position is extracted as a pitch position. At this time, the polarity of the periodic pulse that minimizes the distance in the frame is determined by the shape of the voice signal or sound source signal in the frame section. When viewed, it was not always periodic. For example, as shown in FIG. 9, when the polarity of the periodic pulse in frames (A) and (C) is positive and that in frame (B) is negative, between frames (A) and (B) and between frame (B). ), (C)
Between, the pulses are not periodic.

For this reason, in the speech coding / decoding method using the conventional pitch position extraction method, the pitch position extracted and transmitted from the input voice has no correlation between frames, and the code of the pitch position has an error when transmitting the code. Occurs, the pitch position cannot be estimated even if the decoding side detects that there is an error, and even if all codes other than the code at the pitch position are correctly transmitted, an appropriate reproduced voice cannot be generated, There is a problem that the reproduced sound is greatly deteriorated.

In the above-described conventional pitch position extraction method, when pitch positions are extracted from a plurality of audio signals or sound source signals for which no difference in hearing is perceived, a small distance used for evaluating the pitch position extraction is required. Depending on the difference, which of the periodic pulse trains of the polarity is positive or negative is different, and the pitch position to be extracted may be significantly different. FIG. 10 shows an example of a shift in the pitch position extracted for sounds (A) and (B) having no difference in auditory sense.

For this reason, in the conventional speech encoding / decoding method using the pitch position extraction method which does not require transmission of the pitch position, when a transmission error occurs and the driving excitation vector is affected by the error, the driving The time series vector from the adaptive codebook generated from the excitation vector is also affected by the error. At this time, even if the influence of the error is small, the extracted pitch position may be significantly different from the original correct one. .

Here, once the pitch position is greatly erroneous due to the effect of the transmission error, even if the correct code is transmitted thereafter, the time series vector generated by the pitch synchronization processing is significantly different from the correct one. Vectors cannot be generated correctly. This also causes a vicious cycle in which the pitch position cannot be determined correctly, so that the influence of the transmission error over time is large, and there is a problem that the reproduction sound is greatly deteriorated when the transmission error occurs.

SUMMARY OF THE INVENTION The present invention has been made to solve such a problem, and provides a pitch position extraction method capable of reproducing a speech with little quality deterioration due to a transmission error when applied to a speech encoding / decoding method. Things.

[0017]

In order to solve the above-mentioned problems, a pitch position extracting method according to the present invention uses a distance between a voice signal and a synthesized voice generated by using a periodic pulse as a sound source, or a sound source signal. In the pitch position extracting method for extracting the characteristic positions of the audio signal or the sound source signal arranged at the pitch period interval by evaluating the distance between the periodic pulse and the periodic pulse, the polarity of the periodic pulse is limited.

Further, the pitch position extracting apparatus of the present invention provides a periodic pulse for sequentially generating a time-series vector in which the leading pulse position of a pulse train having a period corresponding to the pitch period of an input voice is shifted by one sample from the leading position of a frame. The time series vector output from the generating means and the periodic pulse generating means is weighted according to a gain given from the optimal gain calculating means, and the weighted using a coefficient corresponding to a linear prediction parameter obtained from the input speech. A synthesis filter for obtaining a synthesized voice of the time-series vector obtained, a distance between the synthesized voice and the input voice, searching for a time-series vector that minimizes the distance, and outputting a leading pulse position of the pulse train as a pitch position. A distance calculation means, and a periodic pulse in a time series vector weighted to generate the synthesized speech. And a pulse polarity limiting means for limiting the sex.

[0019]

Embodiments of the present invention will be described below with reference to the drawings.

Embodiment 1 FIG. 1 in which parts corresponding to those in FIG. 6 are assigned the same reference numerals shows the internal configuration of the pitch position extracting means 9 for performing the first embodiment of the pitch position extracting method according to the present invention, and 105 is a pulse polarity limiting means. .

Hereinafter, the operation will be described. The periodic pulse generating means 101 sequentially generates a time-series vector in which the leading pulse position of the pulse train of the cycle corresponding to the adaptive code S4 is shifted by one sample from the leading position of the frame. The time-series vector output from the periodic pulse generation unit 101 is weighted according to the gain given from the pulse polarity restriction unit 105 and is supplied to the synthesis filter 103. The synthesis filter 103 obtains the weighted synthesized speech of the time-series vector using the coefficient corresponding to the linear prediction parameter S5. The distance calculation means 104 calculates the synthesized voice and the input voice S1.
The optimum gain calculating means calculates a gain that minimizes the distance and outputs the gain. Pulse polarity limiting means 10
5 is weighted for generating a synthesized speech, assuming that, for example, when the sign of the gain input from the optimum gain calculating means 102 is positive, the gain value is output as the gain, and when the sign is negative, zero is output as the gain. Limit the polarity of the periodic pulse in the time series vector. After performing the above processing on all the time-series vectors generated by the periodic pulse generation means 101, a time-series vector that minimizes the distance between the input voice and the synthesized voice is searched, and the top pulse position of the pulse train is determined. Output as pitch position S6.

FIG. 2 in comparison with FIG. 9 is a schematic diagram for explaining the operation of the pitch position extracting method according to the first embodiment of the present invention. Is limited to a positive polarity. In contrast to FIG. 9, in FIG. 2, the polarity of the periodic pulse in frame (B) is positive, and the pulse is periodically generated between frames (A) and (B) and between frames (B) and (C). Has become.

When the pitch position extraction method according to the first embodiment is applied to the speech encoding / decoding method shown in FIG. 5, the pitch positions extracted and transmitted from the input speech have pitch periodicity even between frames. If an error occurs in the code at the pitch position during code transmission, if the decoding side detects that there is an error, the pitch position of the current frame can be estimated from the pitch position and pitch period of the previous frame (FIG. 3).

According to the first embodiment, by limiting the polarity of the pitch period pulse used for the evaluation in the pitch position extraction processing, the extracted pitch position has periodicity even between frames, so that speech coding and decoding is performed. If a transmission error occurs in the code at the pitch position and the correct pitch position cannot be decoded, the pitch position information of the previous frame and the pitch period of the frame where the transmission error occurred can be estimated from the pitch period information even if the correct pitch position cannot be decoded. Therefore, it is possible to avoid the deterioration of the reproduced sound due to the transmission error, and to reproduce the sound with less quality deterioration due to the transmission error.

Embodiment 2 Similar to the first embodiment, for example, FIG. 4 shows a schematic diagram for explaining the operation when the polarity of the pitch period pulse used for evaluation in the pitch position extraction process is restricted to be positive, in comparison with FIG. In FIG. 4, as compared with FIG. 10, the polarity of the periodic pulse with respect to the input voice (B) is positive, and the approximate voice signals (A) and (B)
, The deviation of the pitch position extracted becomes smaller.

When the pitch position extracting method according to the second embodiment is applied to a speech coding / decoding method which does not require transmission of the pitch position shown in FIG. 8, a transmission error occurs and the driving excitation vector is affected by the error. Even if it is received, if the influence of the error is small, the pitch position is found near the original correct one.

According to the second embodiment, by limiting the polarity of the pitch period pulse used for evaluation in the pitch position extraction processing, the pitch position can be determined for a plurality of audio signals or sound source signals for which no audible difference is perceived. When extraction is performed, the pitch positions to be extracted are obtained at almost the same position.Therefore, when applied to a speech coding / decoding method that does not require transmission of pitch positions, a transmission error occurs and the driving excitation vector is affected by the error. Even if it is received, if the influence of the error is small, the pitch position is found in the vicinity of the original correct one, and the time series vector generated by the pitch synchronization process is generated with a small difference from the correct one. Deterioration of generated sound can be avoided, and sound with low quality deterioration due to transmission errors can be reproduced.

Embodiment 3 In the first and second embodiments described above, the polarity of the periodic pulse is not fixed, for example, always limited to positive. The same polarity is always used in continuous voiced sound sections. The polarity may be different in each voiced sound section divided by. According to the third embodiment, since the polarity of the periodic pulse used for pitch position extraction can be either positive or negative, it is possible to select a polarity that is more suitable for the voice mode for each voiced sound section. Thus, a higher quality sound can be reproduced as compared with the case where the polarity is always limited to one polarity.

Embodiment 4 In the above-described first to third embodiments, the polarity of the periodic pulse is limited in the pitch position extraction method for evaluating the distance between the audio signal and the synthesized voice generated using the periodic pulse as a sound source. The same effect can be obtained even if the polarity of the periodic pulse is limited in the pitch position extracting method for evaluating the distance between the sound source signal and the periodic pulse.

[0030]

As described above in detail, according to the present invention, the distance between a speech signal and a synthesized speech generated using a periodic pulse as a sound source, or the distance between a sound source signal and a periodic pulse, is evaluated for each frame. In the pitch position extracting method and the pitch position extracting apparatus for extracting characteristic positions arranged at pitch interval intervals in the audio signal or the sound source signal, the polarity of the periodic pulse is limited. When applied to the method, it is possible to estimate the pitch position even if a transmission error occurs in the code at the pitch position, avoid deterioration of the reproduced sound due to the transmission error, and reproduce the sound with less quality deterioration due to the transmission error. Can be possible. Also, when applied to a speech coding / decoding method that does not require transmission of the pitch position, even if a transmission error occurs and the driving excitation vector is affected by the error, the pitch position is not affected by the error, and the pitch position is not changed. The time series vector that is found near the correct one and the time series vector generated by the pitch synchronization processing can be generated with a small difference from the correct one, avoiding the deterioration of the reproduced sound due to transmission errors, and reproducing the sound with less quality deterioration due to transmission errors Can be possible.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an internal configuration of a pitch position extracting device according to a first embodiment of the present invention.

FIG. 2 is a schematic diagram illustrating the operation of the pitch position extracting device according to the first embodiment of the present invention;

FIG. 3 is a schematic diagram for explaining an operation of a pitch position estimating process at the time of a transmission error in a speech encoding / decoding method to which the embodiment 1 of the pitch position extracting device according to the present invention is applied;

FIG. 4 is a schematic diagram for explaining an operation of a pitch position extracting method according to a second embodiment of the present invention;

FIG. 5 is a block diagram showing an overall configuration of a speech encoding / decoding device using a conventional pitch position extraction method.

FIG. 6 is a block diagram showing an internal configuration of a conventional pitch position extracting device.

FIG. 7 is a schematic diagram for explaining the operation of a conventional periodic pulse generating means.

FIG. 8 is a block diagram illustrating an overall configuration of a speech encoding / decoding method using a conventional pitch position extracting device that does not require transmission of a code of a pitch position.

FIG. 9 is a schematic diagram for explaining the operation of a conventional pitch position extracting method.

FIG. 10 is a schematic diagram for explaining the operation of a conventional pitch position extracting method.

[Explanation of symbols]

1: encoding unit, 2: decoding unit, 3: multiplexing means,
4: separation means, 5: linear prediction parameter analysis means,
6: linear prediction parameter coding means, 7, 15: synthesis filter, 8, 16: adaptive codebook, 9: pitch position extraction means, 10, 18: noise codebook, 1
1, 19: pitch synchronization means, 12: gain encoding means, 13: distance calculation means, 14: linear prediction parameter decoding means, 17: pitch position decoding means, 20: gain decoding means, 21, 22: Pitch position extracting means, 101: periodic pulse generating means, 10
2: optimal gain calculation means, 103: synthesis filter, 104: distance calculation means, 105: pulse polarity restriction means.

Claims (2)

[Claims] An evaluation of a distance between a speech signal and a synthesized speech generated by using a periodic pulse as a sound source or a distance between a sound source signal and a periodic pulse in a frame unit to determine a pitch cycle in the speech signal or the sound source signal. A pitch position extracting method for extracting characteristic positions arranged at intervals, wherein the polarity of the periodic pulse is limited. 2. A periodic pulse generating means for sequentially generating a time series vector in which a leading pulse position of a pulse train having a cycle corresponding to a pitch cycle of an input voice is shifted by one sample from a leading position of a frame, and a periodic pulse generating means. The time-series vector output from is weighted according to the gain given from the optimal gain calculating means, and the synthesized voice of the weighted time-series vector is calculated using a coefficient corresponding to the linear prediction parameter obtained from the input voice. Obtaining a synthesis filter and a distance between the synthesized voice and the input voice,
A distance calculating means for searching for a time series vector which minimizes the distance and outputting the leading pulse position of the pulse train as a pitch position, and limiting the polarity of the periodic pulse in the time series vector weighted to generate the synthesized speech A pitch position extracting device comprising: a pulse polarity limiter.