JPH10124089A - Processor and method for speech signal processing and device and method for expanding voice bandwidth - Google Patents

Abstract

PROBLEM TO BE SOLVED: To clearly reproduce a fricative and an affricate when voice bandwidth is expanded by LPC synthesis. SOLUTION: An input voice signal is analyzed by an LPC analytic filter 2 to retrieve the most matching autocorrelation as a parameter from a narrow- band code book 6, the parameter corresponding to it is outputted from a broadband code book 12, and an LPC synthesizing filter 11 performs synthesis. Consequently, the bandwidth of a voice is expanded. An affricate detecting circuit 7 detects a fricative and an affricate by using the autocorrelation value of the input voice signal and the value of frame power. Once the fricative or affricate is detected, the whole or part of the band of an exciting source is boosted. Consequently, a power deficiency in case of the fricative or affricate is improved to clearly reproduce the fricative and affricate.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio signal processing apparatus and method for generating a wideband audio signal from an audio signal whose frequency band is restricted to a narrow band by passing through a transmission line such as a telephone line. And a bandwidth extension device and method.

[0002]

2. Description of the Related Art The bandwidth of a telephone line is, for example, 300 to 340.
It is as narrow as 0 kHz, and the frequency band of the audio signal transmitted via the telephone line is limited. For this reason, the sound quality of the conventional analog telephone line is not very good. They also complain about the sound quality of digital mobile phones.

Therefore, various systems have been proposed in which the voice bandwidth is expanded on the receiving side to improve the sound quality. Among them, a narrowband codebook in which parameters of a narrowband audio signal obtained in advance from a plurality of audio signal patterns are stored as code vectors, and a wideband audio signal obtained from the same audio signal pattern. By preparing a wideband codebook in which parameters are stored in advance as code vectors, analyzing the input signal with the narrowband codebook, and performing speech synthesis using the wideband codebook based on this analysis result, Systems have been proposed that extend the bandwidth and improve sound quality.

That is, as shown in FIG. 6, when transmitting a voice signal through a transmission line such as a telephone line, the transmitting side 1
The frequency band of the audio signal from 01 is restricted by passing through the transmission line 102. For example, the frequency band of the audio signal from the transmitting side 101 is 300 Hz to 7000 H
Even if z is about z, the frequency band of the audio signal sent to the receiving side 103 is limited to, for example, about 300 Hz to 3400 Hz by passing through the transmission path 102.

Therefore, as shown in FIG. 7, a narrow-band codebook 105 in which parameters of a narrow-band audio signal previously obtained from a plurality of audio signal patterns are stored as code vectors corresponds to the narrow-band codebook 105. A wideband codebook 106 in which parameters of a wideband audio signal obtained from the same audio signal pattern are stored in advance as code vectors is prepared.

The code books 105 and 106 are:
For example, it is created by dividing the same broadband audio signal into frames of a predetermined length to form a plurality of audio signal patterns, and analyzing the spectrum envelope for each frame. That is, when creating a codebook, a wideband audio signal is used, and the wideband audio signal is divided into predetermined frames. The wideband codebook 106 stores, as a code vector, spectrum envelope information when the wideband audio signal is analyzed as it is in a wideband. The narrow-band codebook 105 stores, as a code vector, spectrum envelope information when a wideband audio signal is analyzed with its band limited to, for example, 300 to 3400 Hz.

[0007] As the spectrum envelope information stored in the narrowband codebook 105 and the wideband codebook 106, an LPC cepstrum is conventionally used. LPC
Ceptoram is a ceptoram with a linear prediction coefficient,
It is obtained as shown by the following equation.

(Equation 1)

In FIG. 7, a narrow-band audio signal transmitted from a transmitting side 101 to a receiving side 103 via a transmission line 102 is first transmitted to an analysis circuit 104. Analysis circuit 1
At 04, the input audio signal is divided into predetermined frames,
A spectrum envelope is required. The output of analysis circuit 104 is sent to narrowband codebook 105. In the narrowband codebook 105, the spectrum envelope analyzed by the analysis circuit 104 is compared with the spectrum envelope information stored in the narrowband codebook 105, and a matching process is performed. Then, the output of the narrowband codebook 105 is sent to the wideband codebook 106, and the wideband spectrum envelope information corresponding to the spectrum matching information most closely matched in the narrowband codebook 105 is
Read from wideband codebook 106.

[0009] This broadband spectrum envelope information is sent to the synthesis circuit 107. The synthesis circuit 107 synthesizes the audio signal using the wideband spectrum envelope information read from the wideband codebook 106. Since the synthesized audio signal is synthesized using the wideband codebook 106, it becomes a wideband audio signal.

[0010]

As described above, in the conventional voice bandwidth extension system, an LPC cepstrum is used as a codebook vector. Also, noise and a pulse train are used as excitation sources when synthesizing audio signals. However, in the LPC cepstrum, although the distortion on hearing and the quantization error relatively match, a logarithmic scale is used rather than using a linear scale. Therefore, a portion having a small energy is emphasized, and an error in a portion having a large energy is reduced. growing. For use in such an audio bandwidth extension system, it is preferable that distortion in a vowel portion be suppressed as much as possible in terms of hearing. Therefore, LPC ceptoram is not always optimal. The excitation source should be as close as possible to the LPC residual in a wide band, but the conventional method using noise and a pulse train is far from this.

Therefore, it is conceivable to synthesize a wideband speech signal by LPC synthesis using autocorrelation as a codebook vector and upsampled LPC residuals as an excitation source. Since the autocorrelation is not on a logarithmic scale, it is considered that distortion in vowel parts is improved.
However, using autocorrelation as a codebook vector,
If an upsampled LPC residual is used as an excitation source to form a wideband audio signal by LPC synthesis, there is a problem in that fricative and affricate sounds are insufficient, resulting in a crisp sound. This can be attributed to insufficient prediction of the spectrum envelope, but it is considered to be due to insufficient power of the excitation source as a seed.

That is, in the case of a fricative or affricate, prediction by LPC synthesis is performed relatively well, and the power of the residual is reduced. However, prediction is not sufficient for wideband speech, and the power of the residual does not decrease. For this reason, when the band of the fricative and affricate is extended, the residual power must be as large. However,
Since the residual is created by predicting from the narrowband residual, the power is not sufficiently large. For this reason, in the case of a fricative or affricate, the power of the excitation source is insufficient.

Accordingly, an object of the present invention is to provide an audio signal processing apparatus and method, and an audio bandwidth extending apparatus and method capable of clearly reproducing a fricative sound and an affricate sound when the audio bandwidth is extended. To provide.

[0014]

SUMMARY OF THE INVENTION The present invention relates to an audio signal processing apparatus which analyzes an input audio signal, performs signal processing on the analyzed audio signal, and synthesizes the audio signal. It is characterized by comprising an affricate detecting means for detecting a fricative or affricate of an audio signal, and a boost means for giving a boost to an excitation source when a fricative or affricate is detected. It is an audio signal processing device.

Further, the present invention provides an analyzing means for obtaining an autocorrelation parameter from an input narrow-band audio signal, an excitation source forming means for obtaining an excitation source from an LPC residual of the input narrow-band audio signal, The narrowband codebook in which the parameters of the autocorrelation of the narrowband audio signal obtained from the pattern are stored, and the parameters of the autocorrelation of the wideband audio signal obtained in advance from a plurality of voice signal patterns are stored in the narrowband codebook. A correspondingly stored broadband codebook, fricatives, fricatives detecting means for detecting fricatives, fricatives, boosting means for providing a boost to the excitation source when fricatives are detected, input The autocorrelation parameters of the narrowband speech signal are compared with the autocorrelation parameters of the input narrowband speech signal stored in the narrowband codebook, and the optimum Reading and matching means for searching the meter, based on the search result of the matching means, the corresponding parameter from the parameters of the autocorrelation of the wide band audio signal stored in the wide band code book,
An audio bandwidth extending apparatus comprising: a synthesizing unit that synthesizes an output broadband audio signal based on the read parameters and the excitation source.

In the present invention, the fricative sound detecting means detects fricative sounds and fricative sounds using the value of the autocorrelation of the input audio signal and the value of the frame power.

As described above, when the fricative sound or fricative noise is detected and the fricative sound or fricative noise is detected, a boost is applied to the excitation source. In this case, the power shortage is improved, and the fricative and affricate can be reproduced clearly.

[0018]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows an example of an audio bandwidth extension system to which the present invention is applied. In FIG. 1, the input terminal 1 has a frequency band of, for example, 300 Hz to 3400 Hz and a sampling frequency of 8 Hz.
A kHz narrowband audio signal is provided. This narrowband audio signal is supplied to an LPC (Linear Predictive Coding) analysis filter 2 and also to an up-sampling circuit 3.

The up-sampling circuit 3 is for up-sampling the sampling frequency from 8 kHz to 16 kHz. The output of the up-sampling circuit 3 is 30
The signal is supplied to the addition circuit 5 via the band-pass filter 4 having a pass band of 0 Hz to 3400 Hz. As will be described later, a path leading to the up-sampling circuit 3, the band-pass filter 4, and the adding circuit 5 is a path for adding a signal of an original frequency band component to a high-band sound signal synthesized by voice. It is.

The LPC analysis filter 2 converts the narrow-band audio signal from the input terminal 1 into a frame and performs a tenth-order LPC analysis. In the course of the LPC analysis, a tenth-order autocorrelation is obtained. This autocorrelation is sent to the narrow band codebook 6 and to the affricate detection circuit 7. The LPC residual obtained by the LPC analysis filter 2 is:
It is sent to the up-sampling circuit 8.

The up-sampling circuit 8 up-samples the LPC residual of the narrow-band sound. The output of the up-sampling circuit 8 is a low-pass filter 9 and a boost circuit 1
0 to the LPC synthesis filter 11. The signal obtained by up-sampling the LPC residual and suppressing the high frequency band is used as an excitation source when synthesizing the audio signal, as described later. The boost circuit 10 is for boosting the excitation source when an affricate or a fricative is detected. The boost amount of the boost circuit 10 is controlled by an output of the affricate detecting circuit 7.

The narrow-band codebook 6 stores 10th-order autocorrelation information of a narrow-band audio signal obtained from a plurality of audio signal patterns in advance as a code vector.
In the narrowband codebook 6, the autocorrelation obtained from the LPC analysis filter 2 is compared with the autocorrelation information stored in the narrowband codebook 6, and a matching process is performed. Then, the index of the best matching autocorrelation information is sent to the wideband codebook 12.

The wideband codebook 12 has, in correspondence with the narrowband codebook 6, 20th-order autocorrelation information of a wideband audio signal obtained from an audio signal of the same pattern as when the narrowband codebook 6 was created. It is stored as a code vector. When the best matching autocorrelation information is determined in the narrowband codebook 6, this index is sent to the wideband codebook 12, and the narrowband self-correlation information determined by the wideband codebook 12 to be the best matching is obtained. Broadband autocorrelation information corresponding to the correlation information is read.

The autocorrelation is a parameter in the time domain and is obtained as follows.

(Equation 2)

The wideband codebook 12 is created as follows using a wideband audio signal of 0 to 8000 kHz with a sampling frequency of 16 kHz. That is, when the wideband codebook 12 is created, the wideband audio signal is divided into frames each having a length of 32 ms and a forward period of 20 ms, and a 20th-order autocorrelation is obtained in each frame. Using this, GLA (General Lloyd Algorith
m) The algorithm creates an 8-bit codebook. This is the wideband codebook 4. Here, Ai is the frame number encoded in the i-th code vector of the wideband codebook.

The narrow-band codebook 6 is the same audio signal as used to create the wideband codebook 12, and has a sampling frequency of 8 kHz and a frequency band of 300 Hz to 340.
It is created using the one limited to 0 Hz. This narrow-band audio signal is divided into frames at the same time as when the wideband codebook 12 was created, and a 10th-order autocorrelation is determined for each frame. Then, the center of gravity of the narrow-band autocorrelation of the frame belonging to the frame number Ai is obtained, and the vector is used as the i-th code vector of the narrow-band codebook, thereby coping with the wideband autocorrelation of the wideband codebook of the frame number Ai. Let it do.

In FIG. 1, the wideband autocorrelation information read from the wideband codebook 12 is sent to an autocorrelation-linear prediction coefficient conversion circuit 13. The autocorrelation-linear prediction coefficient conversion circuit 13 converts the autocorrelation into a linear prediction coefficient. This linear prediction coefficient is sent to the LPC synthesis filter 11.

The LPC synthesis filter 11 is supplied with a signal in which the LPC residual from the LPC analysis filter 2 is up-sampled by the up-sampling circuit 8 to generate aliasing distortion, and a low-pass filter 9 is suppressed through the low-pass filter 9. Have been. The LPC synthesis filter 11 up-samples the LPC residual, uses a signal obtained by suppressing the high frequency side of the aliasing distortion as an excitation source, and obtains a linear prediction coefficient from the autocorrelation-linear prediction coefficient conversion circuit unit 13 using: LPC synthesis is performed. As a result, a wideband audio signal of 300 Hz to 7000 Hz is synthesized.

The audio signal synthesized by the LPC synthesis filter 11 is supplied to a band stop filter 14. The band stop filter 14 removes signal components in the frequency band of the input narrowband audio signal. The band stop filter 14 selects 300H included in the original narrowband audio signal from the wideband audio signal having a frequency of 300 Hz to 7000 Hz synthesized by the LPC synthesis filter 11.
Signal components of z to 3400 Hz are removed. The output of the band stop filter 14 is supplied to the adding circuit 5.

The frequency of 300 Hz to 34 passed through the up-sampling circuit 3 and the band-pass filter 4 in the adding circuit 5
The original narrow band audio signal component of 00 Hz and the frequency of 3400 Hz to 70
The component of the voice signal synthesized with the voice of 00 Hz is added. Thereby, the frequency band is 300 to 7000 Hz
Thus, a digital audio signal having a sampling frequency of 16 kHz is obtained. This digital audio signal is output to output terminal 1
5 is output.

As described above, in the audio bandwidth extending apparatus to which the present invention is applied, the input narrowband audio signal is analyzed using the narrowband codebook 6, and the wideband audio signal is synthesized using the wideband codebook 12. Is done. Then, autocorrelation is used as codebook information. Conventionally,
In general, LPC cepstrum was used as the spectrum envelope information. However, as a result of experiments, it has been found that using an autocorrelation that is not on a logarithmic scale is more audible than using LPC cepstrum. This is L
Since PC Ceptrum uses a logarithmic scale,
It is considered that the error is small in a consonant part with low power, but the error is relatively large in a vowel part with high power.

In the speech bandwidth extension system to which the present invention is applied, an excitation source that upsamples the LPC residual, generates aliasing distortion, and suppresses the high frequency side of aliasing distortion is used. In this way, since the power and harmonic structure of the original sound are preserved, sufficient performance as an excitation source can be obtained.

As described above, the autocorrelation is used as the information of the codebooks 6 and 12, the LPC residual is up-sampled, and the audio signal is synthesized by using the high frequency side of the aliasing distortion suppressed as the excitation source. , And the LPC synthesis filter 11, a good wideband audio signal of 300 Hz to 7000 Hz can be obtained.

Thus, the LPC synthesis filter 11
Since the wideband audio signal obtained from the above also includes the signal of the frequency component of the original band, and the frequency components of the original band are distorted by these processes, the LPC synthesis filter 11
If the output signal is used as it is, the influence of the distortion of the frequency component of the original band occurs.

Therefore, the output of the LPC synthesis filter 11 is set to 300 Hz to
The frequency component of the original band of 3400 Hz is removed, and 300 Hz to 34 Hz extracted through the band-pass filter 4.
The component of the original audio signal of 00 Hz and the component of the audio signal of 3400 Hz to 7000 Hz synthesized by the LPC synthesis filter 11 are added.

In the distance calculation at the time of creating the code book, a weighting process may be performed so that the weight of higher-order data is reduced. That is, in the narrow-band codebook 6, the weight from the first to the third order is “1”.
The weight is set to "0" for higher orders, the weight for the first to sixth orders is set to "1" in the wideband codebook 12, and the weight is set to "0" for higher orders.
In this way, not only can the memory capacity be saved, but also the emphasis is placed on the reproduction of a rough spectral envelope as a property of the autocorrelation parameter, and a higher quality voice can be obtained.

By the way, when the autocorrelation is used as the code vector, the LPC residual is up-sampled, and the high frequency band is suppressed and the excitation source is used as the excitation source to form a wideband speech signal by LPC synthesis. In particular, the fricative and affricate are insufficient, resulting in a crisp sound. this is,
The cause may be that the spectrum envelope is not sufficiently predicted, but this is mainly due to insufficient power of the excitation source.

Therefore, in the system to which the present invention is applied, an affricate detection circuit 7 for detecting a fricative or affricate is provided, and when a fricative or affricate is detected, the whole or a part of the excitation source is detected. Is provided. The tenth autocorrelation obtained by the LPC analysis filter 2 is supplied to the affricate detection circuit 7. The affricate detection circuit 7 uses the 0th-order frame power, the 1st-order autocorrelation, and the 2nd-order autocorrelation of the 10th-order autocorrelation to detect whether a fricative or affricate has been input. Is done. When a fricative or affricate is detected by the affricate detection circuit 7, the boost circuit 10 boosts all or a part of the band of the excitation source.

That is, as a result of analyzing the autocorrelation of the input voice signal, it is found that the vowel and the fricative or affricate have 0
The following differences were found in the positional relationship between the next autocorrelation, that is, the frame power, the first order autocorrelation, and the second order autocorrelation. That is, assuming that the 0th-order frame power is R0, the 1st-order autocorrelation is R1, and the 2nd-order autocorrelation is R2, as shown in FIG. , First-order autocorrelation R
The first and second order autocorrelations R2 are arranged on a substantially straight line. On the other hand, as shown in FIG.
The positional relationship between the next frame power R0, the first-order autocorrelation R1, and the second-order autocorrelation R2 is such that they are arranged in a convex shape. From this, the 0th-order frame power is R0, 1
By judging whether or not the positional relationship between the next autocorrelation R1 and the second-order autocorrelation R2 is convex downward, it is possible to detect a fricative or affricate.

Utilizing this, in the system to which the present invention is applied, when the following conditions are satisfied, it is determined that the sound is a fricative or affricate.

Condition (1) If R0 is equal to or greater than a certain value, R1 is equal to or more than a certain value, and R1 / R2 is equal to or less than a certain value, it is determined that the sound is a fricative or affricate.

Condition (2) When R0 is equal to or more than a certain value and equal to or less than a certain value, and R1 is equal to or less than a certain value and 1-R1> R1-R2, it is determined that the sound is a fricative sound or a plosive sound. I do.

Condition (3): R0 is equal to or more than a certain value and equal to or less than a certain value, and (R1-d
c) / (R0-dc) is equal to or less than a certain value, and 1-R
If 1> R1−R2, it is determined that the sound is a fricative sound or a plosive sound. Note that dc is a constant value for each frame band.

If it is determined that the sound is a fricative or affricate according to the condition (1) or the condition (2), the excitation source is set to, for example, 1
It is boosted by 0 dB. When it is determined that the sound is a fricative or affricate according to the condition (3), the excitation source is boosted by, for example, 5 dB.

If the excitation source is instantaneously boosted when the above conditions are satisfied, the sound suddenly changes, giving a sense of incongruity. Therefore, the boost of the excitation source is smoothed for each frame so that the excitation source does not change abruptly, so that the change of the boost of the excitation source is not noticeable.

It is evident from experiments that the voice bandwidth expansion system to which the present invention is applied performs voice bandwidth expansion with good characteristics. That is, FIG.
FIG. 9 shows experimental results when the bandwidth of an audio signal is extended using the audio bandwidth extension system to which the present invention is applied. FIG. 4A is a spectrum diagram of a wideband audio signal serving as a source. The audio signal serving as the source is band-limited as shown in FIG. 4B, and the bandwidth is extended by the audio bandwidth extension system to which the present invention is applied. FIG. 4C is an audio signal obtained by extending the bandwidth of this signal. 4A and 4C, it can be seen that the audio signal bandwidth expansion system to which the present invention is applied can expand the audio signal bandwidth with considerable accuracy.

The present invention can be used for improving the sound quality of an analog telephone line and the sound quality of a digital portable telephone. In particular, VSELP and PSI-CELP are used as modulation schemes in digital mobile phones. Since VSELP and PSI-CELP use linear prediction coefficients and excitation sources, such information can be used for LPC analysis and LPC synthesis in a voice bandwidth extension system.

FIG. 5 shows an example of application to a digital portable telephone. As shown in FIG. 5, in a digital cellular phone, excitation sources and linear prediction coefficients α _{1 to} α ₁₀
Alternatively, an equivalent parameter is sent. This excitation source is supplied to the input terminal 21 and the linear prediction coefficient is supplied to the input terminal 22. The excitation source from the input terminal 21 is
The signal is sent to the LPC synthesis filter 23 and to the up-sampling circuit 24. The autocorrelation coefficient from the input terminal 22 is sent to the LPC synthesis filter 23.

In the LPC synthesis filter 23, the input terminal 21
The audio signal is synthesized using the linear prediction coefficient from the input terminal 22 based on the excitation source from the. The audio signal synthesized by the LPC synthesis filter 23 is output to an up-sampling circuit 25.
Supplied to

The up-sampling circuit 25 is for up-sampling the sampling frequency. The output of the up-sampling circuit 25 is supplied to an adding circuit 27 via a band-pass filter 26. The path leading to the up-sampling circuit 25, the band-pass filter 26, and the adding circuit 27 is a path for adding the signal of the component of the original frequency band to the synthesized audio signal.

The linear prediction coefficient is sent from the LPC synthesis filter 23 to the linear prediction coefficient-autocorrelation conversion circuit 28. The linear prediction coefficient-autocorrelation conversion circuit 28 converts the linear prediction coefficient into autocorrelation. This autocorrelation is sent to the narrowband codebook 29 and to the affricate detection circuit 30.

The excitation source from the input terminal 21 is sent to the up-sampling circuit 24. Upsampling circuit 2
4 is sent to an LPC synthesis filter 33 via a low-pass filter 31 and a boost circuit 32. The boost circuit 32 boosts the excitation source when an affricate or fricative is detected. The boost amount of the boost circuit 32 is controlled by an output of the affricate detector 30.

In the narrowband codebook 29, autocorrelation information of narrowband audio signals obtained from a plurality of audio signal patterns is stored in advance as code vectors. In the narrowband codebook 29, the autocorrelation from the linear prediction coefficient-autocorrelation conversion circuit 28 is compared with the autocorrelation information stored in the narrowband codebook 29, and a matching process is performed. Then, the index of the best matching autocorrelation information is sent to the wideband codebook 34.

In the wideband codebook 34, corresponding to the narrowband codebook 29, autocorrelation information of a wideband audio signal obtained from an audio signal of the same pattern as when the narrowband codebook 29 was created is used as a code vector. Is stored. When the best matching autocorrelation information is determined in the narrowband codebook 29, this index is sent to the wideband codebook 34, and the narrowband self-correlation information determined to be the best matching by the wideband codebook 34 is obtained. Broadband autocorrelation information corresponding to the correlation information is read.

The wideband autocorrelation information read from the wideband codebook 34 is sent to an autocorrelation / linear prediction coefficient conversion circuit 35. The autocorrelation-linear prediction coefficient conversion circuit 35 converts the autocorrelation into a linear prediction coefficient. This linear prediction coefficient is sent to the LPC synthesis filter 33.

The LPC synthesis filter 33 performs LPC synthesis. As a result, a wideband audio signal is synthesized. The audio signal synthesized by the LPC synthesis filter 33 is
The signal is supplied to the band stop filter 36. The output of the band stop filter 36 is supplied to the adding circuit 27.

The up-sampling circuit 2 is added by the adding circuit 27.
5. The component of the original narrow-band audio signal passed through the band-pass filter 26 and the component of the high-band voice signal synthesized through the band-stop filter 36 are added.
Thereby, a wideband audio signal is obtained. This audio signal is output from the output terminal 37.

As described above, in a portable telephone system using VSELP or PSI-CELP as a modulation method, since a linear prediction coefficient and an excitation source are transmitted, it is necessary to extend the voice bandwidth by using such information. I can do it.

[0059]

According to the present invention, the input audio signal is converted to LP
When the C is synthesized to extend the bandwidth, an affricate is detected to detect a fricative or affricate, and a boost is applied to the excitation source when the fricative or affricate is detected. .
For this reason, the power shortage when a fricative or fricative is input is improved, and fricative or fricative can be clearly reproduced.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a voice bandwidth extension system to which the present invention is applied.

FIG. 2 is a graph used for describing an audio bandwidth extension system to which the present invention is applied.

FIG. 3 is a graph used for describing an audio bandwidth extension system to which the present invention is applied.

FIG. 4 is a spectrum diagram used to explain the effect of the voice bandwidth extension system to which the present invention is applied.

FIG. 5 is a block diagram showing an example in which the present invention is applied to a mobile phone.

FIG. 6 is a block diagram used for describing an audio transmission path whose frequency band is restricted.

FIG. 7 is a block diagram used for describing a conventional voice bandwidth extension system.

[Explanation of symbols]

2 ... LPC analysis filter, 6 ... Narrow band codebook, 7 ... Affect detection circuit, 11 ... LPC synthesis filter, 12 ... Broadband codebook

Claims (12)

[Claims] 1. An audio signal processing apparatus which analyzes an input audio signal, performs signal processing on the analyzed audio signal, and synthesizes an audio signal. An audio signal processing device comprising: an affricate detection means for detecting an affricate; and a boost means for providing a boost to an excitation source when the fricative or affricate is detected. 2. The fricative sound detecting means detects fricative sound and fricative sound using at least the value of the autocorrelation of the input audio signal and the value of frame power.
An audio signal processing device as described in the above. 3. The audio signal processing apparatus according to claim 1, wherein said boost means changes a boost value gradually. 4. An audio signal processing method for analyzing an input audio signal, subjecting the analyzed audio signal to signal processing, and then synthesizing the audio signal. An audio signal processing method comprising: detecting an affricate; and boosting the entire band or a part of the band of the excitation source when the fricative or affricate is detected. 5. The audio signal processing method according to claim 4, wherein the detection of the affricate is performed using at least an autocorrelation value and a frame power value of the input audio signal. 6. The audio signal processing method according to claim 4, wherein said boost value is gradually changed. 7. An analysis means for obtaining a parameter from an input narrowband audio signal, an excitation source forming means for obtaining an excitation source from an LPC residual of the input narrowband audio signal, and an excitation source forming means for obtaining an excitation source from a plurality of audio signal patterns in advance. A narrowband codebook in which the parameters of the narrowband audio signal are stored, and a wideband codebook in which the parameters of the wideband audio signal obtained in advance from a plurality of audio signal patterns are stored corresponding to the narrowband codebook, Fricative sound, affricate detection means for detecting affricate, the fricative sound, boost means for boosting the excitation source when the affricate is detected, and parameters of the input narrowband audio signal Matching means for comparing parameters of the input narrowband audio signal stored in the narrowband codebook and searching for optimal parameters; Based on the search result in the matching means, a corresponding parameter is read out from the parameters of the wideband speech signal stored in the wideband codebook, and the output wideband speech is output based on the read parameter and the excitation source. An audio bandwidth extending device comprising: synthesizing means for synthesizing a signal. 8. The fricative sound detecting means detects fricative sound and fricative sound using at least the value of the autocorrelation of the input audio signal and the value of frame power.
A voice bandwidth extender as described. 9. The audio bandwidth extending apparatus according to claim 7, wherein said boost means changes a boost value gradually. 10. A narrow-band codebook in which parameters of a narrow-band audio signal previously obtained from a plurality of audio signal patterns are stored, and parameters of a wide-band audio signal previously obtained from a plurality of audio signal patterns are stored in the narrow-band codebook. A wideband codebook stored in correspondence with the narrowband codebook, and analyzing parameters for the input narrowband audio signal to obtain an excitation source from the LPC residual of the input narrowband audio signal; When a sound is detected, a boost is applied to the excitation source when the fricative or affricate is detected, and a parameter of the input narrowband audio signal and an input narrowband stored in the narrowband codebook are input. Comparing with the parameters of the band audio signal, searching for the optimal parameters, and storing the parameters in the wideband codebook based on the search result of the matching. A corresponding parameter from among the parameters of the broadband audio signal being read, and synthesizing an output broadband audio signal based on the read parameter and the excitation source. . 11. The audio bandwidth extending method according to claim 10, wherein said affricate or fricative is detected by using an autocorrelation value and a frame power value of said input audio signal. 12. The audio bandwidth extending method according to claim 10, wherein the boost value is gradually changed.