JP2003280696A - Voice enhancement device and voice enhancement method - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To reduce voice distortion and sufficiently remove noise. SOLUTION: A second comb filter generation unit 109 controls a second speech / non-speech discrimination unit 1 so as not to be affected by noise information.
A threshold value of 07 is set high, and a reference comb filter for restoring the voice pitch harmonic structure based on the presence or absence of the voice component in each frequency component is generated. Pitch estimation section 111
Estimates the speech pitch period from the speech spectrum output from the frequency division unit 104, and outputs the estimation result to the pitch harmonic structure restoration unit 112. Pitch harmonic structure restoration unit 11
2 restores the pitch harmonic structure based on the estimation result and the result of the second comb filter generation unit 109, and outputs the result to the comb filter correction unit 113. The comb filter correction unit 113 corrects the comb filter by combining the estimation result output from the pitch harmonic structure recovery unit 112 and the result output from the first comb filter generation unit 108.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice enhancement device and a voice enhancement method, and more particularly to a voice enhancement device and a voice enhancement method suitable for use in communication.

[0002]

2. Description of the Related Art In conventional voice coded communication, when voice is coded at a low bit rate for communication, it is possible to provide a voice call with high quality for voice without background noise. , The sound with background noise causes annoying distortion peculiar to low bit rate coding, and the sound quality deteriorates.

In order to deal with this problem of sound quality deterioration, noise suppression processing may be performed. The spectral subtraction method is used as a speech enhancement technology for this noise suppression processing.
And there is a comb filter method.

Spectral subtraction method (SS method)
Pays attention to the noise information, estimates the nature of noise in a silent section, subtracts the short-time power spectrum of noise from the short-time power spectrum of the voice signal containing noise, or multiplies the attenuation coefficient by This is a method of estimating the power spectrum and suppressing noise. The SS method is described in, for example, Document 1 (S. Boll, Suppression of acoustic noise in speech.
using spectral subtraction, IEEE Trans.Acoustics, S
peech, and Signal Processing, vol.ASSP-27, pp.113-12
0,1979), reference 2 (RJMcAulay, MLMalpass, Speech en.
hancement using a soft-decision noise suppression f
ilter, IEEE.Trans.Acoustics, Speech, and Signal Proce
ssing, vol.ASSP-28, pp.137-145.1980).

However, in the SS method, distortion resulting from residual noise between voice pitches occurs as residual noise after the noise is subtracted from the voice.

[0006] For example, in the SS method shown in Document 1 used as a speech enhancement method, only noise information is focused on, noise characteristics in a short time are regarded as stationary, and speech and noise are not distinguished, and a noise base is uniformly applied. Subtract the (estimated noise spectral characteristics). However, the SS method does not use voice information (for example, voice pitch). In actual communication, since the noise characteristic is not stationary, residual noise after the subtraction, particularly residual noise between voice pitches causes an unnatural distortion called "musical noise".

As a method for improving this unnatural distortion, a method of multiplying an attenuation coefficient based on the ratio of voice power to noise power (SNR) to attenuate noise, for example, Japanese Patent No. 27146.
No. 56 and Japanese Patent Application No. 9-518820 have been proposed.

[0008] In the methods disclosed in these, different attenuation coefficients are used by distinguishing a band of large speech (large SNR) and a band of large noise (small SNR) from each other.
Suppressed musical noise and improved sound quality.

However, in the methods disclosed in Japanese Patent No. 2714656 and Japanese Patent Application No. 9-518820, the number of frequency channels to be processed (16 channels) is not sufficient.
It is difficult to separate voice pitch harmonic information from noise and extract it.

Further, since the attenuation coefficient is used for both the voice and noise bands, they influence each other, so that the attenuation coefficient cannot be increased. For example, if the attenuation coefficient is increased, the SNR estimation error may cause distortion of the voice. As a result, noise attenuation is insufficient.

On the other hand, the comb filter method is a method of paying attention to voice information and applying noise to the voice pitch by applying a comb filter. The comb filter is a filter that outputs a signal with or without attenuating a signal input in a frequency domain unit at a predetermined ratio, and has a comb-like attenuation characteristic. When the comb filter method is realized by digital data processing, in the comb filter method, noise can be suppressed by creating attenuation characteristic data of the comb filter for each frequency domain and multiplying the voice spectrum for each frequency.

As a document relating to the comb filter method, a document 3 (JSLim etc., Evaluation of anadaptive comb filt
ering method for enhancing speech degraded by whit
e noise addition, IEEE Trans. Acoustics, Speech, and S
ignal Processing, vol.ASSP26, pp.354-358, 1978).

In the conventional comb filter method, if there is an estimation error in the pitch, which is the fundamental frequency, the error increases in the higher harmonics, and the original higher harmonic component is more likely to deviate from its pass band. Further, there is a problem in feasibility because it is necessary to distinguish between voices having quasi-periodicity and voices that do not. Further, in mobile communication, by simply suppressing noise, ambient noise having a natural feeling is also suppressed, which may cause discomfort in a call.

As a method of eliminating the discomfort of a telephone call, there is a technique of separating voice and noise, and reproducing and reproducing good voice and ambient noise by performing encoding and decoding, respectively. For example, the method shown in Reference 4 (Kimio Mizeki, Masahiro Oshikiri, Low-rate speech coding based on speech / background noise separation, Proceedings of the Acoustical Society of Japan, pp.235-236, March 1998) is used. is there.

The method described in Document 4 is a method in which voice enhancement is performed using the SS method, a noise-suppressed input signal is used as a voice component, and a result obtained by subtracting the input signal and the voice component is used as a noise component. Since it is basically the same method as the SS method, it is difficult to obtain good voice and noise characteristics by the separation method according to this method.

[0016]

As described above, in the conventional device, there is a problem that the distortion of the voice is small and it is difficult to sufficiently remove the noise.

The present invention has been made in view of the above points, and provides a voice emphasizing device which has a small amount of distortion of voice and is capable of sufficiently removing noise, and at the same time, excellent characteristics of voice and noise can be obtained. An object is to provide a voice enhancement device and a voice enhancement method.

[0018]

A speech emphasizing apparatus according to the present invention comprises frequency division means for outputting a frequency division spectrum obtained by dividing the spectrum of an input signal in predetermined frequency units,
A first comb filter creating means for creating a first comb filter for attenuating a frequency domain signal determined to be silent based on the frequency divided spectrum, and removing more noise peaks than the first comb filter based on the frequency divided spectrum Second comb filter creating means for creating a second comb filter which is a filter, and a voice pitch harmonic structure included in the first comb filter with a voice pitch estimated from the second comb filter and the frequency division spectrum. Comb filter modifying means for modifying, suppressing means for suppressing the noise of the frequency-divided spectrum using the first comb filter modified by the comb-filter modifying means, and the frequency-divided spectrum in which the noise is suppressed is continuous in the frequency domain. And a voice frequency synthesizing means for synthesizing the same into the spectrum signal. The take.

According to this configuration, a comb filter for estimating a voice pitch by extracting a peak which is likely to be a voice peak from the spectrum of the voice signal is created, and accurate voice pitch information is acquired from the comb filter. In addition, a comb filter that extracts as much speech information as possible to suppress the noise signal is created, and the peak of the speech spectrum buried in noise is not suppressed by using this comb filter. It is possible to create a comb filter that compensates for the missing voice pitch harmonic structure, and by suppressing noise signals using this comb filter, it is possible to perform voice enhancement with less voice distortion.

The speech enhancement apparatus of the present invention comprises speech / noise frame detection means for determining whether or not the speech spectrum includes a speech component from the first comb filter and the second comb filter, and the comb filter correction means. If the result of the determination by the voice / noise frame detection means is that the voice component is not included, the first comb filter is modified to attenuate the signal at each frequency component.

In the speech emphasizing apparatus of the present invention, the speech / noise frame detecting means includes a sum of power spectra of the input signal in the pass band of the first comb filter and a power spectrum of the input signal in the stop band of the first comb filter. The first result is the ratio of the sums, and the second result is the ratio of the sum of the power spectra of the input signals in the pass band of the second comb filter to the sum of the power spectra of the input signals in the stop band of the second comb filter. If the result of adding the result and the second result is larger than a predetermined threshold value, the addition result is used, and if the addition result is less than or equal to the predetermined threshold value, whether the voice spectrum is included in the voice spectrum using the second result. Take the configuration to judge whether or not.

According to these configurations, it is determined from the first comb filter and the second comb filter whether or not the voice spectrum includes a voice component, and when the determination result is that the voice component does not include the voice component, By attenuating the signal with each frequency component in the first comb filter, it is possible to suppress the noise that is suddenly generated.

In the voice emphasizing device of the present invention, the first comb filter generating means adopts a construction for forming a first comb filter having a predetermined range as a stop band from the minimum value of the power spectrum of the input signal.

According to this configuration, a comb filter having a predetermined range from the minimum value of the power spectrum of the input signal as a stop band is created, and the noise signal is suppressed by using this comb filter, whereby the voice signal and the noise signal are suppressed. Even if the level difference is small, the voice pitch harmonic structure can be extracted and repaired to reduce the voice distortion.

The speech emphasizing device of the present invention comprises pitch estimating means for subtracting the noise base from the power spectrum of the input signal and estimating the speech pitch using the autocorrelation function of the subtraction result. The voice pitch harmonic structure included in the first comb filter is modified by the voice pitch estimated by the pitch estimating means.

According to this configuration, the noise base is subtracted from the power spectrum of the input signal, the voice pitch is estimated based on the autocorrelation function of the subtraction result, and the voice pitch harmonic included in the comb filter is estimated with the estimated voice pitch. By modifying the structure, the pitch harmonic structure of the comb filter can be restored, and the voice enhancement with less voice distortion can be performed.

The speech emphasizing device of the present invention comprises a direct current component generating means for generating a pseudo peak having a predetermined power in the direct current component as a result of subtracting the noise base from the power spectrum of the input signal, and the pitch estimating means is The voice pitch is estimated from the power spectrum in which the pseudo peak is generated by the DC component generating means.

According to this structure, a pseudo peak having a predetermined power in the DC component is generated as a result of subtracting the noise base from the power spectrum of the input signal, and the voice is generated based on the autocorrelation function of the spectrum in which the DC component is generated. By estimating the pitch and modifying the voice pitch harmonic structure included in the comb filter with the estimated voice pitch, pitch information can be obtained by obtaining pitch information even if there are few harmonic peaks in the voice spectrum. Restoration can be performed, and voice enhancement with less voice distortion can be performed.

The speech enhancement apparatus of the present invention has the following result obtained by subtracting the noise base from the power spectrum of the input signal:
The noise characteristic estimating means for calculating the moving average of the number of frequency regions whose power is equal to or higher than the predetermined threshold is provided, and the second comb filter creating means determines whether or not the input signal includes voice from the moving average. The configuration that creates the second comb filter from the result is adopted.

According to this configuration, the distribution of the noise level of the input signal is detected, the reference for generating the comb filter is determined from the voice spectrum based on this distribution, and the pitch information is acquired from the created comb filter. , It is possible to obtain pitch information according to the state of noise and create a comb filter, and it is possible to perform speech enhancement with less speech distortion.

The speech enhancement apparatus of the present invention has the following result obtained by subtracting the noise base from the power spectrum of the input signal.
The second comb filter creating means has a noise characteristic estimating means for calculating a moving average of the number of frequency regions whose power is equal to or higher than a predetermined threshold, and the second comb filter creating means has a moving average calculated by the noise characteristic estimating means at a predetermined value or less. In some cases, a second comb filter having a predetermined frequency range as a stop band is used.

According to this structure, the frequency component is selected based on the estimation result of the noise characteristic, and the selected frequency region is converted into the stop band in the second comb filter. Accurate restoration of the pitch harmonic structure by reducing the generated false pitch harmonics and repairing the pitch harmonic structure based on the pitch harmonics in the low frequency range where false pitch harmonics are less likely to occur. You can

The speech emphasizing apparatus of the present invention comprises SNR estimating means for calculating a signal-to-noise ratio from the power spectrum of the input signal and the noise base, and the suppressing means suppresses the noise of the frequency division spectrum from the signal-to-noise ratio. Adopt a configuration that determines the amount of suppression.

According to this configuration, in the pass band and stop band of the modified comb filter, the amount by which the noise base is subtracted from the input speech power spectrum and the degree of noise attenuation are calculated as SNR.
By adjusting according to the magnitude of the estimated value, it is possible to perform appropriate noise attenuation even in environments with different SNRs, and realize voice enhancement with less voice distortion and residual noise.

In the speech emphasizing apparatus of the present invention, the SNR estimating means calculates the level of the speech component from the moving average value of the power spectrum of the input signal, and weights the noise component level to the noise-based estimation value for each frequency component. It is calculated from the value multiplied by the count, and the signal-to-noise ratio is calculated from the ratio of the level of the voice component and the level of the noise component.

According to this structure, the noise base is subtracted from the moving average value of the input voice power spectrum to calculate the voice level, thereby reducing the influence of noise and reducing the SN.
An accurate voice level can be calculated even under the R environment. In addition, by multiplying each frequency component of the noise-based estimated value by a weighting coefficient for calculation, it is possible to appropriately attenuate different noises and reduce voice distortion.

The speech enhancement apparatus of the present invention calculates the deviation between the signal-to-noise ratio and the moving average value of the signal-to-noise ratio, and uses the deviation to update the moving-average value of the signal-to-noise ratio. The suppressing means is provided with a suppressing means, and the suppressing means determines a noise suppressing amount of the frequency division spectrum from the moving average value of the signal-to-noise ratio updated by the fluctuation suppressing means.

According to this configuration, the SNR estimated value and the SNR
The deviation of the long-term moving average of the estimated value is calculated, and the long-term moving average of the SNR estimated value and a part of the deviation are added and used as the SNR estimated value, thereby effectively suppressing the fluctuation of the SNR and stabilizing Further, the level of noise attenuation can be adjusted according to the magnitude of SNR.

The speech emphasizing device of the present invention calculates two noise-based moving average values having different update speeds in a predetermined time unit, and uses the second moving average value having a faster update speed than the first moving average value as the first moving average value. A configuration is provided that includes a noise base updating unit that changes a moving average value update condition and outputs the first moving average value as a noise base estimated value.

According to this configuration, noise-based estimation can be performed using the moving average coefficient having a high update rate, so that a rapid change in the noise level can be tracked even in the voice section. In addition, the noise-based update with a slow update speed can be accurately estimated by performing the noise-based update with a fast update speed, and the noise-based update can be stopped due to a sudden change in the noise level. Can be prevented.

The wireless communication apparatus of the present invention has a configuration including any one of the above-described voice enhancing apparatuses.

According to this structure, a comb filter for estimating a voice pitch by extracting a peak which is likely to be a voice peak from the spectrum of the voice signal is created, and accurate voice pitch information is obtained from this comb filter. In addition, a comb filter that extracts as much speech information as possible and suppresses the force signal is created, and the peak of the speech spectrum buried in noise is not suppressed by using this comb filter. It is possible to create a comb filter that compensates for the missing voice pitch harmonic structure, and by suppressing noise signals using this comb filter, it is possible to perform voice enhancement with less voice distortion.

The noise suppression apparatus of the present invention passes through a frequency division means for outputting a frequency division spectrum obtained by dividing the spectrum of the input signal in a predetermined frequency unit, and a signal in the frequency domain determined to be silent based on the frequency division spectrum. Noise separating comb filter creating means for creating a noise separating comb filter as a frequency band, a suppressing means for separating a noise component of the frequency divided spectrum using the noise separating comb filter, and the frequency divided spectrum obtained by separating the noise component. And a voice frequency synthesizing means for synthesizing a continuous spectrum signal in the frequency domain.

According to this structure, the noise characteristic can be extracted to the maximum extent by generating the noise dedicated comb filter.

In the noise suppressing device of the present invention, the noise separating means is configured to multiply the real part and the imaginary part of the input speech spectrum by different random numbers and noise-based estimation values in the pass band of the noise separating comb filter. Take.

According to this configuration, noise components are not attenuated in the stop band of the noise separation comb filter, and separate random numbers are provided for the real part and the imaginary part of the input speech spectrum in the pass band of the noise separation comb filter. And the noise-based estimation value are multiplied, the amplitudes and phases of the real and imaginary parts of the noise component are all randomized, and good noise separation characteristics can be obtained.

The noise suppressing device of the present invention comprises a noise component storing means for storing the spectral component of the input voice in the stop band of the voice separating comb filter, and the noise separating means is
A configuration is used in which the spectral components stored in the memory are used in the pass band of the noise separation comb filter.

According to this configuration, the spectral component of the input voice in the stop band of the noise separating comb filter is stored in the memory, and the value is used in the pass band of the noise separating comb filter to obtain the actual noise and characteristics. Can be reconstructed, and good noise separation characteristics can be obtained.

A radio communication apparatus of the present invention has a configuration including any one of the noise suppressing apparatuses described above.

According to this configuration, the noise characteristic can be maximized by generating the noise-only comb filter.

The sound source separation device of the present invention has a configuration including any one of the above speech enhancement device and any one of the above noise suppression devices.

According to this structure, a comb filter for estimating a voice pitch by extracting a peak that is likely to be a voice peak from the spectrum of the voice signal is created, and accurate voice pitch information is acquired from this comb filter. In addition, a comb filter that extracts as much speech information as possible and suppresses the force signal is created, and the peak of the speech spectrum buried in noise is not suppressed by using this comb filter. It is possible to create a comb filter that compensates for the missing voice pitch harmonic structure, and by suppressing noise signals using this comb filter, it is possible to perform voice enhancement with less voice distortion. Further, according to this configuration, the noise characteristics can be maximized by generating the noise-only comb filter.

The speech emphasizing method of the present invention includes a frequency division step of outputting a frequency division spectrum obtained by dividing the spectrum of an input signal in predetermined frequency units, and a signal in a frequency domain determined to be silent based on the frequency division spectrum. Create a first comb filter Create a first comb filter and create a second comb filter that is a filter that removes more noise peaks than the first comb filter based on the frequency division spectrum A step, a comb filter correction step that corrects a voice pitch harmonic structure included in the first comb filter with a voice pitch estimated from the second comb filter and the frequency division spectrum, and is corrected in the comb filter correction step. Suppress the noise of the frequency division spectrum using the first comb filter And as stroke, and to be provided with audio frequency synthesizing step of synthesizing the frequency division spectrum suppressed noise spectrum signals continuous in the frequency region.

According to this method, a comb filter for estimating a voice pitch by extracting a peak which is likely to be a voice peak from the spectrum of a voice signal is created, and accurate voice pitch information is obtained from this comb filter. In addition, a comb filter that extracts as much speech information as possible and suppresses the force signal is created, and the peak of the speech spectrum buried in noise is not suppressed by using this comb filter. It is possible to create a comb filter that compensates for the missing voice pitch harmonic structure, and by suppressing a voice signal using this comb filter, voice enhancement with less voice distortion can be performed.

The noise suppression method of the present invention passes the frequency division process of outputting the frequency division spectrum obtained by dividing the spectrum of the input signal in a predetermined frequency unit, and the signal in the frequency domain determined to be silent based on the frequency division spectrum. A noise separation comb filter creation step for creating a noise separation comb filter as a band, a suppression step for separating a noise component of the frequency division spectrum by using the noise separation comb filter, and the frequency division spectrum obtained by separating the noise component. And a voice frequency synthesizing step for synthesizing a continuous spectrum signal in the frequency domain.

According to this method, the noise characteristics can be extracted to the maximum extent by generating the noise dedicated comb filter.

A speech emphasizing program of the present invention outputs a frequency-divided spectrum obtained by dividing the spectrum of an input signal into predetermined frequency units, and attenuates a signal in a frequency domain determined to be silent based on the frequency-divided spectrum. The first comb filter creation step to create the first comb filter and the second comb filter creation to create the second comb filter, which is a filter that removes more noise peaks than the first comb filter based on the frequency division spectrum Step, a comb filter correction step of correcting a voice pitch harmonic structure included in the first comb filter with a voice pitch estimated from the second comb filter and the frequency division spectrum, and corrected by the comb filter correction means. The frequency division spectrum using the first comb filter Take a suppression step of suppressing Le noise, the structure to execute the audio frequency synthesizing step of synthesizing said frequency division spectrum suppressed noise spectrum signals continuous in the frequency domain, to the computer.

According to this configuration, a comb filter for estimating a voice pitch by extracting a peak which is likely to be a voice peak from the spectrum of the voice signal is created, and accurate voice pitch information is acquired from this comb filter. In addition, a comb filter that extracts as much speech information as possible to suppress the noise signal is created, and the peak of the speech spectrum buried in noise is not suppressed by using this comb filter. It is possible to create a comb filter that compensates for the missing voice pitch harmonic structure, and by suppressing noise signals using this comb filter, it is possible to perform voice enhancement with less voice distortion.

The noise separation program of the present invention passes a frequency division step of outputting a frequency division spectrum obtained by dividing the spectrum of an input signal into predetermined frequency units, and a signal in the frequency domain determined to be silent based on the frequency division spectrum. A noise separation comb filter creating step for creating a noise separation comb filter, a suppression step for separating a noise component of the frequency division spectrum using the noise separation comb filter, and the frequency division spectrum obtained by separating the noise component. An audio frequency synthesizing step of synthesizing a continuous spectrum signal in the frequency domain, and a configuration for causing a computer to execute.

According to this configuration, the noise characteristic can be maximized by generating the noise-only comb filter.

The server device of the present invention outputs a frequency division spectrum obtained by dividing the spectrum of the input signal in predetermined frequency units, and attenuates the signal in the frequency domain determined to be silent based on the frequency division spectrum. First comb filter creation step to create the first comb filter, and second comb filter creation step to create the second comb filter, which is a filter that removes more noise peaks than the first comb filter based on the frequency division spectrum. A comb filter correction step of correcting the voice pitch harmonic structure included in the first comb filter with the voice pitch estimated from the second comb filter and the frequency division spectrum, and the comb filter correction step. The frequency division spectrum using the first comb filter And suppression step of suppressing noise,
Storing a voice emphasizing program characterized by causing a computer to execute a voice frequency synthesizing step of synthesizing the frequency-divided spectrum in which noise is suppressed into a continuous spectrum signal in the frequency domain, and the voice emphasizing program according to a request. Take the configuration to output.

According to this configuration, a comb filter for estimating a voice pitch by extracting a peak that is likely to be a voice peak from the spectrum of the voice signal is created, and accurate voice pitch information is acquired from this comb filter. In addition, a comb filter that extracts as much speech information as possible and suppresses the force signal is created, and the peak of the speech spectrum buried in noise is not suppressed by using this comb filter. It is possible to create a comb filter that compensates for the missing voice pitch harmonic structure, and by suppressing noise signals using this comb filter, it is possible to perform voice enhancement with less voice distortion.

The server device of the present invention outputs a frequency division spectrum obtained by dividing the spectrum of the input signal in predetermined frequency units, and a signal in the frequency domain determined to be silent based on the frequency division spectrum in the pass band. And a noise separation comb filter creation step of creating a noise separation comb filter, a suppression step of separating a noise component of the frequency division spectrum using the noise separation comb filter, and a frequency division of the frequency division spectrum separated noise component. A voice frequency synthesizing step for synthesizing a continuous spectrum signal in a region, and storing a noise separation program characterized by causing a computer to execute,
The noise separation program is output in response to a request.

According to this configuration, the noise characteristic can be extracted to the maximum extent by generating the noise-only comb filter.

[0065]

BEST MODE FOR CARRYING OUT THE INVENTION The essence of the present invention is to generate a comb filter in which more noise peaks are removed than a comb filter used for voice suppression based on a frequency-divided spectrum of a voice signal, and to use this comb filter for voice This is to acquire the pitch information of the signal and supplement the voice pitch of the comb filter.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. (Embodiment 1) FIG. 1 is a block diagram showing a configuration of a speech emphasizing apparatus according to Embodiment 1 of the present invention. In FIG. 1, the speech enhancement device 100 includes a time division unit 101, a windowing unit 102, an FFT unit 103, and a frequency division unit 10.
4, a noise base estimation unit 105, a first voice / non-voice identification unit 106, a second voice / non-voice identification unit 107, a first comb filter generation unit 108, a second comb filter generation unit 109, Voiced / unvoiced discrimination unit 110, pitch estimation unit 111, pitch harmonic structure restoration unit 112, comb filter correction unit 113, speech separation coefficient calculation unit 114, multiplication unit 115, speech frequency synthesis unit 116, IFFT section 1
It is mainly composed of 17.

The time division unit 101 forms a frame divided from the input audio signal in units of a predetermined time, and the window division unit 102 outputs the frame to the window division unit 102.
The frame output from 01 is subjected to windowing processing using a Hanning window or the like and output to the FFT unit 103. FFT section 103 performs FFT (Fast Fourier Transform) on the audio signal output from windowing section 102, and outputs the audio spectrum signal to frequency division section 104.

The frequency division unit 104 divides the voice spectrum output from the FFT unit 103 into frequency components, and the voice spectrum S divided for each frequency component.
_f (k) (where k is a number identifying a frequency component) is used as the noise base estimation unit 105 and the first voice / non-voice identification unit 1
06, the second voice / non-voice identification unit 107, and the multiplication unit 11
Output to 5. The frequency component indicates the minimum unit of the voice spectrum divided by a predetermined frequency unit. S _f (k) is represented by equation (1).

[0069]

[Equation 1] Here, Re {D _f (k)} ² represents the real part of the spectrum of the input speech signal after FFT conversion, and Im {D f
_f (k)} ² indicates the imaginary part of the spectrum of the input voice signal after FFT conversion.

The noise-base estimation unit 105 outputs, when the determination result that the frame does not include a voice component is output,
The noise base is updated using the short-time power spectrum for each frequency component of the audio spectrum output from the frequency division unit 104.

Specifically, the noise base in each frequency component is estimated using equation (2), and the estimated noise base is used as the speech / non-speech discrimination unit 106 and the speech / non-speech discrimination unit 1.
It outputs to 07.

[0072]

[Equation 2] Here, P _base (n−1, k) is a noise base, n is a number identifying a frame to be processed, and k is a number identifying a frequency component. Further, Θ _base represents a threshold for discriminating between speech and noise, and α represents a moving average coefficient.

First voice / non-voice identification unit 106 and second voice /
When the difference between the voice spectrum signal output from the frequency division unit 104 and the noise base value output from the noise base estimation unit 105 is equal to or greater than a predetermined threshold, the non-voice identification unit 107 includes a voiced part including a voice component. Otherwise, in other cases, it is determined to be a silent part including only noise that does not include a voice component.

Then, the first voice / non-voice discrimination section 106
Outputs the determination result to the first comb filter generation unit 108, and the second voice / non-voice identification unit 107 outputs the determination result to the second comb filter generation unit 109.

The first comb filter generation unit 108 sets the threshold value of the first voice / non-voice discrimination unit 106 low so as to extract a large amount of voice pitch harmonic information, and determines whether or not there is a voice component in each frequency component. Based on this, a comb filter that emphasizes the voice pitch harmonic structure is generated, and this comb filter result is output to comb filter correction section 113.

Specifically, the first comb filter COMB_low (k) is generated using the following equation (3).

[0077]

[Equation 3] Where Θ _low is the threshold value for the first comb filter. Also,
HB is the FFT transform length, that is, the number of data items to be subjected to the fast Fourier transform, and is set to HB = 512, for example.

The second comb filter generation unit 109 includes a second voice / non-voice discrimination unit 10 so as not to be affected by noise information.
7 is set to a high threshold value, a reference comb filter for restoring the voice pitch harmonic structure is generated based on the presence or absence of a voice component in each frequency component, and the result of the comb filter is generated by the voiced / unvoiced discrimination unit 110 and the pitch. Output to the harmonic structure restoration unit 112. Specifically, the second comb filter is generated using the following equation (4).

[Equation 4] Where Θ _high is the threshold value for the second comb filter, and Θ _high
Is greater than Θ _low .

Voiced / unvoiced discrimination section 110 discriminates between voiced and unvoiced based on the result output from second comb filter generation section 109, and outputs the discrimination result to pitch estimation section 111.

Specifically, the passband of the second comb filter (that is, COMB_hi) is divided into the low frequency region and the high frequency region of the input audio signal using the following equations (5) and (6).
Add the number of frequency components of gh (k) = 1).

[0081]

[Equation 5]

[0082]

[Equation 6] Here, if both Expressions (5) and (6) are larger than the set threshold value, or if Expression (5) is larger than the set threshold value and Expression (6) is smaller than the set threshold value, it is voiced. If not, it is determined to be unvoiced.

When it is determined that the voice is present, the voice pitch is estimated and the voice pitch harmonic structure is restored. When it is determined that the voice is unvoiced, the voice pitch is not estimated and the voice pitch harmonic structure is not restored.

The pitch estimating section 111 includes a frequency dividing section 10
The speech pitch period is estimated from the speech spectrum output from No. 4 and the estimation result is output to the pitch harmonic structure restoration unit 112.

The pitch harmonic structure restoration unit 112 restores the pitch based on the estimation result output from the pitch estimation unit 111 and the result of the second comb filter generation unit 109, and the result is sent to the comb filter correction unit 113. Output.

Specifically, the restoration of the voice pitch harmonic structure is performed in the following steps. In the first step, the power peak of the speech spectrum for each pass band of the second comb filter COMB_high (k) is extracted, and the pitch reference comb filter COMB_int that serves as a reference for the pitch harmonic structure restoration is extracted from all pass bands.
Generate (k).

In the second step, the peak-to-peak interval of the pitch reference comb filter is calculated, and when a predetermined threshold value (for example, a pitch period of 1.5 times) is exceeded, the missing pitch is determined based on the result of pitch estimation. Harmonics are inserted to generate a pitch harmonic insertion comb filter COMB_rec (k).

In the third step, the comb of the pitch harmonic insertion comb filter, that is, the pitch harmonic restoration comb filter COMB_ext that widens the width of the pass band according to the value of the pitch period.
Generate (k).

The comb filter modification unit 113 modifies the comb filter by combining the estimation result output from the pitch harmonic structure repair unit 112 and the result output from the first comb filter generation unit 108, and outputs the result to the voice. It outputs to the separation coefficient calculation unit 114.

Specifically, a pitch harmonic restoration comb filter
COMB_ext (k) passband and first comb filter COMB_low (k)
By comparing the passbands of, and if there is an overlapping part, the passband of the first comb filter is set as the passband of the modified comb filter, and the other parts are set as the stopband of the modified comb filter. A modified comb filter COMB_res (k) is generated.

The voice separation coefficient calculation unit 114 multiplies the comb filter generated in the comb filter correction unit 113 by the separation coefficient based on the frequency characteristic, and sets the separation coefficient of the input signal for each frequency component. The separation coefficient of each frequency component is output to the multiplication unit 115.

For example, from the following equation (7), the separation coefficient seps
It is also possible to calculate (k) and multiply the input signal.

[0093]

[Equation 7] Where gc is a constant, k is a variable that identifies the frequency component, and γ
Is a coefficient for adjusting the noise-based subtraction amount. Also,
P _MAX (n) indicates the maximum value of P _base (n, k). In addition, gc · P _MAX (n) / P _base (n, k) is an attenuation coefficient that normalizes the noise base estimation value for each frame and uses its reciprocal, and COMB_res (k) is a modification of the comb filter. The result.

FIG. 2 and FIG. 3 are examples of the speech enhancement result of the present invention. The steps of generating the comb filter, repairing the pitch harmonic structure, modifying the comb filter and the speech separation coefficient (attenuation coefficient) are shown in FIG. Show. FIG. 2 is a diagram showing an example of a comb filter created by the voice enhancement device according to the present embodiment. In FIG. 2, the vertical axis represents the power of the spectrum and the attenuation of the filter, and the horizontal axis represents the frequency.

The comb filter has the attenuation characteristic shown in S1, and the attenuation characteristic is set for each frequency component. The first comb filter generation unit 108 creates a comb filter having an attenuation characteristic that attenuates a frequency domain signal that does not include a voice component and does not attenuate a frequency domain signal that includes a voice signal.

The speech spectrum S2 including the noise component is S
By applying a comb filter with an attenuation characteristic of 1,
The signal in the frequency domain including the noise component is attenuated to reduce the power, and the portion including the voice signal is not attenuated and the power does not change. The obtained voice spectrum has a spectrum shape in which the frequency region of the noise component becomes lower and the peak is emphasized without being lost, and the voice spectrum S3 in which the noise in which the pitch harmonic information is not suppressed is suppressed is output.

FIG. 3 shows an example of restoration of the comb filter in the voice processing device according to this embodiment. In FIG. 3, the vertical axis represents the degree of attenuation and the horizontal axis represents the frequency component.
Specifically, the horizontal axis has 256 frequency components, and
The region from 4 kHz to 4 kHz is shown.

C1 is a generated comb filter, C2 is a comb filter in which the pitch of the comb filter C1 is restored, and C3 is a comb filter in which the pitch width is corrected in the comb filter C2.

In the comb filter C1, pitch information is lost in frequency components from 100 to 140. The pitch harmonic structure restoration unit 112 supplements the pitch harmonic information in the frequency components 100 to 140 of the comb filter C1 based on the pitch period information estimated by the pitch estimation unit 111. As a result, the comb filter C2 is obtained.

Next, the pitch harmonic structure restoration section 112 corrects the pitch harmonic width of the comb filter C2 based on the voice spectrum output from the frequency division section 104. As a result, the comb filter C3 is obtained.

The multiplication section 115 adds the speech separation coefficient calculation section 114 to the speech spectrum output from the frequency division section 104.
The separation coefficient output from is multiplied by each frequency component. Then, the result of the multiplication is output to the audio frequency synthesis unit 116.

The voice frequency synthesizer 116 has a multiplier 115.
The spectrum of each frequency component output from the above is synthesized into a continuous voice spectrum in the frequency domain in a predetermined processing time unit and output to the IFFT unit 117.

The IFFT unit 117 is a voice frequency synthesis unit 1.
The IFFT (Invers
e Fast Fourier Transform) is performed to output a signal converted into an audio signal.

As described above, according to the voice emphasizing device of this embodiment, a comb filter for estimating a voice pitch by extracting a peak which is likely to be a voice peak from the spectrum of a voice signal is created, and this comb filter is used. We obtain a precise voice pitch information from the voice, and create a comb filter that suppresses the noise signal by extracting as much voice information as possible, and do not suppress the peak of the voice spectrum buried in noise using this comb filter. As a result, it is possible to create a comb filter that compensates for the missing voice pitch harmonic structure based on the correct voice pitch information. By suppressing noise signals using this comb filter, voice enhancement with less voice distortion can be achieved. It can be carried out.

Specifically, according to the speech emphasizing device of the embodiment of the present invention, the noise pitch estimation and the speech / non-speech discrimination for each frequency component are performed so that the speech pitch harmonic information is obtained in the frequency domain. A first comb filter to extract can be generated.

Further, since the second comb filter provides the basic structure of pitch harmonics, the harmonic component of the voice due to the pitch estimation error does not deviate from the pass band of the comb filter.

Further, voiced / unvoiced discrimination is performed based on the generation result of the second comb filter, the voice pitch is estimated only in the case of voiced voice having a pitch harmonic structure, and the missing pitch harmonic is estimated based on the result. By repairing the structure, it becomes possible to restore voice information buried in noise, and reduce voice distortion due to voice pitch harmonic loss. Also,
Since whether or not to attenuate the voice spectrum is determined for each frequency component based on the result of the modified comb filter, it is possible to perform voice enhancement with less voice distortion even if the attenuation is increased.

Further, by setting the first voice / non-voice discrimination threshold to be low and generating the first comb filter, more voice information can be extracted. Meanwhile, the second voice
/ By setting the non-voice discrimination threshold to a high value and generating the second comb filter, it is possible to generate a comb filter that is not easily affected by noise information. Based on the result, the voice pitch harmonic structure can be accurately restored.

Further, by distinguishing between voiced and unvoiced based on the result of the second comb filter generation, it is possible to easily distinguish between voiced and unvoiced with a small amount of calculation. Also,
In the case of the unvoiced section, the unvoiced section without the pitch harmonic structure can be dealt with by not performing the speech pitch estimation and the restoration of the speech pitch harmonic structure.

Also, the voice pitch harmonic can be restored by inserting the pitch harmonic based on the voice pitch estimation result. Further, the width of the pitch harmonic is automatically adjusted according to the pitch estimation result, so that the influence of the voice pitch estimation error can be reduced and the voice pitch harmonic structure can be more reliably restored. Furthermore, the results of voice pitch harmonic structure restoration and the results of the first comb filter are compared. If there is an overlapping part, the pass band of the first comb filter is set as the corrected pass band of the comb filter, and the others are corrected. By setting it as the stop band of the subsequent comb filter, only the voice pitch harmonic information can be extracted and the noise information between the pitch harmonics can be suppressed.

(Second Embodiment) FIG. 4 is a block diagram showing an example of the configuration of a voice emphasizing device according to the second embodiment.
However, the same components as those in FIG. 1 are denoted by the same reference numerals as those in FIG. 1, and detailed description thereof will be omitted.

The speech emphasizing apparatus 300 of FIG. 4 comprises a speech / noise frame detector 301, judges whether or not the speech spectrum contains a speech component from the first comb filter and the second comb filter, and When the determination result is that the voice component is not included, the point that the first comb filter is modified to attenuate the signal at each frequency component is different from the voice enhancement apparatus of FIG.

Specifically, the voice emphasizing device 300 shown in FIG.
Is the ratio of the sum of the input voice power spectrum in the pass band of the first comb filter and the sum of the input voice power spectrum in the stop band of the first comb filter as the first result, and the input voice power in the pass band of the second comb filter is If the ratio of the sum of the spectrum and the sum of the input speech power spectrum in the stop band of the second comb filter is the second result, and if it is larger than a predetermined threshold value, the first result and the second result are added and When it is small, the speech / noise frame is detected by using the second result, which is different from the speech enhancement apparatus of FIG.

In FIG. 4, the first comb filter generator 1
08 and the result output from the second comb filter generation unit 109 and the input voice power spectrum are input to the voice / noise frame detection unit 301, and the voice / noise frame detection unit 30 is input.
The speech / noise frame detection result calculated in 1 is output to the comb filter correction unit 113.

Specifically, first, the SN ratio of voice and noise based on the first comb filter and the second comb filter is calculated using the following formulas (8) and (9).

[0116]

[Equation 8]

[0117]

[Equation 9] Next, the SN ratio (SNR_frame (n)) of the frame is calculated by the following equation (10).

[0118]

[Equation 10] Where Θ _sn is a threshold. Then, the speech / noise frame is detected by comparing SNR_frame (n) with Θ _sn . If the detection result of the voice / noise frame is a noise frame (that is, SNR_frame (n) <Θ _sn ), all the frequency components of the modified comb filter COMB_res (k) are set to the stop band.

As described above, according to the voice emphasizing device of the present embodiment, it is determined from the first comb filter and the second comb filter whether or not the voice spectrum includes a voice component, and this determination is a voice component. In the case where the result does not include the above, the first comb filter can be modified so as to attenuate the signal at each frequency component, so that noise that occurs suddenly can be suppressed.

Specifically, according to the voice emphasizing device of the present embodiment, the SN ratio of voice and noise is calculated based on the result of the first comb filter having a low voice / non-voice discrimination threshold. And noise becomes easier to detect. On the other hand, by calculating the SN ratio of voice and noise based on the result of the second comb filter having a high voice / non-voice discrimination threshold, it is possible to reduce false detection due to sudden noise. By using the above voice / noise frame detection, both advantages can be utilized, the voice / noise frame can be detected more reliably, and the influence of sudden noise on the detection of the voice / noise frame can be minimized. .

(Third Embodiment) FIG. 5 is a block diagram showing an example of the configuration of a voice emphasizing device according to the third embodiment.
However, the same components as those in FIG. 1 are denoted by the same reference numerals as those in FIG.

The speech emphasizing apparatus 400 of FIG. 5 is provided with a local minimum value calculating section 401, and a point of creating a comb filter having a predetermined range from the minimum value of the power spectrum of the input signal as a comb filter is shown in FIG. Different from the highlighter.

Specifically, the voice emphasizing device 400 of FIG.
1 is that, in a predetermined frequency region, the local minimum value of the input speech power spectrum is used as the stop band of the first comb filter, and the other frequency components are used as the pass band to generate the first comb filter. Different from the voice enhancement device.

In FIG. 5, the input speech spectrum is input to the local minimum value calculation unit 401, and the local minimum value calculation unit 401 is input.
Is output to the first comb filter generation unit 108 together with the threshold value set by the first voice / non-voice identification unit 106,
The result of the first comb filter generation unit 108 is output to the comb filter correction unit 113.

The local minimum value calculation unit 401 determines the frequency component located at the local minimum value in a predetermined frequency region based on the input voice power spectrum, as the frequency components at the conversion points of the pass band and stop band of the first comb filter. And Specifically, in the predetermined frequency domain, the first comb filter is generated by the following procedure.

Speech spectrum S _f (k) divided for each frequency component (where k is a number identifying the frequency component)
For adjacent frequency components S _f (k−1) and S
_If both powers of _f (k + 1) are smaller than the above, the first comb filter is set as a stop band, and if the above conditions are not satisfied, the first comb filter is set as a pass band. For frequency regions other than the predetermined frequency region, the first comb filter is generated by the same means as in the first embodiment.

As described above, according to the voice emphasizing device of the present embodiment, a comb filter having a predetermined range from the minimum value of the power spectrum of the input signal as a stop band is created, and the input signal is output using this comb filter. By suppressing the noise, even if the level difference between the voice and the noise is small, the voice pitch harmonic structure can be extracted and restored to reduce the voice distortion.

Specifically, according to the voice emphasizing device of the present embodiment, the local minimum value is extracted in a predetermined frequency region (particularly, the low frequency band), and the frequency component of the local minimum value is extracted. By making the stop band of the first comb filter and the other frequency components the pass band, it is possible to more reliably extract and restore the voice pitch harmonic structure even in a low SN ratio environment where voice is easily buried in noise. Therefore, it is possible to reduce the voice distortion due to the lack of the voice pitch harmonic structure.

(Embodiment 4) FIG. 6 is a block diagram showing an example of the configuration of a voice emphasizing device according to Embodiment 4. In FIG. However, the same components as those in FIG. 1 are denoted by the same reference numerals as those in FIG. 1, and detailed description thereof will be omitted.

The speech emphasizing apparatus 500 of FIG. 6 comprises a noise base subtraction unit 501, subtracts the noise base from the power spectrum of the input signal, and estimates and estimates the speech pitch based on the autocorrelation function of the subtraction result. The point that the voice pitch harmonic structure included in the first comb filter is corrected by the voice pitch is different from the voice enhancing apparatus of FIG.

Specifically, the voice emphasizing device 500 shown in FIG.
1 adopts a method of subtracting a noise base from the power spectrum of the input speech to calculate an autocorrelation function, and calculating a pitch estimation value based on the autocorrelation function of the power spectrum of the input speech in the frequency domain. Different from the voice enhancement device.

In FIG. 6, the input speech power spectrum and the noise base estimation value estimated by the noise base estimation unit 105 are input to the noise base subtraction unit 501, and the noise base subtraction unit 501 extracts the noise base from the input speech power spectrum. The estimated value is subtracted and the result is input to pitch estimation section 111. Further, the signal output from the voiced / unvoiced discrimination unit 110 indicating whether or not to perform pitch estimation is also input to the pitch estimation unit 111. The pitch period estimated by the pitch estimation unit 111 is output to the pitch harmonic structure restoration unit 112.

Pitch estimating section 111 calculates an autocorrelation function using the result of subtracting the noise base from the input speech power spectrum, and sets the delay corresponding to the maximum value of the autocorrelation function as the pitch period.

Specifically, the noise base is subtracted from the input speech power spectrum using the following equation (11), and the autocorrelation function is calculated using equation (12).

[0135]

[Equation 11]

[0136]

[Equation 12] Here, K _M is the upper limit of the frequency. Let τ corresponding to the maximum value of the autocorrelation function calculated by equation (12) be the pitch period.

As described above, according to the voice emphasizing apparatus of the present embodiment, the noise base is subtracted from the power spectrum of the input signal, the voice pitch is estimated based on the autocorrelation function of the subtraction result, and the estimated voice pitch is obtained. By correcting the voice pitch harmonic structure included in the first comb filter, the pitch harmonic structure can be restored, and voice enhancement with less voice distortion can be performed.

Specifically, according to the speech emphasizing apparatus of this embodiment, the pitch estimating unit 111 calculates the autocorrelation function using the result of subtracting the noise base from the input speech power spectrum, and The pitch estimation error can be reduced, and the pitch harmonic structure can be more accurately restored.

(Fifth Embodiment) FIG. 7 is a block diagram showing an example of the configuration of a voice emphasizing device according to a fifth embodiment. However, the same components as those in FIG. 1 are denoted by the same reference numerals as those in FIG. 1, and detailed description thereof will be omitted.

The speech emphasizing apparatus 600 of FIG. 7 generates a pseudo peak having a predetermined power in the DC component as a result of subtracting the noise base from the power spectrum of the input signal, and calculates the DC component as an autocorrelation function of the generated spectrum. 1 is different in that the voice pitch is estimated based on the estimated voice pitch and the voice pitch of the comb filter is corrected with the estimated voice pitch.

Specifically, the voice emphasizing device 600 shown in FIG.
1 includes a DC component generator 601. When calculating an autocorrelation function, a pseudo power spectrum having appropriate energy is generated as a DC component, and the autocorrelation function is calculated based on the generated DC spectrum. Different from the device.

In FIG. 7, the DC component generator 601 generates a power spectrum having an appropriate energy for the DC component and inputs it to the pitch estimator 111. Further, a signal indicating whether or not pitch estimation is performed by the voiced / unvoiced discrimination unit 110 is also input to the pitch estimation unit 111. The pitch period estimated by the pitch estimation unit 111 is output to the pitch harmonic structure restoration unit 112.

Specifically, an autocorrelation function is calculated using an input spectrum in which a pseudo power spectrum having the same energy as the power spectrum of the first voice pitch harmonic is added as a DC component, and the pitch period is calculated based on the result. To estimate.

As described above, according to the voice emphasizing device of the present embodiment, a pseudo peak having a predetermined power in the DC component is generated as a result of subtracting the noise base from the power spectrum of the input signal, and the DC component is generated. By estimating the voice pitch based on the autocorrelation function of the spectrum, and restoring the pitch harmonic structure with the estimated voice pitch, even if a part of the pitch harmonic is buried in noise, the pitch information is obtained and the pitch tuning is performed. The wave structure can be restored, and the voice enhancement with less voice distortion can be performed.

Specifically, according to the voice emphasizing device of the present embodiment, even if the input voice signal does not include a DC component,
In the frequency domain, the DC component can be used as a base point for pitch harmonics. By generating a pseudo power spectrum having an appropriate energy for a DC component, one reference pitch harmonic is increased. If the autocorrelation function is calculated using this, pitch estimation can be performed more accurately. Particularly, it is very effective for reducing the pitch estimation error when the energy of the first voice pitch harmonic is large and the energy of the higher order pitch harmonic is small, or when the noise level in the low frequency region is high.

The fifth embodiment can be combined with the fourth embodiment. That is, if the noise-based subtraction unit 501 described in the fourth embodiment is used in the voice enhancement device in FIG. 7, the effect of the fourth embodiment can be obtained.

(Sixth Embodiment) FIG. 8 is a block diagram showing an example of the configuration of a voice emphasizing device according to the sixth embodiment.
However, the same components as those in FIG. 1 or 4 are denoted by the same reference numerals as those in FIG.

The voice emphasizing device 700 of FIG.
Non-voice identification unit 701 and third comb filter generation unit 702
And a noise characteristic estimation unit 703, estimate the noise variance value of the input signal, calculate the moving average of the number of frequency components in the pass band of the third comb filter, and if the value is large, the noise variance It is judged that the value is large, and conversely it is judged that the noise variance value is small, and based on the result, the voice / non-voice discrimination threshold when the second comb filter is generated is set. Different from the device.

Specifically, the voice emphasizing device 700 shown in FIG.
Generates a third comb filter that estimates the characteristics of noise in the frequency domain, adds the number of frequency components in the passband of the third comb filter in the noise frame, and then outputs the second voice / The difference from the voice emphasizing device in FIG. 1 is that the non-voice discrimination threshold is determined.

The third speech / non-speech discrimination section 701, when the difference between the speech spectrum signal output from the frequency division section 104 and the noise base value output from the noise base estimation section 105 is equal to or larger than a predetermined threshold value, It is determined to be the pass band, and otherwise it is determined to be the stop band. The determination result is output to the third comb filter generation unit 702.

The third comb filter generation unit 702 generates a pass band / stop band of the comb filter based on the voice / non-voice discrimination result output from the third voice / non-voice discrimination unit 701, and outputs the result as noise. It is output to the characteristic estimation unit 703. The noise characteristic estimation unit 703 uses the speech / noise frame detection unit 3
In the noise frame detected from 01, the number of frequency components in the pass band of the third comb filter is added, the average value is calculated over a predetermined number of frames, and the result is calculated by the second voice / non-voice discriminating unit 107. Output to. In particular,
The noise characteristic is estimated using the following equation (13).

[0152]

[Equation 13] Where COMB_var (k) is the third comb filter, NS_var (n)
Is a noise characteristic estimation result, and α _V is a moving average coefficient.

Then, the second voice / non-voice discrimination threshold is NS_v
ar (n) is adaptively controlled, and if the value of NS_var (n) is large, it is determined that the variance is large as a noise characteristic,
The second voice / non-voice discrimination threshold is set high, and conversely, if the value is small, it is determined that the variance of the noise characteristics is small, and the second voice / non-voice discrimination threshold is set low.

As described above, according to the voice emphasizing device of the present embodiment, the distribution of the noise level of the input signal is detected, and the reference for generating the comb filter is determined from the voice spectrum based on this distribution to determine the noise level. Noise can be suppressed according to the type, and speech enhancement with less speech distortion can be performed.

Specifically, according to the voice emphasizing device of the present embodiment, the third comb filter for estimating the noise characteristic is provided to indirectly estimate the noise variance value by a simple calculation. By setting the second voice / non-voice discrimination threshold based on the result, it is possible to reduce the mixing of false pitch harmonics due to noise with large variance when generating the second comb filter. it can. Further, in the case of noise having a small variance value, more voice pitch harmonic information can be left.

(Seventh Embodiment) FIG. 9 is a block diagram showing an example of the configuration of a voice emphasizing device according to a seventh embodiment. However, the configuration common to FIG. 1 and FIG.
4, and the same numbers as in FIG. 4 are given, and detailed description is omitted.

The speech emphasizing apparatus 800 of FIG. 9 is equipped with a frequency domain selection unit 801, and when the speech pitch harmonic structure is restored, a predetermined frequency domain of the second comb filter is obtained based on the result of the third comb filter. The point that all the frequency components in are converted to the stop band is different from the speech enhancement apparatus of FIG.

In FIG. 9, noise characteristic estimating section 703
Estimates the noise characteristic based on the result output from the third comb filter generation unit 702, and outputs the result to the frequency domain selection unit 801. Frequency domain selection unit 801
Determines the intermediate frequency region in which the second comb filter is in the stop band based on the noise characteristic estimation result, and outputs the result to the second comb filter generation unit 109.

Specifically, if the moving average value of the noise characteristic calculated by the noise characteristic estimating unit 703 exceeds a certain threshold, it is determined that the noise has a large variance value, and the intermediate frequency region of the second comb filter, for example, 1 kHz. All frequency components between 2 kHz are converted to the stop band.

As described above, according to the voice emphasizing device of the present embodiment, the frequency component is selected based on the estimation result of the noise characteristic, and the second comb filter converts all the selected frequency region into the stop band. By doing so, false pitch harmonics generated by noise having a large variance value are reduced, and the pitch harmonics described in the first embodiment are based on the pitch harmonics in the low frequency region in which false pitch harmonics are less likely to occur. If the structure is repaired, the pitch harmonic structure can be accurately repaired.

The seventh embodiment can be combined with the sixth embodiment. That is, if the noise characteristic estimation unit 703 described in the sixth embodiment is used in the speech enhancement apparatus of FIG. 9, the effect of the sixth embodiment can be obtained.

(Embodiment 8) FIG. 10 shows Embodiment 8.
It is a block diagram showing an example of composition of a voice emphasis device concerning. However, the same components as those in FIG. 1 are denoted by the same reference numerals as those in FIG. 1, and detailed description thereof will be omitted.

The speech emphasizing apparatus 900 of FIG. 10 comprises an SNR estimating section 901, and in the speech separation coefficient calculating means,
The difference from the speech enhancement apparatus of FIG. 1 is that the amount of noise attenuation is adjusted according to the magnitude of the estimated SNR value.

In FIG. 10, SNR estimating section 901 is input speech power spectrum and noise base estimating section 105.
The ratio of the voice level to the noise level is calculated based on the noise-based estimated value output from the voice separation coefficient calculation unit 114, and the result is output to the voice separation coefficient calculation unit 114. Speech separation coefficient calculation unit 114
Calculates the amount of noise attenuation for each frequency component according to the magnitude of the SNR estimated value, and multiplies the input speech spectrum by the multiplication unit 115. Specifically, the following equations (14) to (16)
To calculate the SNR.

[0165]

[Equation 14]

[0166]

[Equation 15]

[0167]

[Equation 16] Here, S _p (n) is the moving average value of the voice level, N
_s (n) is a moving average value of noise, and α _S is a moving average coefficient.

According to the SNR value, the voice separation coefficient (amount of noise attenuation) in the pass band and stop band of the modified comb filter is calculated using the following equation (17).

[0169]

[Equation 17] Here, γ (n) is a coefficient indicating the amount by which the noise base is subtracted, and gc (n) is a noise attenuation degree coefficient. Γ (n) and gc are set so that they can handle environments with different SN ratios.
The value of (n) can be automatically adjusted from the estimated value of SNR. For example, the magnitude of γ (n) is directly proportional to the value of SNR (n), and the magnitude of gc (n) is automatically adjusted to be inversely proportional to the value of SNR (n).

As described above, according to the speech separation apparatus of the present embodiment, the amount of subtraction of the noise base from the input speech power spectrum and the degree of noise attenuation in the pass band and stop band of the modified comb filter are used as the SNR estimated value. By adjusting in accordance with the magnitude of, the appropriate noise attenuation can be performed even under the environment of different SNR, and the voice enhancement with less voice distortion and residual noise can be realized.

(Ninth Embodiment) FIG. 11 shows a ninth embodiment.
It is a block diagram showing an example of composition of a voice emphasis device concerning. However, components common to those in FIGS. 1 and 10 are designated by the same reference numerals as those in FIGS. 1 and 10, and detailed description thereof is omitted.

The speech emphasizing apparatus 1000 of FIG. 11 comprises a noise base subtraction unit 1001 and a weighting coefficient calculation unit 1002, calculates the level of the speech component from the moving average value of the power spectrum of the input signal, and calculates the noise component 1 is that a level is calculated from a value obtained by multiplying a noise-based estimated value by a weighting coefficient for each frequency component, and a signal-to-noise ratio is calculated from a ratio between the level of the voice component and the level of the noise component. Different from the highlighter.

In FIG. 11, the noise base estimation unit 10
5 estimates the noise base, and outputs the result to the noise base subtraction unit 1001 and the weight coefficient calculation unit 1002. The noise-based subtraction unit 1001 calculates the moving average value of the input speech power spectrum, subtracts the noise-based estimated value from the moving average value, and outputs it to the SNR estimation unit 901. Weighting coefficient calculation section 1002 calculates a weighting coefficient for each frequency component of the noise-based estimated value, and outputs the result to SNR estimation section 901. The SNR estimation unit 901 calculates the ratio between the voice level and the noise level, and the result is used as the voice separation coefficient calculation unit 114.
Output to. Specifically, the following equations (18) to (2
0) is used to calculate the SNR.

[0174]

[Equation 18]

[0175]

[Formula 19]

[0176]

[Equation 20] Where β is a coefficient indicating the amount by which the noise base is subtracted,
δ (k) is a weighting coefficient. The weighting factor δ (k) is set by utilizing the characteristics of voice. For example, since the energy of the voice spectrum in the intermediate frequency region is small, but the influence on the intelligibility of the voice is great, if the noise level is calculated by increasing the value of the weighting coefficient for the noise in the intermediate frequency region, different noises To provide appropriate damping.

As described above, according to the voice emphasizing apparatus of the present embodiment, the noise base is calculated by subtracting the noise base from the moving average value of the input voice power spectrum, thereby reducing the influence of noise and reducing the noise level. An accurate voice level can be calculated even in an SNR environment. In addition, by multiplying each frequency component of the noise-based estimated value by a weighting coefficient for calculation, it is possible to appropriately attenuate different noises and reduce voice distortion.

Note that the ninth embodiment can be combined with the eighth embodiment. That is, if the SNR estimation unit 901 described in the eighth embodiment is used in the speech enhancement apparatus in FIG. 11, the effect of the eighth embodiment can be obtained.

(Tenth Embodiment) FIG. 12 is a block diagram showing an example of the configuration of a voice emphasizing device according to the tenth embodiment. However, the same components as those in FIG. 1, FIG. 10 and FIG.
Detailed explanation is omitted.

The speech enhancement apparatus 1100 shown in FIG.
SNR fluctuation suppressing section 1101 for suppressing fluctuation of
The NR variation suppressing unit 1101 is different from the speech enhancement apparatus of FIG. 1 in that the NR variation suppressing unit 1101 suppresses the SNR variation based on the result of the SNR estimated value and the long-term moving average value of the SNR estimated value.

In FIG. 12, the SNR estimation unit 901 calculates the ratio between the voice level and the noise level, and outputs the result as the SNR.
Output to the fluctuation suppressing unit 1101. SNR fluctuation suppressing section 11
01 calculates the long-term moving average value of SNR based on the SNR estimated value, calculates the deviation between the result and the SNR estimated value, and
The long-term moving average value of the NR estimated value and a part of the deviation are added and used as the SNR fluctuation suppression result. Then, the SNR estimated value in which the fluctuation is suppressed is output to the voice separation coefficient calculation unit 114.

Specifically, S is calculated using the following equation (21).
Calculate the long-term moving average value of NR and use equation (22) to calculate S
An estimated value of SNR that suppresses fluctuations in NR is calculated.

[0183]

[Equation 21]

[0184]

[Equation 22] Here, α _r is a moving average coefficient, and μ is a coefficient that determines the magnitude of deviation to be added.

As described above, according to the speech emphasizing apparatus of the present embodiment, the deviation between the SNR estimated value and the long-term moving average value of the SNR estimated value is calculated, and the long-term moving average value of the SNR estimated value and one of the deviations are calculated. By adding the parts and using them as the SNR estimation value, the SNR variation is effectively suppressed, and the SN is stable.
The level of noise attenuation can be adjusted according to the magnitude of R.

Note that the tenth embodiment can be combined with the eighth or ninth embodiment. That is, the SN described in the eighth embodiment in the voice emphasizing device of FIG.
If the R estimation unit 901 is used, the effect of the eighth embodiment can also be obtained, and SN can be obtained by using the noise-based subtraction and weighting factor calculation means described in the ninth embodiment in the speech enhancement apparatus of FIG.
If R is estimated, the effect of the ninth embodiment can be obtained.

(Embodiment 11) FIG. 13 is a block diagram showing an example of the configuration of a speech emphasizing device according to Embodiment 11. However, the same components as those in FIG. 1 are denoted by the same reference numerals as those in FIG. 1, and detailed description thereof will be omitted.

The speech emphasizing apparatus 1200 of FIG. 13 is different from the speech emphasizing apparatus of FIG. 1 in that it is equipped with a noise base updating unit 1201 having a high updating speed, and noise-based tracking can be performed even in a voice section.

In FIG. 13, the noise base updating unit 12
01 estimates a noise base based on the input speech power spectrum using a moving average coefficient with a high update speed, and outputs the result to the noise base estimation unit 105. The noise base estimation unit 105 estimates the noise base using the moving average coefficient having a slow update rate, and outputs the result to the first voice / non-voice identification unit 106 and the second voice / non-voice identification unit 107.

Specifically, the following equations (23) and (2)
4) is used to estimate a noise base with a fast update speed and a noise base with a slow update speed.

[0191]

[Equation 23]

[0192]

[Equation 24] Here, α _f and α _s are a fast update coefficient and a slow update coefficient, respectively, and Θ _fast is a threshold value for distinguishing between speech and noise.

As described above, according to the speech emphasizing apparatus of the present embodiment, noise-based estimation is performed using the moving average coefficient with a high update rate, so that a rapid change in the noise level is tracked even in the speech section. can do. In addition, the noise-based update with a slow update speed can be accurately estimated by performing the noise-based update with a fast update speed, and the noise-based update can be stopped due to a sudden change in the noise level. Can be prevented.

(Embodiment 12) FIG. 14 is a block diagram showing the structure of a noise separating apparatus according to Embodiment 12 of the present invention. The noise separation device 1300 of the present embodiment separates and extracts a noise signal from a speech signal containing noise.

In FIG. 14, the noise separation device 1300
Is a time division unit 101, a windowing unit 102, an FFT unit 103, a frequency division unit 104, a noise base estimation unit 105, a voice / non-voice identification unit 1301, a noise comb filter generation unit 1302, and a real number. An imaginary number separating unit 1303,
The noise separation coefficient calculation unit 1304, the multiplication unit 1305, the noise frequency synthesis unit 1306, and the IFFT unit 1307 are mainly included.

Here, the same components as those in FIG. 1 are designated by the same reference numerals as those in FIG. 1, and detailed description thereof will be omitted.

The speech / non-speech discriminating section 1301 discriminates the speech if the difference between the speech spectrum signal outputted from the frequency dividing section 104 and the noise base estimation value outputted from the noise base estimating section 105 is equal to or larger than a predetermined threshold value. It is determined to be a voiced portion including a component, and in other cases, it is determined to be a silent portion including only noise that does not include a voice component, and the result is output to the noise comb filter generation unit 1302. The noise comb filter generation unit 1302 generates a noise separation comb filter based on the result of the speech / non-speech discrimination unit 1301 and outputs this comb filter to the real number imaginary number separation unit 1303.

Specifically, the noise comb filter generator 13
02 sets a low threshold value for voice / non-voice discrimination so as to suppress voice information, and generates a noise separation comb filter using the following equation (25).

[0199]

[Equation 25] Where Θ _nos is a noise separation threshold.

The real imaginary number separation unit 1303 separates the real number part and the imaginary number part of the input speech spectrum and outputs the result to the noise separation coefficient calculation unit 1304. Noise separation coefficient calculation unit 1
304 calculates different separation coefficients for each frequency component with respect to the pass band and the stop band of the noise separation comb filter, and outputs the result to the multiplication unit 1305.

Specifically, the following equations (26) and (2)
7) is used, the noise separation coefficient is set to 1 in the stop band of the noise separation comb filter, and the noise separation coefficient is set separately for the real part and the imaginary part of the input speech spectrum in the pass band of the noise separation comb filter. Multiply a random number by a noise-based estimate.

[0202]

[Equation 26]

[0203]

[Equation 27] Here, rd _re (i) is a random function used for the real part and is composed of uniformly distributed random numbers, and rd _im (i) is
It is a random function used for the imaginary part and is composed of uniformly distributed random numbers.

The multiplying unit 1305 calculates the noise separation coefficient calculating unit 13 based on the speech spectrum output from the frequency dividing unit 104.
The separation coefficient output from 04 is multiplied for each frequency component. Then, the noise spectrum obtained as a result of the multiplication is output to the noise frequency synthesis unit 1306.

The noise frequency synthesizer 1306 has a multiplier 13
The frequency component spectrum output from 05 is combined into a continuous noise spectrum in the frequency domain in a predetermined processing time unit and output to the IFFT unit 1307.

The IFFT unit 1307 adds IFFT (In) to the noise spectrum output from the noise frequency synthesis unit 1306.
verse Fast Fourier Transform) is performed to output a signal converted into a noise signal.

As described above, according to the noise separating apparatus of the present embodiment, the noise characteristic can be extracted to the maximum extent by generating the noise dedicated comb filter. In the stopband of the noise separation comb filter, the noise component is not attenuated, and in the passband of the noise separation comb filter,
By multiplying the real and imaginary parts of the input speech spectrum by separate random numbers and noise-based estimates,
The amplitude and phase of the real and imaginary parts of the noise component are all randomized, and good noise separation characteristics can be obtained.

(Thirteenth Embodiment) FIG. 15 is a block diagram showing an example of the configuration of a noise separation device according to the thirteenth embodiment. However, the same components as those in FIGS. 1 and 14 are denoted by the same reference numerals as those in FIGS. 1 and 14, and detailed description thereof is omitted.

The noise separation device of FIG. 15 comprises a noise component storage unit 1401, stores the spectral component of the input voice in the stop band of the noise separation comb filter in the memory, and passes the value through the noise separation comb filter. The points used for the area are
Different from FIG.

In FIG. 15, noise component storage unit 1401
Stores the input speech spectrum in the stop band of the noise comb filter output from the noise comb filter generation unit 1302, and outputs the noise component storage result to the noise separation coefficient calculation unit 1304 in the pass band of the noise comb filter.

Specifically, for example, a predetermined number of memories are prepared to sequentially store the input speech spectrum in the stop band of the noise comb filter from the low frequency region to the high frequency region, and the same low frequency band is used in the pass band of the noise comb filter. In order from the frequency domain to the high frequency domain, the latest stored data and the one having the closest frequency component are selected and used as the input speech spectrum in the pass band of the noise comb filter.

As described above, according to the noise separating apparatus of the present embodiment, the spectral component of the input voice in the stop band of the noise separating comb filter is stored in the memory, and the value is stored in the pass band of the noise separating comb filter. , It is possible to reconstruct pseudo noise having characteristics close to those of actual noise, and obtain good noise separation characteristics.

The thirteenth embodiment is the twelfth embodiment.
Can be combined with. That is, in the noise separation device of FIG. 15, the noise separation coefficient calculation unit 1304 according to the twelfth embodiment is used.
By using, the effect of the twelfth embodiment can be obtained.

The present invention is not limited to the above-described embodiments, and a plurality of embodiments can be combined or variously modified and carried out. For example, although cases have been described with the above embodiments where a speech enhancement apparatus or noise suppression apparatus is used, the present invention is not limited to this, and this speech enhancement method or noise suppression method can also be implemented as software. .

For example, a program for executing the above speech enhancement method or noise suppression method is stored in advance in a ROM (Read Only Me).
mory) and store the program in the CPU (Cent
ralProcessor Unit).

Further, a program for executing the above speech enhancement method or noise suppression method is stored in a computer-readable storage medium, and the program stored in the storage medium is recorded in a RAM (Random Access Memory) of the computer, The computer may be operated according to the program.

It is also possible to store a program for performing the above speech enhancement or noise suppression in a server, transfer the program stored in the server to a client, and execute the program on the client. Even in such a case, the same operation and effect as those of the above-described embodiment are exhibited.

[0218] Further, the voice emphasizing device or the noise suppressing device according to any one of the above-mentioned embodiments can be installed in a radio communication device, a communication terminal, a base station device, or the like. As a result, the voice during communication can be emphasized or noise can be extracted.

[0219]

As described above, according to the voice emphasizing device and the voice emphasizing method of the present invention, a comb in which more noise peaks are removed than the comb filter used for the voice suppression based on the frequency-divided spectrum of the voice signal. By generating a filter, using this comb filter to obtain pitch information of the voice signal, and supplementing the voice pitch of the comb filter,
The distortion of the voice is small and the noise can be sufficiently removed.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a voice emphasizing device according to a first embodiment of the present invention.

FIG. 2 is a diagram showing an example of a comb filter created by the voice enhancement device according to the above embodiment.

FIG. 3 is a diagram showing an example of restoration of a comb filter in the voice processing device according to the embodiment.

FIG. 4 is a block diagram showing a configuration of a voice emphasizing device according to a second embodiment of the present invention.

FIG. 5 is a block diagram showing a configuration of a voice emphasizing device according to a third embodiment of the present invention.

FIG. 6 is a block diagram showing a configuration of a voice emphasizing device according to a fourth embodiment of the present invention.

FIG. 7 is a block diagram showing a configuration of a voice emphasizing device according to a fifth embodiment of the present invention.

FIG. 8 is a block diagram showing a configuration of a voice emphasizing device according to a sixth embodiment of the present invention.

FIG. 9 is a block diagram showing a configuration of a voice emphasizing device according to a seventh embodiment of the present invention.

FIG. 10 is a block diagram showing a configuration of a voice emphasizing device according to an eighth embodiment of the present invention.

FIG. 11 is a block diagram showing a configuration of a voice emphasizing device according to a ninth embodiment of the present invention.

FIG. 12 is a block diagram showing a configuration of a voice emphasizing device according to a tenth embodiment of the present invention.

FIG. 13 is a block diagram showing a configuration of a voice emphasizing device according to an eleventh embodiment of the present invention.

FIG. 14 is a block diagram showing a configuration of a noise separation device according to a twelfth embodiment of the present invention.

FIG. 15 is a block diagram showing a configuration of a noise separation device according to a thirteenth embodiment of the present invention.

[Explanation of symbols]

103 FFT section 104 Frequency division unit 105 noise base estimation unit 106 First voice / non-voice discrimination unit 107 Second voice / non-voice discrimination section 108 First Comb Filter Generation Unit 109 Second comb filter generator 110 voiced / unvoiced discriminator 111 Pitch estimation unit 112 Pitch harmonic structure restoration section 113 Comfilter correction unit 114 Speech Separation Coefficient Calculation Unit 115, 1305 multiplication unit 116 Speech frequency synthesizer 117 IFFT section 301 voice / noise frame detector 401 Local minimum value calculation unit 501, 1001 Noise base subtraction unit 601 DC component generator 701 Third voice / non-voice discrimination unit 702 Third comb filter generation unit 703 Noise characteristic estimation unit 801 Frequency domain selector 901 SNR estimation unit 1002 Weighting factor calculation unit 1101 SNR fluctuation suppression unit 1201 Noise base update unit 1301 voice / non-voice discriminator 1302 Noise comb filter generation unit 1303 Real number imaginary number separation part 1304 Noise Separation Coefficient Calculation Unit 1306 Noise frequency synthesizer 1401 Noise component storage unit

Claims (24)

[Claims] 1. A frequency division means for outputting a frequency division spectrum obtained by dividing the spectrum of an input signal by a predetermined frequency unit, and a first comb filter for attenuating a signal in a frequency domain determined to be silent based on the frequency division spectrum. A first comb filter creating means for creating, a second comb filter creating means for creating a second comb filter which is a filter in which more noise peaks are removed than the first comb filter based on the frequency division spectrum; A comb filter correcting unit that corrects a voice pitch harmonic structure included in the first comb filter with a voice pitch estimated from a comb filter and the frequency-divided spectrum, and a first comb filter corrected by the comb filter correcting unit. Suppressing means for suppressing the noise of the frequency division spectrum using the And a voice frequency synthesizing unit for synthesizing the frequency-divided spectrum into a continuous spectrum signal in the frequency domain. 2. A voice / noise frame detection means for determining whether or not a voice component is included in a voice spectrum from the first comb filter and the second comb filter, wherein the comb filter correction means comprises the voice / noise. The speech enhancement apparatus according to claim 1, wherein, when the result of the frame detection means does not include a speech component, the first comb filter is modified to attenuate the signal at each frequency component. 3. The speech / noise frame detection means sets a ratio of a sum of a power spectrum of an input signal in a pass band of the first comb filter and a sum of power spectra of an input signal in a stop band of the first comb filter to a first ratio. As a result, the ratio of the sum of the power spectrum of the input signal in the pass band of the second comb filter and the sum of the power spectrum of the input signal in the stop band of the second comb filter is set as the second result. When the addition result is larger than a predetermined threshold value, the addition result is used, and when the addition result is less than or equal to the predetermined threshold value, it is determined whether or not the voice spectrum includes a voice using the second result. The voice enhancement device according to claim 2. 4. The first comb filter generation means creates a first comb filter having a stop band within a predetermined range from the minimum value of the power spectrum of the input signal. The voice enhancement device according to any one of claims. 5. A pitch estimating means for subtracting a noise base from a power spectrum of an input signal and estimating a voice pitch by using an autocorrelation function of the subtraction result, wherein a comb filter correcting means estimates the speech pitch in the pitch estimating means. The speech enhancement apparatus according to any one of claims 1 to 4, wherein the speech pitch harmonic structure included in the first comb filter is modified with the defined speech pitch. 6. A direct current component generating means for generating a pseudo peak having a predetermined power in a direct current component as a result of subtracting a noise base from a power spectrum of an input signal, wherein the pitch estimating means comprises the direct current component generating means. The speech enhancement apparatus according to claim 5, wherein the speech pitch is estimated from the power spectrum in which the pseudo peak is generated. 7. A second comb filter is provided, comprising noise characteristic estimating means for calculating a moving average of the number of frequency regions in which the power is equal to or more than a predetermined threshold, as a result of subtracting the noise base from the power spectrum of the input signal. 7. The voice emphasizing device according to claim 1, wherein the means creates a second comb filter based on a result of determining whether or not the input signal includes a voice from the moving average. 8. A second comb filter is provided, comprising noise characteristic estimating means for calculating a moving average of the number of frequency regions in which the power is equal to or more than a predetermined threshold, as a result of subtracting the noise base from the power spectrum of the input signal. The means for creating a second comb filter having a predetermined frequency range as a stop band when the moving average calculated by the noise characteristic estimating means is equal to or smaller than a predetermined value. The voice enhancement device according to any one of 1. 9. An SNR estimating means for calculating a signal-to-noise ratio from a power spectrum of an input signal and a noise base, wherein the suppressing means determines a noise suppression amount of the frequency division spectrum from the signal-to-noise ratio. The voice enhancement device according to any one of claims 1 to 8. 10. The SNR estimating means calculates a level of a voice component from a moving average value of a power spectrum of an input signal,
A noise component level is calculated from a value obtained by multiplying a noise-based estimated value by a weighting coefficient for each frequency component, and a signal-to-noise ratio is calculated from a ratio between the voice component level and the noise component level. The voice enhancement device according to claim 9. 11. A fluctuation suppressing means for calculating a deviation between a signal-to-noise ratio and a moving average value of the signal-to-noise ratio, and updating the moving average value of the signal-to-noise ratio using the deviation, The speech enhancement apparatus according to claim 9 or 10, wherein the suppression unit determines a noise suppression amount of the frequency division spectrum from the moving average value of the signal-to-noise ratio updated by the fluctuation suppression unit. . 12. A condition for updating a first moving average value with a second moving average value having a faster updating speed than the first moving average value, by calculating two noise-based moving average values having different updating speeds in a predetermined time unit. The speech enhancement apparatus according to any one of claims 1 to 11, further comprising a noise base updating unit configured to change the first moving average value and output the first moving average value as a noise base estimation value. 13. A wireless communication device comprising the noise suppressing device according to claim 1. Description: 14. A frequency division means for outputting a frequency division spectrum obtained by dividing the spectrum of an input signal in a predetermined frequency unit, and a noise separation comb having a passband for a signal in a frequency domain determined to be silent based on the frequency division spectrum. A noise separation comb filter creation means for creating a filter, a suppression means for separating a noise component of the frequency division spectrum using the noise separation comb filter, and a spectrum in which the frequency division spectrum obtained by separating the noise component is continuous in the frequency domain. And a voice frequency synthesizing means for synthesizing into a signal. 15. The noise separating means multiplies the real part and the imaginary part of the input speech spectrum by different random numbers and a noise-based estimation value in the pass band of the noise separating comb filter. 14. The noise suppression device according to 14. 16. A noise component storage means for storing a spectral component of an input voice in a stop band of a voice separation comb filter, wherein the noise separation means stores the spectral component stored in the memory in a pass band of the noise separation comb filter. 16. The noise suppression device according to claim 14 or 15, which is used for. 17. A wireless communication device comprising the noise suppressing device according to claim 14. 18. A sound source comprising the speech enhancement apparatus according to any one of claims 1 to 12 and the noise suppression apparatus according to any one of claims 14 to 16. Separation device. 19. A frequency division process for outputting a frequency division spectrum obtained by dividing the spectrum of an input signal in a predetermined frequency unit, and a first comb filter for attenuating a signal in a frequency domain determined to be silent based on the frequency division spectrum. A first comb filter creation step to create, a second comb filter creation step to create a second comb filter which is a filter in which more noise peaks are removed than the first comb filter based on the frequency division spectrum, and the second A comb filter correction step for correcting a voice pitch harmonic structure included in the first comb filter with a voice pitch estimated from a comb filter and the frequency division spectrum, and a first comb filter corrected in the comb filter correction step. The suppression process of suppressing the noise of the frequency division spectrum by using A voice frequency synthesizing step of synthesizing the compressed frequency-divided spectrum into a continuous spectrum signal in the frequency domain. 20. A frequency division step of outputting a frequency division spectrum obtained by dividing the spectrum of an input signal in a predetermined frequency unit, and a noise separation comb having a passband of a signal in a frequency domain determined to be silent based on the frequency division spectrum. A noise separation comb filter creation process for creating a filter, a suppression process for separating a noise component of the frequency division spectrum using the noise separation comb filter, and a spectrum in which the frequency division spectrum obtained by separating the noise component is continuous in the frequency domain. And a voice frequency synthesizing step of synthesizing into a signal. 21. A frequency division step of outputting a frequency division spectrum obtained by dividing the spectrum of an input signal in a predetermined frequency unit, and a first comb filter for attenuating a signal in a frequency domain determined to be silent based on the frequency division spectrum. A first comb filter creating step, a second comb filter creating step that creates a second comb filter that is a filter with more noise peaks removed than the first comb filter based on the frequency division spectrum, and the second A comb filter correcting step of correcting a voice pitch harmonic structure included in the first comb filter with a voice pitch estimated from a comb filter and the frequency division spectrum; and a first comb filter corrected by the comb filter correcting means. Suppressor for suppressing the noise of the frequency division spectrum using A speech enhancement program characterized by causing a computer to execute a step and a speech frequency synthesizing step of synthesizing the frequency-divided spectrum in which noise is suppressed into a continuous spectrum signal in the frequency domain. 22. A frequency division step of outputting a frequency division spectrum obtained by dividing the spectrum of an input signal in a predetermined frequency unit, and a noise separation comb having a passband of a signal in a frequency domain determined to be silent based on the frequency division spectrum. A noise separation comb filter creation step of creating a filter, a suppression step of separating a noise component of the frequency division spectrum using the noise separation comb filter, and a spectrum in which the frequency division spectrum obtained by separating the noise component is continuous in the frequency domain. A noise separation program characterized by causing a computer to execute a voice frequency synthesizing step of synthesizing into a signal. 23. A frequency division step of outputting a frequency division spectrum obtained by dividing the spectrum of an input signal by a predetermined frequency unit, and a first comb filter for attenuating a signal in a frequency domain determined to be silent based on the frequency division spectrum. A first comb filter creating step, a second comb filter creating step that creates a second comb filter that is a filter with more noise peaks removed than the first comb filter based on the frequency division spectrum, and the second A comb filter correction step of correcting a voice pitch harmonic structure included in the first comb filter with a voice pitch estimated from a comb filter and the frequency division spectrum, and a first comb filter corrected in the comb filter correction step. Suppressing to suppress the noise of the frequency division spectrum using Storing a voice emphasizing program characterized by causing a computer to execute a pressure step and a voice frequency synthesizing step of synthesizing the frequency-divided spectrum in which noise is suppressed into a continuous spectrum signal in the frequency domain, A server device which outputs the voice enhancement program. 24. A frequency division step of outputting a frequency division spectrum obtained by dividing the spectrum of an input signal in a predetermined frequency unit, and a noise separation comb having a passband as a frequency domain signal determined to be silent based on the frequency division spectrum. A noise separation comb filter creation step of creating a filter, a suppression step of separating a noise component of the frequency division spectrum using the noise separation comb filter, and a spectrum in which the frequency division spectrum obtained by separating the noise component is continuous in the frequency domain. A server device, which stores a noise separation program characterized by causing a computer to execute a voice frequency synthesis step of combining with a signal, and outputs the noise separation program in response to a request.