JP2000047697A - Noise canceler - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a noise canceler which satisfies simultaneously two conditions being contrary each other such that attenuation quantity of a background noise is fully secured and voice quality is kept, and which has superior speech quality. SOLUTION: A voice content discriminating section 9 monitors always variation of power of an input signal before analyzing a time frequency, a signal to noise ratio of an input signal is estimated from variation speed of power and converted into a scale of voice content. A attenuation quantity calculating section 15 monitors variation of power of all frequency region data, and calculates attenuation quantity using voice content as an aid of judgment of a voice component and a noise component. An attenuator 5 attenuates a noise component of frequency region data time-frequency-analyzed by FFT 4 in accordance with calculated attenuation quantity. Frequency region data of which only a noise component is attenuated is inversely converted into real time region data, made to continuous data by a window function overlap 7, and outputted from an output terminal 8.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a noise canceller, and more particularly to a noise canceller (background noise canceling device) used in a device such as a stationary telephone, a portable telephone, and a voice conference device that handles an audio signal such as voice as an input signal.
About.

[0002]

2. Description of the Related Art Conventionally, in this type of noise canceller, generally, a fast Fourier transform (F
FT: Fast Fourier Transform
m) Performs time-frequency analysis by performing signal domain transformation from the time domain to the frequency domain by orthogonal transformation as in m), and monitors power fluctuations for each band in the frequency domain. The band whose power has been increased is regarded as an audio component and passed without being attenuated.

On the other hand, a band in which no power fluctuation is observed or a band in which the power is weak is regarded as a noise component and attenuated, and an inverse fast Fourier transform (IFFT: Inverse Fast) is performed.
The function of removing only the noise component is realized by means such as performing an inverse transform such as Fourier Transform to restore the time domain signal.

[0004] The principle of the background noise attenuation of the noise canceller will be described. FIG. 10 is an example of a real-time signal in which a noise component is superimposed on speech. When monitoring power fluctuations, the difference between speech and noise can be identified only in sections.

FIG. 11 shows an example of a waveform obtained by converting an audio signal into a frequency domain. Generally, an audio signal is composed of a combination of a plurality of frequencies. FIG. 12 is an example of a waveform obtained by converting a background noise signal into a frequency domain.

[0006] Background noise has power over almost all frequency components. An example of a real-time domain waveform of an audio signal on which a noise component is superimposed is shown in FIG. The background noise component power is generally smaller than the voice component power, and the power fluctuation with time is small and gradual.

On the other hand, the sound component power tends to be larger than the noise power, and the power fluctuates greatly with time. Therefore, if the power fluctuation is monitored in the frequency domain, the noise component can be identified even in the voice section in the real time domain, and only the noise component can be attenuated.

However, in recent years, it has been required to more accurately separate noise components and voice components to increase the amount of noise component attenuation, thereby improving the communication quality. In order to respond to this request, for example, Japanese Unexamined Patent Publication No. 9-171397 discloses that in the frequency domain, a peak band having the maximum power is searched from all the bands, and a band corresponding to a multiple frequency thereof is determined as a voice component. It has been proposed to improve the communication quality by using a technique that does not attenuate.

[0009]

In the above-mentioned conventional noise canceller, in the above-mentioned publication, it is proposed to improve the speech quality by a technique of not attenuating. However, in this technique, speech contains a plurality of frequency components. It is assumed that they are in a multiple relationship specific to the audio signal. Therefore, when a noise canceller is positioned as preprocessing of the voice recognition device, the above technique is an effective means.

[0010] However, the above technique cannot exert an effect when a signal other than voice is included in the input signal. Therefore, the present invention cannot be applied to a case where an input signal of a stationary telephone, a mobile telephone, a voice conference device, or the like also considers a component other than voice as a signal, and another solution must be provided.

In a noise canceller of the type in which noise components are attenuated or subtracted in the frequency domain, when the power fluctuation is small, the speech quality largely depends on the criteria for determining the voice components and the noise components. ing. In other words, there is a problem that if the determination is wrong, narrow-band (tone) noise remains, or a signal having only weak power such as a harmonic component of voice is lost.

In particular, with respect to the problem of residual narrow-band noise, the larger the maximum value of the noise attenuation, the greater the relative power between the remaining narrow-band noise and the neighboring maximum-attenuated band. However, there is a contradiction that the speech quality in terms of audibility is degraded despite the large total noise attenuation in the entire region. Under such a background, it has been a challenge to more accurately separate noise components and voice components to increase the amount of attenuation of the noise components and to improve the communication quality.

Accordingly, an object of the present invention is to solve the above-mentioned problems, to simultaneously satisfy two contradictory conditions, that is, to ensure sufficient attenuation of background noise and to maintain speech quality, and to achieve speech communication. An object of the present invention is to provide a high-quality noise canceller.

[0014]

A noise canceller according to the present invention is a noise canceller for attenuating a noise component by attenuating the voice component in a frequency domain when a voice including the noise component is input, wherein the signal in the frequency domain is attenuated. First calculating means for calculating a noise-to-noise ratio, second calculating means for calculating a signal-to-noise ratio in a real time domain, and the frequency domain based on the calculation results of the first and second calculating means. And an attenuation amount calculating means for calculating the amount of attenuation at the time.

Another noise canceller according to the present invention is a noise canceller which attenuates a noise component by attenuating the voice in a frequency domain when a voice including the noise component is input, wherein the amount of attenuation in the frequency domain is set in advance. Setting means for setting to no attenuation, a maximum value, or an intermediate value between them based on the converted characteristic, and means for dynamically changing the conversion characteristic used in the setting means. I have.

Another noise canceller according to the present invention is a noise canceller for attenuating a noise component by attenuating the voice in a frequency domain when a voice including the noise component is input, wherein the signal-to-noise ratio in the frequency domain is reduced. First calculating means for calculating, second calculating means for calculating a signal-to-noise ratio in a real-time domain, and attenuation in the frequency domain based on the calculation results of the first and second calculating means. Setting means for setting any one of a no attenuation value, a maximum value, and an intermediate value thereof based on a preset conversion characteristic;
Means for converting the output of the calculation means into the probability of containing the voice component, and attenuation amount changing means for dynamically changing the conversion characteristic used by the setting means, the probability of containing the voice component Is used as a reference value of the attenuation amount changing means.

That is, the noise canceler of the present invention
As a reference parameter to a part of the attenuation calculation for outputting the noise attenuation for each band for performing the attenuation control of the noise component calculated in the frequency domain, a voice content determination unit for performing the calculation in the real time domain is provided. I have.

This voice content determination unit constantly monitors the power fluctuation of the input signal before performing the time-frequency analysis, and estimates the signal-to-noise ratio of the input signal from the power fluctuation speed of the real-time domain signal. The signal-to-noise ratio is logarithmized according to the auditory characteristics, and is made to correspond linearly to the signal-to-noise ratio on the audibility.

From the logarithmic value, the probability that the input signal contains a voice component is calculated as probability, converted into a scale of voice content, and used as an output signal of the voice content determination unit. The attenuation control unit at the next stage operates by using the voice content and the frequency domain signal as input, and using them as factors for determining the amount of attenuation for each band.

If it is difficult to determine only by monitoring the power fluctuation in the frequency domain, and it is difficult to determine the weak power fluctuation in a certain band, it is used to assist in determining whether the voice content is regarded as a voice component or a noise component. The noise component can be attenuated without damaging the harmonic components included in the audio component.

That is, when the signal-to-noise ratio of the real-time signal is large, there is a high probability that many voice components are contained, so that it is possible to judge a slight power fluctuation as a double voice of voice. Become. On the other hand, when the signal-to-noise ratio is small, the probability that a speech component is included is low, and it is possible to determine a weak power fluctuation as a noise component.

Therefore, the noise component can be attenuated more than when the noise attenuation control is performed based only on the power fluctuation of the frequency domain signal, and the sound components including the harmonic components are not damaged. Can be

In this means, the voice content is calculated based on the monitoring of the power fluctuation in the real time domain.
Even for non-noise signals other than speech, the same effect as speech is exhibited.

[0024]

Next, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a noise canceller according to one embodiment of the present invention. In the figure, a noise canceller 1 has an input terminal 2 and a window function 3
And FFT (Fast Fourier Transf)
orm: fast Fourier transform) 4, attenuator 5, and IFF
T (Inverse Fast Fourier Tr
ansform: inverse fast Fourier transform) 6, window function overlap 7, output terminal 8, and audio content determination unit 9
And an attenuation calculator 15. Hereinafter, non-noise signals other than voice are also representatively referred to as voice or voice signals.

A signal (data) input from the input terminal 2 of the noise canceller 1 is supplied to a window function 3, and data of several points of the FFT 4 are cut out from a signal sequence continuously input.

The input of the FFT 4 is real-time domain data arranged in a time-series order. The input data is subjected to time-frequency analysis processing by the FFT 4 and converted into frequency domain data of several FFT point bands. In order to perform an operation on the frequency domain data, the attenuation calculator 15 and the attenuator 5 are provided for several bands.

The noise component of the frequency domain data is attenuated by the attenuator 5 in accordance with the attenuation calculated by the attenuation calculator 15, the voice component is attenuated, and only the noise component is attenuated. The frequency domain data is inversely transformed into real time domain data. Further, the cut-out data is connected to continuous data by the window function overlap 7 and output from the output terminal 8.

The voice content determining section 9 is a signal power monitor 1
0, stationary power monitor 11, and signal to noise ratio calculator 1
2, a logarithmic calculator 13, and a voice content converter 14, which constantly monitor the power fluctuation of the input signal before performing the time-frequency analysis, and determine the signal-to-noise ratio of the input signal of the real-time domain signal. It is estimated from the power fluctuation speed and converted into a measure of voice content.

At this time, the voice content judging unit 9 converts the estimated signal-to-noise ratio into a logarithm according to the auditory characteristics, and makes the signal-to-noise ratio in the audibility linearly correspond. After that, the voice content determining section 9 calculates the probability that the voice signal is included in the input signal from the logarithmic value, converts the ratio into a measure of the voice content, and outputs the scale.

The signal power monitor 1 of the audio content determination section 9
0 quickly follows the power fluctuation of the input signal and estimates the power of the audio component of the real-time domain signal. The stationary power monitor 11 follows the power fluctuation of the input signal at a low speed and estimates the power of the background noise component of the real time domain signal.

The signal-to-noise ratio calculator 12 calculates the signal-to-noise ratio from these power fluctuations, and the logarithmic calculator 13 converts the signal-to-noise ratio calculated by the signal-to-noise ratio calculator 12 into a logarithm. The voice content rate converter 14 converts the signal-to-noise ratio logarithmized by the logarithmic calculator 13 into a voice content rate, and outputs the voice content rate to the band-specific attenuation calculator 20 of the attenuation calculator 15. .

The attenuation calculator 15 includes a signal power monitor 16
, A stationary power monitor 17 and a signal-to-noise ratio calculator 18
, A logarithmic calculator 19, and a band-specific attenuation calculator 20. The signal power monitor 16 and the steady-state power monitor 17 monitor power fluctuations of all frequency domain data.

The attenuation calculating unit 15 makes a determination on whether to consider the speech content rate as a speech component or a noise component for a weak power variation in a certain band, which is difficult to determine only by monitoring the power variation in the frequency domain. Calculate attenuation using as an aid. Thus, for example, the noise component can be attenuated without damaging the overtone component included in the audio component.

That is, when the signal-to-noise ratio of the real-time signal is large, the attenuation amount calculating section 15 determines that a slight power fluctuation is equivalent to the double voice of the voice because the probability that many voice components are included is high. It is possible to do. In addition, when the signal-to-noise ratio is small, the attenuation amount calculation unit 15 has a low probability that a speech component is included, and can determine a slight power fluctuation as a noise component.

The signal power monitor 16 follows the power fluctuation of the frequency domain data at a high speed and estimates the audio component power of the frequency domain data. The stationary power monitor 17 follows the power fluctuation of the frequency domain data at a low speed,
Estimate the background noise component power of the frequency domain data. Signal to noise ratio calculator 18 calculates a signal to noise ratio from these power fluctuations.

The signal-to-noise ratio is logarithmically calculated by a logarithmic calculator 19, and the attenuation is calculated for each band by a band-specific attenuation calculator 20. When calculating this attenuation, the band-specific attenuation calculator 20 uses the voice content from the voice content converter 14 as a reference parameter.

The signal-to-noise ratio calculators 12 and 18 perform a division operation of dividing the signal power by the noise power. The realization method may be a method of actually performing a division calculation, or simply,
A known method or the like that refers to a conversion table in which a reciprocal value is stored may be used.

The logarithmic calculators 13 and 19 may be realized by a method of actually performing logarithmic calculation, or by a known method that simply refers to a conversion table storing logarithmic values.

The signal power monitor 16 and the steady power monitor 17 and the signal power monitor 10 and the steady power monitor 11 are constituted by the circuits shown in FIG. The circuit shown in FIG. 2 is a known circuit called a leak integration circuit,
By selecting a multiplication coefficient of 02 and 104 to a proper value in accordance with the application, the operation of smoothing the signal is performed, and it is often used as a smoothing circuit or a simple power calculation circuit. Since the leak integration circuit, the window function 3, the window function overlap 7, the FFT 4 and the IFFT 6 are known technologies,
Detailed description is omitted.

FIG. 3 is a diagram showing the characteristic of converting the logarithmic signal-to-noise ratio in the time domain into the voice content by the operation of the voice content converter 14 of FIG. 1, and FIG. FIG. 6 is a diagram illustrating a characteristic of converting a logarithmic signal-to-noise ratio in a frequency domain into an attenuation amount by an operation of an attenuation calculator for each band.

FIG. 5 is a diagram showing how the characteristic of converting the signal-to-noise ratio into the amount of attenuation is changed according to the speech content rate by the operation of the band-specific attenuation calculator 20 of FIG. 1, and FIG. FIG. 7 is a diagram illustrating a state in which the attenuation is changed when the signal-to-noise ratio-attenuation conversion characteristic is changed according to the voice content in one embodiment. The processing operation of the noise canceller 1 will be described in detail with reference to FIGS.

The voice content converter 1 of the voice content calculator 9
Reference numeral 4 denotes an operation for converting the logarithmic signal-to-noise ratio in the real-time domain into the voice content. FIG. 3 shows conversion characteristics when converting to the audio content rate. If the signal-to-noise ratio in the time domain is large, there is a high possibility that a speech component is included in the input signal, so the speech content rate is set to 1. vice versa,
If the signal-to-noise ratio is small, it is unlikely that an audio component is included in the input signal, so the audio content rate is set to 0.

Further, when the signal-to-noise ratio is intermediate between the two, an intermediate value between 1 and 0 which is proportional to the value of the signal-to-noise ratio is set. This intermediate value is a continuous characteristic curve indicating the probability of including a voice estimated from the magnitude of the signal-to-noise ratio.

This characteristic curve is a known arctan function as shown in FIG. 3, but may be approximated without departing from the object of the present invention, and may be a polygonal line, for example. Further, the voice content rate converter 14 may be a method of actually calculating the characteristic curve, or a known method that simply refers to a conversion table.

The attenuation calculator 15 monitors the power fluctuation for each band in the frequency domain. As with the power fluctuation monitoring in the real time domain, there are two signal power monitors 16 that follow fast and a steady power monitor 17 that follows slowly.
Let the ratio of the two powers be the signal to noise ratio.

The band-specific attenuation calculator 20 determines the attenuation based on the signal-to-noise ratio in the frequency domain. FIG. 4 shows a conversion characteristic for obtaining an attenuation amount from a signal-to-noise ratio. This conversion characteristic is divided into three regions, and is a polygonal line composed of a straight line in the maximum attenuation region, a straight line in the non-attenuation region, and a straight line in a region where the attenuation is proportional to the signal-to-noise ratio. .

Although a method of providing an intermediate region as control when clear separation is not easy is known, it is indispensable for realizing the attenuation control method according to one embodiment of the present invention. Will be described.

In the case of a point PS having a large signal-to-noise ratio, since a speech component having a large power exists, the signal-to-noise ratio is converted into an attenuation LS (= 0). In the case of a point PN having a small signal-to-noise ratio, since only a noise component exists, the signal-to-noise ratio is set to the attenuation LN (= maximum attenuation).
Convert to

In the case where the signal-to-noise ratio is at an intermediate point PM, both the state in which the noise is only slightly larger and the state in which a weak sound component such as a harmonic component of the sound is contained. Therefore, the signal-to-noise ratio is set to an attenuation amount LM (= intermediate attenuation amount) as an attenuation amount proportional to the signal-to-noise ratio.
Convert to

In addition to the basic operation based on these techniques,
The attenuation calculator 15 inputs the speech content as a result calculated in the real time domain as a determination factor for determining the attenuation. In order to change the width of the attenuation amount proportional area based on the magnitude of the voice content, an operation of increasing or decreasing the slope of the hatched portion of the attenuation characteristic broken line is performed.

A method of increasing / decreasing the inclination of the diagonal line of the attenuation characteristic broken line will be described. As shown in FIG. 5, according to the magnitude of the audio content, which is the output of the audio content converter 14,
If the audio content is large, the slope will increase.
If it is small, the slope of the hatched portion of the attenuation characteristic polygonal line is increased or decreased in the direction in which the slope becomes smaller. In other words, the attenuation amount calculation unit 15 refers to the voice content rate and performs an operation of increasing or decreasing the inclination of the attenuation characteristic proportional line and the width of the attenuation amount proportional region and the non-attenuation region.

As described above, the fact that dynamically changing the attenuation characteristic has the effect of improving the communication quality will be described below. With reference to FIG. 6, a description will be given of how the position of the point PM in the attenuation proportional region of an arbitrary frequency band is dynamically changed.

First, it is assumed that the RM is in the attenuation proportional region which is an intermediate region where it is difficult to determine whether the signal-to-noise ratio is a voice component or a noise component from monitoring of power fluctuation in the frequency domain. From the signal-to-noise ratio RM, the point PM on the attenuation characteristic line
Is found. Here, since the band-specific attenuation calculator 20 operates to increase or decrease the slope of the oblique line by referring to the audio content, the position of the point PM moves in the vertical axis direction according to the magnitude of the audio content.

When the sound content is high, the sound moves in the direction of decreasing the attenuation, and when the sound content is low, the sound moves in the direction of increasing the attenuation. The attenuation LM converted from the signal-to-noise ratio takes a value from the minimum value LMmin to the maximum value LMmax. That is, even when it is difficult to determine whether the voice component or the noise component is used, it is possible to perform more accurate determination and improve the communication quality.

FIG. 7 is a block diagram showing a configuration of a noise canceller according to another embodiment of the present invention. In the figure,
The other embodiment of the present invention has the same configuration as that of the embodiment of the present invention shown in FIG. 1 except that the noise canceller 21 is provided with the voice content smoothing unit 22, and the same components are denoted by the same reference numerals. I have. The operation of these same components is the same as in the embodiment of the present invention.

In the noise canceller 21, a voice content smoothing unit 22 is provided immediately before the output of the voice content calculation unit 9. The voice content calculating unit 9 has a function of monitoring power fluctuations in the real time domain. However, if the tracking speed of the signal power monitoring unit 10 is set to be high in order to increase the tracking of voices with rapidly changing power, natural conversation will occur. There is a case where the voice content becomes unstable due to the fluctuation of the voice power caused by the inflection of the voice in the above.

The voice content smoothing unit 22 is provided for the purpose of improving the degradation of the communication quality due to this phenomenon. The audio content rate smoothing unit 22 is configured by the circuit shown in FIG. 2 and has a function of smoothing the fluctuation of the audio content rate.

As described above, in another embodiment of the present invention, the voice content can be stabilized by the voice content smoothing device 22, and the determination between the noise component and the voice component can be made more accurate. .

FIG. 8 is a block diagram showing a configuration of a noise canceller according to another embodiment of the present invention. In the figure,
Another embodiment of the present invention has the same configuration as the other embodiment of the present invention shown in FIG. 7 except that a frequency weighting corrector 24 is provided in the noise canceller 23, and the same components are denoted by the same reference numerals. I have. The operation of these same components is the same as in the other embodiments of the present invention.

In the noise canceller 23, the attenuation calculator 1
A frequency weighting corrector 24 is provided immediately before the input of the band-specific attenuation calculator 20 of FIG. One of the factors reducing the quality of a signal after background noise attenuation is the presence of narrow-band noise generated due to insufficient attenuation of noise components compared to neighboring bands. In particular, high-frequency narrow-band noise is unpleasant due to human auditory characteristics, and is a factor that greatly impairs speech quality.

In another embodiment of the present invention, a frequency weighting corrector 24 is provided to improve this problem. As shown in FIG. 9, the correction is performed so that the voice content rate is gradually reduced as the frequency increases in accordance with the human auditory characteristics.
Therefore, an effect of further increasing the noise attenuation effect on the audibility is obtained, and the object of the present invention is achieved.

As described above, in the calculation of the amount of noise attenuation, three regions of no attenuation, maximum attenuation, and intermediate attenuation are provided, and the band which can be clearly judged as a voice component has no attenuation, and the noise component is clear. It is possible to increase the amount of noise attenuation by setting the attenuation band to the maximum attenuation amount and setting the intermediate variable attenuation amount to the band where determination is difficult.

Further, by adopting a configuration in which the attenuation amount for each band is calculated by incorporating the voice content based on the real-time power fluctuation of the input signal as a determination factor, it is difficult to determine by increasing or decreasing the intermediate attenuation amount. It is possible to accurately calculate the amount of attenuation even in the band, and it is possible to reduce the residual noise component and the sound component loss.

Further, by smoothing the voice content, it is possible to stably attenuate noise even for voices including intonation. Furthermore, by performing frequency weighting on the audio content rate, it is possible to suppress generation of harsh high-frequency residual noise.

It should be noted that the present invention is not limited to the configuration of each embodiment described above, and it is clear that the configuration of each embodiment can be appropriately changed within the scope of the technical idea of the present invention. It is not limited to these.

[0066]

As described above, according to the present invention, in a noise canceller for attenuating a noise component by attenuating the voice component in the frequency domain when the voice including the noise component is input, the signal-to-noise ratio in the frequency domain is reduced. Calculate the amount of attenuation in the frequency domain based on the calculation results of the above and the signal-to-noise ratio in the real-time domain to ensure sufficient attenuation of background noise and maintain voice quality. There is an effect that excellent speech quality can be obtained while simultaneously satisfying two conditions that conflict with the above.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a noise canceller according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a leak integrator used in one embodiment of the present invention.

FIG. 3 is a diagram illustrating a characteristic of converting a logarithmic signal-to-noise ratio in a time domain into a voice content by an operation of the voice content converter of FIG. 1;

FIG. 4 is a diagram showing a characteristic of converting a logarithmic signal-to-noise ratio in a frequency domain into an attenuation amount by an operation of the attenuation calculator for each band of FIG. 1;

FIG. 5 is a diagram showing how the characteristic of converting a signal-to-noise ratio into an amount of attenuation is changed in accordance with the audio content rate by the operation of the band-specific attenuation calculator of FIG. 1;

FIG. 6 is a diagram showing a state where the attenuation is changed when the signal-to-noise ratio-attenuation conversion characteristic is changed according to the voice content in one embodiment of the present invention.

FIG. 7 is a block diagram showing a configuration of a noise canceller according to another embodiment of the present invention.

FIG. 8 is a block diagram showing a configuration of a noise canceller according to another embodiment of the present invention.

FIG. 9 is a diagram showing characteristics of frequency-speech content correction coefficient conversion by the operation of the frequency weighting corrector of FIG. 8;

FIG. 10 is a diagram illustrating an example of a conventional time waveform and power of an audio signal on which a background noise component is superimposed.

FIG. 11 is a diagram illustrating an example of frequency analysis of a conventional signal including only an audio component.

FIG. 12 is a diagram illustrating an example of frequency analysis of a conventional signal including only background noise.

FIG. 13 is a diagram illustrating an example of frequency analysis of a conventional audio signal on which a background noise component is superimposed.

[Explanation of symbols]

 1,21,23 Noise canceller 3 Window function 4 FFT 5 Attenuator 6 IFFT 7 Window function overlap 9 Voice content calculator 10,16 Signal power monitor 11,17 Steady power monitor 12,18 Signal to noise ratio calculator 13, 19 Logarithmic Calculator 14 Speech Content Converter 15 Attenuation Calculator 20 Attenuation Calculator for Each Band 22 Speech Content Smoother 24 Frequency Weighting Corrector

Claims (8)

[Claims] 1. A noise canceller for attenuating a noise component by attenuating the voice in a frequency domain when a voice including a noise component is input, wherein a first calculation for calculating a signal-to-noise ratio in the frequency domain is performed. Means for calculating a signal-to-noise ratio in a real-time domain;
And an attenuation calculating means for calculating the attenuation in the frequency domain based on the calculation result of the second calculating means. 2. The noise canceller according to claim 1, further comprising means for converting an output of said second calculating means into a probability of containing an audio component. 3. A noise canceller for attenuating a noise component by attenuating the voice in a frequency domain when a voice including a noise component is input, wherein the attenuation in the frequency domain is determined based on a predetermined conversion characteristic. A noise canceller comprising: setting means for setting any one of a non-attenuation, a maximum value, and an intermediate value between them, and means for dynamically changing the conversion characteristic used in the setting means. 4. The noise canceller according to claim 3, further comprising: means for smoothing a probability of containing the voice component. 5. A noise canceller for attenuating a noise component by attenuating the voice in a frequency domain when a voice including a noise component is input, wherein a first calculation for calculating a signal-to-noise ratio in the frequency domain is performed. Means for calculating a signal-to-noise ratio in a real-time domain;
Setting means for setting the amount of attenuation in the frequency domain based on a calculation result of the second calculating means to one of no attenuation, a maximum value, and an intermediate value based on a predetermined conversion characteristic; Means for converting the output of the second calculation means into a probability of containing a speech component, and attenuation amount changing means for dynamically changing the conversion characteristic used by the setting means, wherein the speech component The noise canceller is configured to use the probability of containing as a reference value of the attenuation amount changing means. 6. The noise canceller according to claim 5, further comprising a smoothing means for smoothing the probability of containing said voice component. 7. The noise canceller according to claim 5, further comprising means for weighting and correcting the probability of containing the voice component in accordance with the frequency. 8. The noise canceller according to claim 6, further comprising means for weighting and correcting the probability of containing said audio component smoothed by said smoothing means according to frequency.