JP2005037617A - Noise reduction system of voice signal - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a technology for improving the quality of noise reduction voice by reducing noise of a voice signal in response to an environmental noise level. <P>SOLUTION: A noise reduction system of the voice signal comprises: an analysis means for inputting voice data as an input signal and analyzing characteristic information concerning a frequency; a means for specifying a noise component and a voice component in the input signal; a means for removing the noise component from the characteristic information and extracting signal information; a pitch extraction means for extracting a pitch component from the input signal or the characteristic information and extracting pitch information of the signal information; a pitch emphasis control means for controlling a ratio of the pitch information increased to the signal information from a ratio of the noise component and the signal component in the input signal or the characteristic information; and a means for converting the signal information controlling the ratio of the pitch information into the voice data. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a technique for reducing a noise component from a voice on which noise such as environmental noise is superimposed.
[0002]
[Prior art]
2. Description of the Related Art Conventionally, there has been known a technique for reducing noise and improving the S / N ratio of a voice signal that has been reduced by enhancing the voice signal by pitch enhancement (see, for example, Patent Document 1).
[0003]
An example of the prior art is shown in FIG. FIG. 1 is a block diagram of a conventional system. The conventional method is realized by the analysis unit 101, the noise removal unit 102, the pitch extraction unit 104, the pitch enhancement unit 103, and the synthesis unit 105.
[0004]
The analysis unit 101 determines whether or not sound is included in the frame data of the input signal (sampled data of the input signal accumulated for a certain period of time). If it is determined that only noise is present, the noise removing unit 102 stores noise feature information. If it is determined that the voice is included, the pitch extraction unit 104 calculates the pitch information and stores the voice feature information. The noise removing unit 102 removes noise based on the noise feature information stored from the frame data, and the pitch emphasizing unit 103 enhances the voice signal based on the stored voice feature information.
[0005]
[Patent Document 1]
Japanese Patent Application No. 6-131568.
[Patent Document 2]
Japanese Patent No. 2830276.
[Patent Document 3]
JP-A-7-334189.
[0006]
[Problems to be solved by the invention]
However, in the above-described prior art, when the environmental noise level is high, it is difficult to accurately calculate the pitch information of the voice due to the influence of noise. In this case, since the audio signal is emphasized based on incorrect pitch information, the audio signal subjected to the enhancement process is deteriorated.
[0007]
In addition, when the environmental noise level is low, the naturalness of the voice is impaired because the voice signal is further emphasized even though the noise can be appropriately removed. In other words, it is desirable not to emphasize the voice when the environmental noise level is low.
[0008]
On the other hand, when the environmental noise level is high to some extent and the pitch information of the voice can be calculated to some extent, the merit of making the voice easier to hear exceeds the demerit that impairs the naturalness of the voice by the emphasis It will be. In such a situation, it is desirable to perform speech enhancement.
[0009]
The present invention has been made in view of such problems of the prior art. That is, the problem to be solved by the present invention is to provide a technique capable of reducing the noise of a voice signal according to the environmental noise level and improving the quality of the noise-reduced voice.
[0010]
[Means for Solving the Problems]
The present invention employs the following means in order to solve the above problems.
[0011]
The noise reduction apparatus according to the present invention inputs voice data as an input signal, analyzes characteristic information relating to frequency, means for specifying a noise component and a voice component in the input signal, and the feature information. Means for removing noise components and extracting signal information; pitch extracting means for extracting pitch information from the input signal or the feature information; and extracting pitch information of the signal information; and the input signal or the feature information Pitch emphasis control means for controlling the ratio of the pitch information to be increased to the signal information from the ratio of the noise component to the signal component, and means for converting the signal information in which the ratio of the pitch information is controlled into audio data And features.
[0012]
With this configuration, if it is not desirable to emphasize pitch information, such as when it can be estimated that the performance of extracting pitch components has deteriorated, or when noise removal has been sufficiently performed, the pitch information is emphasized. It is possible to control to reduce the ratio to be performed. Conversely, when it is desirable to enhance pitch information, such as when it is estimated that the performance of extracting pitch components is high or when noise removal is not sufficient, it is possible to control to increase the pitch information enhancement ratio. . As a result, the noise of the voice signal can be reduced according to the environmental noise level, and the quality of the noise-reduced voice can be improved.
[0013]
Further, the pitch emphasis control means, when the ratio of the signal and noise exceeds a first predetermined value, or when the ratio of the signal and noise is less than a second predetermined value, You may control so that pitch information may not increase.
[0014]
When such control is executed, the enhancement of the pitch information can be stopped when the enhancement of the pitch information is not desirable. As a result, the quality of noise-reduced speech can be improved more comprehensively than in the prior art.
[0015]
Further, the present invention may be one in which the above program is stored in a readable storage medium.
[0016]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings of FIGS. According to the present invention, in a technology using a voice communication system such as a mobile phone system or an IP phone system, noise components can be reduced from a voice signal on which environmental noise input from a microphone is superimposed, thereby improving call quality.
[0017]
FIG. 2 shows an example in which the present invention is applied to a voice communication system. In FIG. 2, first, the noise reduction unit receives an audio signal on which noise is superimposed. Then, the noise reduction unit 201 removes noise from the audio signal and transmits it to the audio encoding unit 202. After that, the speech encoding unit 202 encodes the speech signal from which noise has been removed, and transmits the encoded speech signal to the transmission unit 203. The present invention is used in the noise reduction unit shown in FIG.
[0018]
FIG. 3 is a diagram for explaining the principle of the present invention, and FIGS. 4 to 7 are diagrams showing a processing flow of the present invention.
<Principle>
The principle of the present invention will be described with reference to FIG. FIG. 3 is a block diagram for explaining the principle of the present invention.
[0019]
The present invention includes an analysis unit 301 that inputs an input signal for each frame and decomposes it into frequency components to calculate feature information, and a noise removal unit that removes a noise component contained in the feature information of the input signal based on the feature information. 302, pitch extraction means 304 for extracting pitch information such as pitch frequency, pitch enhancement means 303 for enhancing the pitch characteristics of noise-removed feature information, contribution of pitch enhancement means 303 and contribution of noise removal means 302. A control information calculation unit 307 that calculates control information including the control information, a control unit 306 that calculates final feature information based on the control information, and a synthesis unit 305 that combines the feature information calculated by the control unit into an audio signal. And realized.
[0020]
In the present embodiment, a frame refers to an audio signal that is processed during a predetermined period. In the present embodiment, the audio signal is a sampled discrete signal.
[0021]
As a method for expressing a voice signal, there are an expression expressed in the time domain with time and amplitude as axes, and an expression expressed in the frequency domain with amplitude and frequency as axes. And as characteristic information, the spectrum amplitude and spectrum phase at the time of expressing an audio | voice signal in a frequency domain can be illustrated, for example. The height of the sound determined by the vibration period of the sound is called a pitch. The reciprocal of this pitch is called the fundamental frequency, that is, the pitch frequency.
[0022]
Hereinafter, each component will be further described.
[0023]
The analysis unit 301 inputs an input signal for each frame, analyzes the frame, and calculates feature information such as a spectrum. Then, the analyzing unit 301 outputs the calculated feature information to the noise removing unit 302 and the pitch extracting unit 304.
[0024]
The noise removing unit 302 receives the feature information from the analyzing unit 301 and removes a noise component from the feature information of the input signal. The noise removing unit 302 outputs the feature information from which the noise component has been removed to the pitch emphasizing unit 303 and the control unit 306.
[0025]
The pitch extraction unit 304 inputs feature information from the analysis unit 301 and outputs pitch information. This pitch information is information including the presence / absence of pitch characteristics and the pitch frequency when there is pitch characteristics.
[0026]
The pitch emphasizing unit 303 receives pitch information from the pitch extracting unit 304 and inputs feature information from which noise has been removed from the noise removing unit 302. The pitch emphasizing unit 303 emphasizes the pitch property of the feature information based on the pitch information and outputs the feature information. Emphasizing the pitch property means, for example, increasing the frequency component (power) corresponding to the pitch frequency.
[0027]
The control information calculation unit 307 receives the input signal, calculates control information, and outputs the control information to the control unit. This control information is used to calculate the contribution of the noise removal unit 302 and the contribution of the pitch enhancement unit 303. Here, contribution means a weight indicating how much the output of the pitch emphasis means is added to the output of the noise removing means 302.
[0028]
The control unit 306 receives the pitch-enhanced feature information from the pitch enhancement unit 303, the noise-removed feature information from the noise removal unit 302, and the control information calculation unit 307 receives the control information.
[0029]
Then, the control unit 306 determines the contribution of the output of the pitch enhancement unit 303 and the output of the noise removal unit 302 based on the control information, and calculates feature information. Then, the control unit 306 outputs the calculated feature information to the synthesis unit.
[0030]
The synthesizing unit 305 receives the feature information from the control unit 306 and executes a process of synthesizing the voice using the feature information. Then, the synthesizing unit 305 outputs the processed sound as an output signal.
<Processing flow>
Hereinafter, a processing flow of audio processing in the audio communication method of the present embodiment will be shown. This voice processing can be realized as a program on a processor such as a CPU (Central Processing Unit) or a DSP (Digital Signal Processor). Moreover, you may comprise as hardware for exclusive use (gate array, custom LSI, etc.) which implement | achieves such a process.
[0031]
With reference to FIG. 4, the entire process from the input of the input signal to the output of the output signal will be described. FIG. 4 shows the overall flow of processing in the present embodiment. The processing shown in FIG. 4 is executed by the analysis unit 301, noise removal unit 302, pitch extraction unit 304, pitch enhancement unit 303, control information calculation unit 307, control unit 306, and synthesis unit 305.
[0032]
First, the analysis unit 301 inputs an input signal for one frame (S401). The audio is sampled and stored in a buffer (not shown) in sequence as an A / D converted audio signal. Therefore, the analysis unit 301 cuts out an audio signal from the buffer as an input signal and performs analysis for each frame.
[0033]
The analysis unit 301 performs FFT analysis on the input signal (S402). At this time, the analysis unit 301 acquires the spectrum amplitude Sa [i] and the phase Sp [i]. The spectrum amplitude Sa [i] and the phase Sp [i] represent the audio signal in the frequency domain, and each processing means performs processing such as noise removal and pitch enhancement on the spectrum amplitude and phase. To do. Here, i indicates an index of the frequency band.
[0034]
In the present embodiment, analysis processing is executed by FFT (Fast Fourier Transform: Fast Fourier Transform), and synthesis processing is executed by IFFT (Inverse Fast Fourier Transform: Fast Inverse Fourier Transform). The processing procedure of the input signal executed during the analysis / synthesis by FFT / IFFT is described in detail in, for example, “Computer Music” (Curtis Roads, translated and supervised by Tatsuya Aoyagi et al., Tokyo Denki University Press, pages 452-457). Explained.
[0035]
The noise removing unit 302 removes a noise component from the spectrum amplitude Sa [i] of the input signal (S403). Here, the noise removing unit 302 calculates the spectrum amplitude Sa ′ [i] after the noise removal, and transmits it to the pitch emphasizing unit 303 and the control unit 306. As a typical method of noise removal, a spectral subtraction method (JS Lim and AV Oppenheim: “Enhancement and Bandwidth Compression of Noise Speed,” Proc. IEEE, Vol. 67, No. 12). , Pp. 1586-1604, 1979.).
[0036]
The pitch extraction unit 304 performs pitch extraction (S404). Here, the pitch extracting unit 304 acquires the pitch frequency F (Hz) and transmits it to the pitch emphasizing unit 303. The pitch extraction method is described in detail in “Computer Music” (Curtis Roads, translated and supervised by Tatsuya Aoyagi et al., Tokyo Denki University Press, pages 417-425).
[0037]
The pitch emphasizing unit 303 performs pitch emphasis on the spectrum amplitude Sa ′ [i] after noise removal (S405). Here, the pitch emphasizing unit 303 performs pitch emphasis on the pitch frequency F (Hz) extracted by the pitch extracting unit 304 and its high frequency out of the spectrum amplitude Sa ′ [i] calculated by the noise removing unit 302, and thereby the spectral amplitude. Sa ″ [i] is calculated. The spectrum amplitude Sa ″ [i] calculated here is transmitted to the control means 306.
[0038]
The control information calculation unit 307 calculates control information (S406). That is, the control information calculation unit 307 calculates control information using the input signal in accordance with the magnitude of noise, the signal-to-noise ratio (hereinafter referred to as SNR (signal to noise ratio)), and the like. This control information indicates a ratio contributed by spectrum amplitudes Sa ′ [i] and Sa ″ [i] used when calculating the final spectrum amplitude S [i]. The calculated control information is transmitted to the control means 306. The control information calculation method will be described later in detail.
[0039]
The control unit 306 calculates feature information (S407). The control unit 306 weights Sa ′ [i] and Sa ″ [i] using the control information calculated by the control information calculation unit 307, and obtains the final spectrum amplitude S [i].
[0040]
The synthesizing unit performs IFFT using the final spectrum amplitude S [i] and phase Sp [i] to synthesize an audio signal (S408).
[0041]
Then, the analysis unit 301 determines whether or not the input of the audio signal is finished (S409).
[0042]
If the analysis unit 301 determines in step S409 that the audio signal input signal has ended, the analysis unit 301 ends the process. On the other hand, if it is determined in S409 that the input signal continues, the analysis unit 301 proceeds to S401 and processes the next frame.
<Processing of control information calculation means>
A processing flow of the control information calculation unit will be described with reference to FIG. FIG. 5 shows the flow of processing of the control information calculation means. The process of the control information calculation unit 307 is executed in S406 shown in FIG. Hereinafter, a process in which the control information calculation unit 307 calculates control information from an input signal will be described. In the present embodiment, the control information is represented by an output contribution A of the pitch emphasis means and an output contribution B of the noise removal means.
[0043]
The control information calculation unit 307 inputs an input signal for one frame (S501). Then, the control information calculation unit 307 calculates the frame power E (S502). The frame power is the signal power of the input signal in one frame, and is used when calculating the SNR. The SNR (signal to noise ratio) is the ratio of signal to noise, and can be calculated by dividing the signal power by the noise power.
[0044]
The control information calculation unit 307 calculates the pitch correlation C (S503). The pitch correlation is a function for indicating how much a frequency component corresponding to the pitch frequency is included in the input signal. The pitch correlation can be calculated by, for example, calculating a correlation value for the sample having the highest correlation between the samples with respect to the sampled input signal X [i] ((i = 1 to N), N: sampling frequency). This can be calculated by sequentially shifting a specific input signal X [k] on the time axis and calculating a correlation value with another input signal sample X [k ′].
[0045]
The control information calculation unit 307 determines whether the pitch correlation C exceeds a threshold value (S504). If the control information calculation unit 307 determines in step S504 that pitch correlation C> threshold, the control information calculation unit 307 updates the average Ev of the audio power according to the following formula 1 (S505).
[0046]
[Expression 1]
Ev = ξ × E + (1−ξ) × Ev ′ (ξ (0 ≦ ξ ≦ 1): a predetermined constant)
Here, the average Ev ′ of voice power is the average value of voice power up to the previous time, and this average Ev ′ is stored in a storage area such as a memory. When the average Ev is updated by adding the frame power E extracted this time, the average Ev can be calculated from Equation 1 using a predetermined constant ξ. The constant ξ is a value indicating the contribution of the current frame power E to the average value of the audio power. The voice power refers to the power spectrum of the input signal included in the voice signal.
[0047]
On the other hand, if the control information calculation unit 307 determines in step S504 that pitch correlation C ≦ threshold, the average En of the noise power is updated according to the following formula 2.
[0048]
[Expression 2]
En = η × E + (1−η) × En ′ (η (0 ≦ η ≦ 1): a predetermined constant)
This average noise power En ′ is an average value of noise power up to the previous time, and this noise power average En ′ is stored in a storage area such as a memory, in the same manner as the average of sound power. When the average En is updated by adding the frame power E extracted this time, the average En can be calculated from Equation 2 using a predetermined constant η. The constant η is a value indicating the contribution of the current frame power E to the average noise power. The average En of the noise power refers to the power spectrum of the noise signal included in the audio signal.
[0049]
Next, the control information calculation unit 307 calculates the SNR according to the following Equation 3 (S507).
[0050]
[Equation 3]
SNR = Ev / En
Then, the control information calculation unit 307 calculates the value of Expression 4 below (S508).
[0051]
[Expression 4]
A = SNR × γ (γ: predetermined constant)
This value is used to determine whether or not to use the output of the pitch emphasizing means after voice synthesis. This value is called contribution A.
[0052]
Next, the control information calculation unit 307 calculates the value of Equation 5 below (S509).
[0053]
[Equation 5]
B = SNR × η (η: a predetermined constant)
This value is used for calculating a weighting coefficient when the output of the pitch emphasizing means is used for an audio signal. This value is called contribution B.
[0054]
The control information calculation means 307 outputs control information A and B (S510). The control information A and B is also referred to as pitch extraction performance. The control information calculation unit 307 transmits the output contribution A of the pitch enhancement unit and the output B of the noise removal unit 302 to the control unit 306 as control information.
<Processing of control means>
With reference to FIG. 6, the process of a control means is demonstrated. FIG. 6 shows the flow of processing of the control means. The process of the control means is a process executed in S407 shown in FIG.
[0055]
Based on the control information calculated by the control information calculation unit 307, the feature information calculated by the noise removal unit 302, and the feature information calculated by the pitch enhancement unit, the control unit 306 uses the spectrum amplitude S [ i] and the spectral phase Sp [i] are calculated. Hereinafter, a process in which the control unit 306 calculates feature information will be described.
[0056]
The control means 306 inputs control information A and B (S601). The control unit 306 receives, as control information, the output contribution A of the pitch enhancement unit 303 and the output contribution B of the noise removal unit from the control information calculation unit 307.
[0057]
The control unit 306 inputs the feature information (spectrum amplitude Sa ′ [i] and phase Sp [i]) output from the noise removing unit 302 (S602). The control unit 306 receives the spectrum amplitude Sa ′ [i] and the phase Sp ′ [i] from the noise removing unit 302 as the feature information.
[0058]
The control unit 306 inputs the feature information (spectrum amplitude Sa ″ [i] and phase Sp ″ [i]) output from the pitch emphasizing unit (S603). The control unit 306 receives the spectral amplitude Sa ″ [i] and the phase Sp ″ [i] from the pitch emphasizing unit 303 as feature information. Then, the control means determines the weighting factor W from the relationship between the control information A and the threshold 1 and threshold 2 (threshold 1> threshold 2).
[0059]
The control unit 306 determines whether the control information A (hereinafter also referred to as pitch extraction performance A) exceeds the threshold value 1 (S604).
[0060]
In S604, when it is determined that the pitch extraction performance A> the threshold value 1, the control unit 306 sets the weighting coefficient W = 0 (S607). On the other hand, in S04, when it is determined that the pitch extraction performance A ≦ the threshold value 1, the control unit 306 determines whether the pitch extraction performance A exceeds the threshold value 2 (S605).
[0061]
In S605, when it is determined that the pitch extraction performance A <threshold value 2, the control unit 306 sets the weighting factor W = 0 (S607). On the other hand, if the control unit 306 determines in step S605 that contribution A ≧ threshold value 2, the weighting factor is set to W = αB + β (S606). In this case, the contribution A is in the range of threshold 2 ≦ A ≦ threshold 1. Here, α and β are predetermined constants. Contribution B is the contribution of the output of the noise removal means 302 calculated by the control information calculation means 307.
[0062]
The control means 306 calculates the spectrum amplitude S [i] according to the following formula 6 (S608).
[0063]
[Formula 6]
S [i] = W × Sa ″ [i] + (1−W) Sa ′ [i]
That is, when the pitch extraction performance A is between the threshold value 2 and the threshold value 1, the control means 206 considers (weights and adds) the spectrum amplitude Sa ″ [i] with the pitch emphasis, and the spectrum amplitude S [i ], And when the pitch extraction performance A is not between the threshold value 2 and the threshold value 1, the spectrum amplitude Sa ′ [i] from which noise is removed is set as the final spectrum amplitude S [i].
[0064]
The control means 306 outputs the final spectrum amplitude S [i] and spectrum phase Sp [i] as feature information (S609). The control unit 306 transmits the calculated spectrum amplitude S [i] and spectrum phase Sp [i] to the synthesis unit 305.
<Processing of analysis means and pitch extraction means>
The processing of the pitch extraction unit 307 will be described with reference to FIG. FIG. 7 shows the processing flow of the analysis means 301 and the pitch extraction means 304. The process of the analysis means 301 is performed by S402 shown in FIG. Further, the processing of the pitch extraction unit 304 is executed in S404 shown in FIG. Hereinafter, a process of calculating the pitch frequency by the analysis unit 301 receiving the input signal and the pitch extraction unit 304 determining whether or not there is pitch property will be described.
[0065]
The analysis unit 301 inputs an input signal for one frame (S701). The analysis unit 301 cuts out the audio signal for each frame and executes the analysis, as in S401 shown in FIG.
[0066]
The analysis unit 301 performs FFT to calculate the spectrum amplitude and the like (S702). That is, here, the analysis unit 301 converts the time domain audio signal into a frequency domain audio signal.
[0067]
The pitch extraction means calculates the peak of the spectrum amplitude in the frequency band of 1000 Hz or less (S703). Then, the pitch extracting means extracts three of the peaks of the spectrum amplitude from the largest (S704). The frequencies extracted here are P1, P2, and P3 (Hz) from those having a lower frequency.
[0068]
The pitch extraction means determines whether or not the peak frequencies P1, P2, P3 (Hz) satisfy the following Expression 7 (S705). Here, the pitch extraction unit 301 determines whether or not the extracted frequency has pitch characteristics based on the following Equation 7. The peak calculation performed by the pitch extraction means may use spectrum power which is the square of the spectrum amplitude.
[0069]
[Expression 7]
| P2 / P1-2 | <threshold and | P3 / P1-3 | <threshold
In S705, the pitch extraction unit 307 outputs the pitch frequency P1 when it is determined that the above mathematical formula 7 is established (S706). That is, when the audio signal has pitch characteristics, the lowest frequency P1 is the fundamental frequency, the second frequency P2 is twice the fundamental frequency, and the third largest frequency P3 is three times the fundamental frequency. Become. Therefore, if the value obtained by subtracting 2 from the value obtained by dividing the frequency P2 by the basic frequency and the value obtained by subtracting 3 from the value obtained by dividing the frequency P3 by the basic frequency are within the predetermined range, the pitch extracting means 207 determines the voice. Judge that the signal has pitch characteristics.
[0070]
On the other hand, if it is determined in S706 that the above formula 7 is not satisfied, the pitch extracting means outputs that there is no pitch property (S707).
<Effect of embodiment>
According to the present embodiment, the control information calculation means calculates control information for controlling the feature information calculated by the pitch emphasis means and the feature information calculated by the noise removal means. Then, the control means can calculate final feature information based on the control information by using the feature information that has been subjected to noise removal and the feature information that has been subjected to both noise removal and pitch enhancement.
[0071]
Further, according to the present embodiment, when pitch enhancement is not desirable, such as when it can be estimated that the pitch extraction performance is deteriorated or when noise removal is sufficiently performed, the contribution of pitch enhancement is reduced, or Stop pitch emphasis.
[0072]
Conversely, when it is desirable to perform pitch enhancement, such as when the pitch extraction performance is high to some extent and noise removal is not sufficient, pitch enhancement is performed. Further, in that case, the contribution of pitch enhancement can be increased according to the pitch extraction performance.
[0073]
For this reason, the conventional technique may deteriorate when pitch emphasis is not desired, whereas in the present invention, when pitch emphasis is desired, sound quality equivalent to that of the prior art can be obtained, and pitch emphasis is not desirable. Since it is possible to prevent deterioration due to pitch enhancement, it is possible to improve the quality of noise-reduced speech more comprehensively than the prior art.
<Computer-readable storage medium>
A program that causes a computer to execute any of the processes in the above-described embodiments can be recorded on a computer-readable storage medium. The system described in the above embodiment can be provided by causing a computer to read and execute the program of the storage medium.
[0074]
Here, the computer-readable storage medium refers to a storage medium that stores information such as data and programs by electrical, magnetic, optical, mechanical, or chemical action and can be read from the computer. . Examples of such storage media that can be removed from the computer include a flexible disk, a magneto-optical disk, a CD-ROM, a CD-R / W, a DVD, a DAT, an 8 mm tape, and a memory card.
[0075]
Further, as a recording medium fixed to the computer, there are a hard disk, a ROM (read only memory), and the like.
[0076]
The above embodiment does not limit the scope of the present invention at all, and various modifications can be appropriately made within the scope that can be understood by those skilled in the art.
<Others>
Furthermore, this embodiment discloses the following invention.
[0077]
(Additional remark 1) The analysis means which inputs audio | voice data as an input signal, and analyzes the characteristic information which concerns on a frequency,
Means for identifying a noise component and a speech component in the input signal;
Means for removing noise components from the feature information and extracting signal information;
Pitch extraction means for extracting a pitch component from the input signal or the feature information and extracting pitch information of the signal information;
Pitch emphasis control means for controlling a ratio of the pitch information to be increased to the signal information from a ratio of a noise component and a signal component in the input signal or the characteristic information;
And a means for converting the signal information, in which the ratio of the pitch information is controlled, into audio data.
[0078]
(Additional remark 2) The said pitch extraction means is a noise reduction device of the audio | voice signal of Additional remark 1 which extracts the pitch component of an input signal using the spectrum amplitude or spectrum power of an input signal.
[0079]
(Supplementary note 3) The speech signal noise reduction device according to supplementary note 1, wherein the pitch emphasis control means does not increase the pitch information when the ratio of the signal component and the noise component exceeds a first predetermined value. .
[0080]
(Supplementary note 4) The speech signal noise reduction device according to supplementary note 1, wherein the pitch emphasis control means does not add the pitch information when the ratio between the signal component and the noise component is less than a second predetermined value.
[0081]
(Appendix 5) Inputting voice data as an input signal and analyzing characteristic information related to frequency;
Identifying a noise component and a speech component in the input signal;
Removing noise components from the feature information and extracting signal information;
Extracting a pitch component from the input signal or the feature information and extracting pitch information of the signal information;
Controlling a ratio of the pitch information to be increased to the signal information from a ratio of a noise component and a signal component in the input signal or the feature information;
A program for causing a computer to execute the step of converting the signal information in which the ratio of the pitch information is controlled into audio data.
[0082]
(Supplementary note 6) The program according to supplementary note 5, wherein the step of extracting the pitch information extracts a pitch component of the input signal using a spectral amplitude or a spectral power of the input signal.
[0083]
(Supplementary note 7) The program according to supplementary note 5, wherein the step of controlling the ratio of the pitch information does not increase the pitch information when the ratio of the signal component and the noise component exceeds a first predetermined value.
[0084]
(Supplementary note 8) The program according to supplementary note 5, wherein the step of controlling the ratio of the pitch information does not add the pitch information when the ratio of the signal component and the noise component is less than a second predetermined value.
[0085]
【The invention's effect】
As described above, the present invention can provide a technique capable of reducing the noise of a voice signal according to the environmental noise level and improving the quality of the noise-reduced voice.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of a conventional method.
FIG. 2 is a diagram showing an example in which the present invention is applied to a voice communication system.
FIG. 3 is a diagram illustrating the principle of the present invention.
FIG. 4 is a diagram showing an entire processing flow.
FIG. 5 is a diagram showing a processing flow of control information calculation means.
FIG. 6 is a diagram showing a processing flow of a control means.
FIG. 7 is a diagram showing a processing flow of pitch extraction means.
[Explanation of symbols]
101 ... Analyzing means in the conventional system
102: Noise removal means in the conventional system
103 ... Pitch emphasis means in the conventional system
104: Pitch extraction means in the conventional system
105. Combining means in the conventional system
201: Noise reduction unit
202: Speech encoding unit
203: Transmitter
301 ... Analyzing means
302: Noise removing means
303 ... Pitch emphasis means
304 ... Pitch extraction means
305 .. Synthesis means
306 ... Control means
307: Control information calculation means

Claims (5)

Analysis means for inputting voice data as an input signal and analyzing characteristic information related to frequency;
Means for identifying a noise component and a speech component in the input signal;
Means for removing noise components from the feature information and extracting signal information;
Pitch extraction means for extracting a pitch component from the input signal or the feature information and extracting pitch information of the signal information;
Pitch emphasis control means for controlling a ratio of the pitch information to be increased to the signal information from a ratio of a noise component and a signal component in the input signal or the characteristic information;
And a means for converting the signal information, in which the ratio of the pitch information is controlled, into audio data. 2. The audio signal noise reduction device according to claim 1, wherein the pitch extraction unit extracts a pitch component of the input signal using a spectral amplitude or a spectral power of the input signal. 2. The noise reduction device for an audio signal according to claim 1, wherein the pitch emphasis control means does not increase the pitch information when a ratio between the signal component and the noise component exceeds a first predetermined value. 2. The noise reduction apparatus for an audio signal according to claim 1, wherein the pitch emphasis control means does not add the pitch information when the ratio between the signal component and the noise component is less than a second predetermined value. Inputting voice data as an input signal and analyzing frequency-related feature information;
Identifying a noise component and a speech component in the input signal;
Removing noise components from the feature information and extracting signal information;
Extracting a pitch component from the input signal or the feature information and extracting pitch information of the signal information;
Controlling a ratio of the pitch information to be increased to the signal information from a ratio of a noise component and a signal component in the input signal or the feature information;
A program for causing a computer to execute the step of converting the signal information in which the ratio of the pitch information is controlled into audio data.

Images