JP2018066963A - Sound processing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce each noise in the case where wind noise and drive noise have been superimposed on sound.SOLUTION: A sound processing device includes conversion means that converts an input sound signal into a sound signal spectrum, storage means that stores a profile concerning the amplitude per frequency of drive noise, calculation means that calculates a subtraction coefficient on the basis of an output from the conversion means, correction amount determination means that determines a correction amount in accordance with the magnitude of wind noise included in the input sound signal, correction means that corrects a subtraction coefficient with the correction amount, determination means that determines a reduction value of noise per frequency from the corrected subtraction coefficient and the profile, noise reduction means that performs drive noise reduction processing by subtracting the reduction value from the sound signal spectrum output from the conversion means, inverse conversion means that converts the sound signal spectrum from the noise reduction means into a sound signal of a time domain, and reduction means that reduces wind noise from the sound signal output from the inverse conversion means.SELECTED DRAWING: Figure 5

Description

  The present invention relates to a voice processing apparatus.

  Conventionally, in an imaging apparatus such as a digital camera, audio is recorded together with a captured moving image. Digital cameras are often used outdoors. For this reason, wind hits the microphone, and noise caused by the wind is included in the voice. Digital cameras are smaller in size than video cameras, and the optical system such as a lens is close to the microphone. For this reason, driving sound generated by the motor that drives the aperture and the lens is included in the sound as noise.

  Conventionally, techniques for reducing noise caused by wind (wind noise) and noise (driving noise) caused by driving of an optical system have been proposed. As a noise reduction method, a spectral subtraction method (hereinafter referred to as SS method) is known (Patent Document 1).

JP 2000-330597 A

  However, assuming that the drive noise and wind noise reduction processing is performed by the SS method when a sound is input such that the drive noise band and the wind noise band overlap, there are the following problems.

  FIG. 12 is a diagram illustrating the phase vectors of the Lch and Rch audio signals when the wind noise (wind noise) and the driving sound are superimposed. Here, for example, the driving sound is assumed to be noise (zoom sound) generated by driving the zoom lens. In FIG. 12A, it is assumed that the wind noise R1201 and the wind noise L1202 are input in opposite phases, and the zoom sound R1203 and the zoom sound L1204 are input in the same phase.

  At that time, the wind noise reduction process and the drive noise reduction process are performed as shown in FIG. In FIG. 12B, the drive noise reduction processing 1207 and 1208 subtract the scalar amount from the zoom sound + wind noise 1205 and 1206. Further, the wind noise reduction processes 1209 and 1210 subtract the scalar amount of the phase difference. Therefore, there is a problem that noise 1211 having the same phase remains in the zoom sound and cannot be suppressed.

  An object of this invention is to provide the apparatus which can reduce each noise, when a wind noise and a drive noise are superimposing on the audio | voice.

  Input means, driving means, conversion means for dividing the time-domain audio signal obtained by the input means at regular intervals and converting it into a frequency-domain audio signal spectrum, and driving by noise generated by the driving means Storage means for storing a profile relating to amplitude for each frequency of noise, calculation means for calculating a subtraction coefficient based on an output signal from the conversion means, and magnitude of wind noise included in the audio signal obtained from the input means Accordingly, a correction amount determining means for determining a correction amount of the subtraction coefficient, a correction means for correcting the subtraction coefficient obtained by the calculation means with a correction amount obtained by the correction amount determination means, and the correction means And a determination means for determining a noise reduction value for each frequency from the subtraction coefficient corrected in accordance with the profile stored in the storage means and an output from the conversion means. Noise reduction means for reducing the drive noise by subtracting the reduction value obtained by the determination means from the spectrum of the audio signal in the frequency domain; and the audio in the frequency domain from the noise reduction means Inverse conversion means for converting a signal spectrum into an audio signal in the time domain, and wind noise reduction means for reducing wind noise from the audio signal output from the inverse conversion means.

  According to the present invention, even when wind noise and driving noise are superimposed on speech, each noise can be reduced.

It is a block diagram of an imaging device including a voice processing device. It is a flowchart which shows an audio | voice process. It is a figure which shows the reduction circuit of a wind noise. It is a figure which shows the drive noise reduction circuit. It is a figure which shows the drive noise reduction circuit. It is a flowchart showing the process of the correction amount determination part. It is a figure which shows the detection amount and subtraction coefficient (gamma) of a wind noise. It is a flowchart which shows the process of the correction amount determination part. It is a figure which shows a subtraction coefficient. It is a figure which shows the drive noise reduction circuit. It is a figure which shows the production | generation process of a noise profile. It is a figure explaining a drive noise reduction process and a wind noise reduction process.

Example 1
Embodiments of the present invention will be described below.

  FIG. 1 is a block diagram showing a configuration example of an imaging apparatus to which a sound processing apparatus of the present invention is applied. The imaging apparatus 100 in FIG. 1 mainly performs noise reduction processing by the audio processing unit 107. Many of the components described below are connected to the memory bus 117 and exchange data with the memory 116.

  The memory 116 is a dynamic RAM that can be randomly accessed at high speed. The memory includes an audio data area, an image data area, and a control signal area. In addition, the audio data stored in the audio data area, the image data stored in the image data area, and the timing signal in the timing signal area are managed so as to identify the time of each frame.

  In FIG. 1, a lens 101 performs a zoom operation by a motor 102 controlled by a motor driving unit 103. The imaging unit 104 photoelectrically converts an optical image of a subject formed through the lens 101 by an imaging element such as a CCD sensor or a CMOS sensor to generate an analog image signal. Then, the generated analog image signal is converted into a digital signal and transmitted to the image processing unit 105. The image processing unit 105 performs an image quality adjustment process for adjusting the white balance, color, brightness, and the like according to the set value on the input digital image signal, and transmits the processed image to the memory 116. Remember.

  The voice input unit 106 is a built-in microphone or an external microphone connected via a voice input terminal. The sound processing unit 107 converts an analog sound signal obtained by collecting sound around the apparatus by the sound input unit 106 into a digital signal, and performs sound-related processing such as appropriate level processing and specific frequency reduction processing. Do. The audio processing unit 107 performs drive noise reduction processing and wind noise reduction processing as will be described later. Then, the audio processing unit 107 transmits the digital audio signal to the memory 116 and stores it in the audio data area in the memory 116. Also, the audio processing unit 107 reads out the audio data stored in the audio data area of the memory 116 from the communication unit 109, performs the same processing, and stores it again in the audio data area of the memory 116.

  The display unit 108 may be anything as long as it is a display device such as a liquid crystal display, an organic EL display, and electronic paper. For example, display image data is temporarily stored in the image data area of the memory 116 from the image processing unit 105. The display unit 108 reads out the image data from the image data area and displays it on the display.

  The communication unit 109 receives the digital audio signal transmitted from the wireless microphone and temporarily stores it in the control signal area of the memory 116. The communication unit 109 communicates with an external device, and transmits and receives control signals for shooting operations such as audio signals, image signals, shooting start and end commands, and the like. The communication unit 109 is, for example, a wireless communication module such as an infrared communication module, a Bluetooth (registered trademark) communication module, or a wireless LAN module.

  The recording / reproducing unit 110 reads (reproduces) the image data recorded on the recording medium 111 and transmits it to the display unit 108 and the image output unit 114. The recording medium 111 is a random access recording medium such as a memory card or HDD.

  The image output unit 114 includes, for example, an image output terminal, and transmits an image signal to display an image on an external display or the like connected to the imaging device 100. The audio output unit 115 includes an audio output terminal, for example, for reading audio data stored in the audio data area of the memory 116 and outputting audio from an earphone or a speaker connected to the imaging apparatus 100. Send an audio signal. The audio output unit 115 may be a speaker that is built in the imaging apparatus 100 and outputs audio corresponding to an audio signal. In addition, the image output unit 114 and the audio output unit 115 may be a single integrated terminal, for example, a terminal such as HDMI (registered trademark) (High Definition Multimedia Interface).

  The operation unit 113 is, for example, a button or a dial, and transmits an instruction signal corresponding to a user operation to the control unit 112. The control unit 112 controls each block by transmitting a control signal to each block of the imaging apparatus 100 based on the instruction signal transmitted from the operation unit 113. The operation unit 113 is, for example, a power button, a recording start button, a menu display button, a determination button, a cursor key, a touch panel, and the like. The control unit 112 includes, for example, a CPU, DRAM, SRAM, and the like for executing various processes. The memory bus 117 is used for transmitting various data and control signals to each block of the imaging apparatus 100.

  Next, the internal configuration of the audio processing unit 107 will be described. FIG. 2 is a flowchart showing the flow of audio processing of the conventional audio processing unit 107. In S201, drive noise reduction processing is performed by the SS method. Details of the drive noise reduction processing by the SS method will be described later. In S202, a wind noise reduction process is performed using the audio output subjected to the camera drive noise process in S201. Details of the wind noise reduction processing will be described later. In S203, an auto level control process (ALC) is performed for the audio signal that has undergone a series of noise reduction processes up to S202 in order to adjust the audio signal to a constant volume. For example, when the sound collected by the microphone is small, the amplitude is amplified, and when the sound volume is appropriate, the amplification factor of the amplitude is restored. In general, an upper limit is set for the amplification factor.

  Hereinafter, a wind noise reduction process and a camera drive noise reduction process using the SS method will be described.

<Wind noise reduction method>
When picking up outdoor sound using a microphone or the like, wind may hit the microphone and wind noise may be recorded. Therefore, in the present embodiment, wind noise reduction processing is performed on each of the left and right audio signals acquired by a microphone (stereo microphone) that independently collects sound from the left and right directions. The sound picked up by two microphones that are omnidirectional and close to each other generally has the following characteristics.
(1) When sound is picked up in an environment where there is no wind, in-phase sound is picked up by the left and right microphones. In particular, the low frequency is in phase.
(2) When sound is collected in an environment with wind, the wind noise is concentrated at a low frequency (about 50 Hz to several kHz), and a phase difference occurs in the sound signal when the left and right microphones collect sound.

  The wind noise reduction processing is performed in consideration of the above two features. FIG. 3A is a diagram illustrating a circuit block 300 that performs wind noise reduction processing in the audio processing unit 107. FIG. 3B is a graph showing an example of the amplitude characteristic of the high pass filter 303. As will be described later, the wind noise processing circuit 300 in FIG. 3A receives the Lch and Rch audio signals after the noise reduction processing by the SS method is performed, and performs the wind noise reduction processing. .

  In FIG. 3A, the left channel (Lch) and the right channel (Rch) of the stereo audio signal input to the wind noise reduction circuit 300 are added by the addition circuit 301 and subtracted by the subtraction circuit 302. In the high-pass filter circuit 303, for example, a component of 50 kHz or less, which is a frequency band included in a lot of wind noise as shown in FIG. 3B, is attenuated. The high pass filter 303 removes the low frequency of the difference value of the stereo audio signal obtained by the subtraction circuit 302 and outputs the result. Here, for example, a filter that attenuates 3 dB at 1 kHz is used.

  The output from the high-pass filter 303 and the output from the adder circuit 301 are added by the adder circuit 304, and the controller 307 adjusts the volume of the output from the adder circuit 304 in half and outputs it to the Lch audio output unit. Further, the subtracter circuit 305 subtracts the output of the high-pass filter 303 from the output of the adder circuit 301, and the controller 307 adjusts the volume of the output of the subtractor circuit 305 by half and outputs it to the Rch audio output unit. Thus, the wind noise included in the input voice is reduced by the wind noise reduction circuit 300.

<Drive noise reduction processing method by SS method>
The configuration of the drive noise reduction process using the SS method will be described with reference to FIG. In FIG. 4A, noise profiles 401 and 402 store the frequency components of noise to be reduced as noise profiles. Specifically, an audio signal consisting only of noise to be reduced is Fourier transformed to obtain a frequency component. At this time, when the noise to be reduced continues for a certain period of time (for example, 4 seconds), a noise profile is obtained by peak holding with respect to the temporal change of the frequency component within the time during which the noise continues. The noise profiles stored in the noise profiles 401 and 402 may be compressed as long as they can be restored to some extent. Compression or non-compression of the stored noise profile is not limited. The noise profiles 401 and 402 transmit the stored noise profile to the frequency component comparison units 405 and 406 and the adder / subtractors 411 and 412.

  The Fourier transform units 403 and 404 divide the input audio signal at regular time intervals (frames). Then, Fourier transform is performed on the divided digital audio signal in the time domain to convert it into an audio signal spectrum in the frequency domain. As a result, the phase information for each frequency of the audio signal and the absolute value (frequency component) of the amplitude for each frequency are calculated. The Fourier transform units 403 and 404 transmit the calculated frequency components to the frequency component comparison units 45 and 406 and the noise reduction units 413 and 414. Further, the Fourier transform units 403 and 404 transmit the calculated phase information for each frequency to the inverse Fourier transform units 415 and 416.

  The frequency component comparison units 405 and 406 are dividers, and divide the frequency components of the input speech transmitted from the Fourier transform units 403 and 404 for each frequency by the noise profiles from the noise profile storage units 401 and 402. The frequency component comparison units 405 and 406 transmit the calculated calculation results for each frequency to the time change control units 407 and 408.

  The time change control units 407 and 408 smooth the division result for each frequency transmitted from the frequency component comparison units 405 and 406 by applying a low-pass filter (LPF) in the time direction for each frequency. The time change controllers 407 and 408 transmit the calculated determination results for each frequency to the subtraction coefficient calculators 409 and 410.

  The subtraction coefficient calculation units 409 and 410 calculate the subtraction coefficient for each frequency using the calculation result for each frequency transmitted from the time change control units 407 and 408. The table is such that the subtraction coefficient γ [n] gradually decreases as the LPF output level, which is the output of the time change controllers 407 and 408, increases. For example, a table as shown in FIG. When the LPF output level is sufficiently high, a sufficiently large frequency component of the desired voice is superimposed on the frequency component of the noise to be reduced, and the human hearing cannot substantially perceive the noise due to the masking effect. Therefore, by reducing the subtraction coefficient γ [n], the size of the profile that is subtracted from the input speech by the noise reduction units 413 and 414 can be reduced, and deterioration of the desired speech can be suppressed. The subtraction coefficient calculation units 409 and 410 transmit the calculated subtraction coefficient for each frequency to the multipliers 411 and 412.

  The multipliers 411 and 412 multiply the subtraction coefficient for each frequency transmitted from the subtraction coefficient calculation units 409 and 410 and the noise profile transmitted from the noise profile storage units 401 and 402 for each frequency. The calculation result for each frequency of the multipliers 411 and 412 is defined as a subtraction coefficient spectrum. The multipliers 411 and 412 transmit the calculated subtraction coefficient spectrum to the noise reduction units 413 and 414.

  The noise reduction units 413 and 414 are subtracters, and the frequency at which noise is reduced by subtracting the subtraction coefficient spectrum transmitted from the multipliers 411 and 412 from the frequency component transmitted from the Fourier transform units 403 and 404. Get every spectrum. The spectrum obtained by the noise reduction units 413 and 414 is defined as a noise reduction spectrum. The noise / noise reduction units 413 and 414 transmit the noise reduction spectrum to the inverse Fourier transform units 415 and 416.

  The inverse Fourier transform units 415 and 416 use the phase information transmitted from the Fourier transform units 403 and 404 for the noise reduction spectrum transmitted from the noise reduction units 413 and 414, and perform inverse Fourier transform (Fourier inverse transform). I do. That is, an inverse conversion process is performed to convert a frequency domain audio signal spectrum into a time domain audio signal. Then, a digital audio signal with reduced noise is obtained. The inverse Fourier transform units 415 and 416 transmit the restored digital audio signal after noise reduction to the signal output control units Lch and Rch.

  The wind noise reduction processing circuit and the noise reduction processing circuit using the SS method have been described above. Next, noise reduction processing by the voice processing unit 107 in this embodiment will be described.

  FIG. 5 is a diagram illustrating a circuit for noise reduction processing in the present embodiment. The noise reduction circuit 500 is included in the audio processing unit 107. The noise reduction circuit 500 performs drive noise reduction processing by the SS method. The driving noise is noise generated by driving the motor 102 in FIG. In addition, the same number is added about the block which is common in Fig.4 (a), and detailed description is abbreviate | omitted.

  The noise reduction circuit 500 receives a digital audio signal obtained by converting the Lch and Rch audio signals input from the audio input unit 106 into digital signals. Also, the Lch and Rch audio signals from the noise reduction circuit 500 are output to the wind noise processing circuit 300 of FIG.

  The noise reduction circuit 500 adjusts the correction amount of the noise profile in accordance with the magnitude of the wind noise component included in the input voice. In FIG. 5, the correction amount determination unit 511 detects low-frequency frequencies having different phases of the Lch and Rch audio signals using the audio data for each frequency band output from the fast Fourier transform circuits 503 and 504, and attenuates them. A correction value α for correcting the coefficient γ [n] is determined.

  FIG. 6A is a flowchart showing processing for determining the correction amount α of the attenuation coefficient γ [n] by the correction amount determination unit 511. The noise reduction circuit 500 adjusts the correction amount of the noise profile according to the magnitude of the wind noise component included in the input voice. For this reason, processing for detecting wind noise components included in the input speech is performed. In S601, the wind detection evaluation value is initialized. In S602, it is determined that the frequency band of the voice spectrum transmitted from the Fourier transform units 401 and 402 does not include a frequency band equal to or higher than the threshold thf. When the frequency band equal to or higher than the threshold thf is not included, the process proceeds to S603. If the frequency band equal to or higher than the threshold thf is included, the flow is terminated. The threshold value of the frequency of the voice spectrum at this time, thf, is set to about 1 kHz, which is a band where a lot of wind noise is concentrated in the voice signal.

  In S603, the phase characteristics of the audio spectrum of the stereo microphone that is the output of the Fourier transform units 401 and 402 are compared. As described above, when wind noise is included in the sound, the phase of the frequency is uncorrelated in the low frequency band of the sound, so if there is a phase difference equal to or greater than a certain threshold thφ, the process proceeds to S604. On the other hand, if the phase difference is equal to or smaller than the threshold thφ, it is determined that no wind noise is included, and the process proceeds to S602. At this time, the digital camera as the imaging apparatus 100 in FIG. 1 is considered to have a small distance between the Lch and Rch microphones. Normally, when the distance between the microphones is small, the sound entering the stereo microphone can be picked up with substantially the same phase at thf or less, which is the low frequency band.

  In S604, the amplitude of the audio spectrum of the stereo microphone that is the output of the Fourier transform units 401 and 402 is compared. The sound spectrum including wind noise has a large amplitude. Therefore, if the amplitudes of the left and right audio spectra are equal to or greater than a certain threshold thA, it is determined that wind noise is included, and the process proceeds to 605. If the amplitude is equal to or less than a certain threshold thA, the process proceeds to S602.

  In S605, if the conditions from S602 to S604 are met, wind noise is included in the speech spectrum, and the wind detection evaluation value is incremented by one. In S606, if the wind detection evaluation value is equal to or greater than a certain threshold thn, the process proceeds to S607. If it is equal to or less than the threshold value thn, the process proceeds to S602. In S607, for example, the correction amount α of the attenuation coefficient γ with respect to the wind detection evaluation value as shown in FIG. The correction amount α is determined with reference to the wind detection evaluation value obtained in S606, and the flow ends. If the evaluation value for wind detection is large, the correction amount α is also increased. At this time, in FIG. 7A, when the wind detection evaluation value is equal to or greater than a certain value, the correction amount α is set to be constant.

  The attenuation coefficient correction units 512 and 513 correct the attenuation coefficient γ [n] calculated by the attenuation coefficient calculation unit 509 using the correction amount α obtained by the correction amount determination unit 511 as described above. The multipliers 411 and 412 multiply the subtraction coefficient for each frequency output from the subtraction coefficient correction units 512 and 513 and the noise profile transmitted from the noise profile storage units 401 and 402 for each frequency. Thereby, a noise reduction value for each frequency based on the noise profile is determined.

  FIG. 7B shows a graph in which the attenuation coefficient γ [n] is corrected by the correction amount α. In FIG. 7B, the correction amount α obtained by the correction amount determination unit 511 is added to the attenuation coefficient γ for each audio level. Here, the correction amount α is added as the correction method, but the correction amount α may be changed according to the situation.

  Thus, in the present embodiment, when wind noise is included, the attenuation coefficient of the noise profile is corrected according to the magnitude. Specifically, the attenuation coefficient correction amount is determined such that the larger the wind noise is, the larger the profile is. Therefore, when it is determined that wind noise is included, the noise profile component subtracted from the input voice is corrected to be larger. By adding the correction amount α to the attenuation coefficient γ [n], it is possible to solve the problem that the drive sound remains in a low frequency during the drive noise reduction process using the SS method.

(Example 2)
In this embodiment, a description will be given of a method for correcting the attenuation coefficient γ. Since the configuration of the noise reduction circuit of this embodiment has the same configuration as that of the first embodiment, description thereof is omitted. However, since the configuration of the correction amount determination unit 511 is different from that of the first embodiment, details will be described below.

  FIG. 8 is a flowchart for explaining the processing of the correction amount determination unit 511 in the present embodiment. FIG. 9 shows a graph in which the attenuation coefficient γ [n] is corrected by the correction amount α. In step S801, the correction amount determination unit 511 determines whether the level of the audio signal is greater than or equal to a threshold value. In S802, if the level of the audio signal is equal to or higher than the threshold value, the correction amount α is added to the attenuation coefficient γ according to the level. Further, after correction as shown in FIG. 9, the value of the attenuation coefficient of 1.0 or more is set to 1.0.

  In S803, if the audio signal is equal to or smaller than the threshold value, the correction amount α is not set, and the attenuation coefficient correction units 510 and 511 do not correct the attenuation coefficient γ. In step S804, the profile is updated by applying the attenuation coefficient γ [n] corrected in steps S802 and S803 to the profiles 501 and 502.

  In this embodiment, when the level of the audio signal obtained from the stereo microphone is smaller than the threshold value, the correction amount α is not added even if wind noise enters. If the audio signal is at a level equal to or higher than the threshold, the correction amount α is added.

  Thus, by determining whether or not to add the correction amount α in accordance with the level of the audio signal, it is possible to reduce drive noise and wind noise so that the correction is not excessive when the level of the audio signal is small. .

(Example 3)
The noise reduction circuit 1000 in the present embodiment will be described with reference to FIGS. The configuration of the noise reduction circuit 1000 according to the present embodiment is different from the configuration of the noise reduction device in FIG.

  FIG. 10 is a diagram illustrating a configuration of the noise reduction circuit 1000 according to the present embodiment. FIG. 11 is a conceptual diagram for generating a profile for performing drive noise reduction processing by the SS method. The noise profile generation units 1001 and 1002 store noise profiles indicating frequency components of noise to be reduced. Specifically, in FIG. 11, the audio signal composed of the audio 1101 before the noise to be reduced and the noise 1102 to be reduced are Fourier-transformed, and the frequency component Rt1 [i] of the audio signal 1101 and the frequency of the audio signal 1102 The component Rt2 [i] is obtained. At this time, the frequency components Rt1 [i] and Rt2 [i] are scalar quantities of amplitude. Further, i is the number of frequency samples, and if the number of audio data samples is N according to the sampling theorem, i is N / 2. That is, when the number of audio data samples N = 1024, the frequency sample number is i = 512.

By subtracting the frequency component (Rt1) of the voice not including the noise at time t1 from the noise frequency component (Rt2) at time t2, the frequency component of noise alone, that is, the noise profile Rp is calculated.
Rp [i] = Rt2 [i] −Rt1 [i] Formula (1)

  The calculated frequency component Rp of only noise to be reduced is stored as a noise profile, and the noise reduction units 413 and 414 perform noise reduction processing.

  In this manner, by generating a noise profile each time noise is generated, reduction processing suitable for each noise can be performed.

Claims (7)

Input means;
Driving means;
A conversion unit that divides the time-domain audio signal obtained by the input unit at regular intervals and converts the signal into a frequency-domain audio signal spectrum;
Storage means for storing a profile relating to amplitude for each frequency of driving noise, which is noise generated by the driving means;
Calculation means for calculating a subtraction coefficient based on the signal output from the conversion means;
Correction amount determination means for determining the correction amount of the subtraction coefficient according to the magnitude of wind noise included in the audio signal obtained from the input means;
Correction means for correcting the subtraction coefficient obtained by the calculation means with the correction amount obtained by the correction amount determination means;
A determining unit that determines a noise reduction value for each frequency from the subtraction coefficient corrected by the correcting unit and the profile stored in the storage unit;
Noise reduction means for reducing the drive noise by subtracting the reduction value obtained by the determination means from the spectrum of the frequency domain audio signal output from the conversion means;
Inverse transform means for transforming the frequency domain speech signal spectrum from the noise reduction means into a time domain speech signal;
Wind noise reducing means for reducing wind noise from the audio signal output from the inverse transform means;
A speech processing apparatus comprising:   2. The correction amount determination unit according to claim 1, wherein the correction amount determination unit determines the correction amount so that the reduction value increases as the wind signal contained in the audio signal obtained from the input unit increases. Voice processing device.   The correction amount determination unit obtains an evaluation value of wind noise from an amplitude characteristic and a phase characteristic of an audio signal spectrum output from the conversion unit, and determines the correction amount based on the evaluation value. Item 6. The speech processing apparatus according to Item 1. The input means inputs a stereo audio signal,
The correction amount determination unit obtains the evaluation value according to a phase difference and an amplitude magnitude in the audio signal spectrum of the left channel and the right channel of the stereo audio signal output from the conversion unit. The speech processing apparatus according to claim 3.   The audio processing apparatus according to claim 1, wherein the correction unit corrects the correction amount determined by the correction amount determination unit by adding the correction amount to a subtraction coefficient for each audio level.   The speech processing apparatus according to claim 1, wherein the calculation unit decreases the subtraction coefficient as the time change of the result of dividing the profile by the speech signal spectrum output from the conversion unit increases.   The calculation means obtains a ratio between the audio signal spectrum output from the conversion means and the profile for each frequency, and based on a result of smoothing the ratio for each frequency for each frequency with respect to a fixed time. The audio processing apparatus according to claim 6, wherein the subtraction coefficient is calculated.

Images