JP2013168856A - Noise reduction device, audio input device, radio communication device, noise reduction method and noise reduction program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a noise reduction device capable of suppressing reduction of sound pressure level.SOLUTION: A noise reduction device includes: an audio interval determination unit 11 which determines the audio interval on the basis of a sound collection signal 21; a noise reduction processing unit 12 which reduces noise components contained in the sound collection signal 21 using the sound collection signal 22; a sound pressure level variation amount calculation unit 13 which calculates the variation amount of the sound pressure level of a signal 25 after noise reduction processing with respect to the sound collection signal 21 using the sound collection signal 21 and the signal 25 after noise reduction processing in the audio interval; and a sound pressure level compensation unit 14 which compensates the sound pressure level of the signal 25 after noise reduction processing in accordance with the variation amount calculated by the sound pressure level variation amount calculation unit 13.

Description

  The present invention relates to a noise reduction device, a voice input device, a wireless communication device, a noise reduction method, and a noise reduction program.

  There is a noise reduction processing technique that makes it easy to hear a sound by reducing a noise component included in the sound signal. In the noise reduction processing technology, for example, noise (for example, unnecessary sound other than the desired sound) is mainly obtained from a sound signal collected by a microphone that mainly collects sound (for example, desired sound such as a sound emitted from a caller). By subtracting the noise signal (reference signal) collected by the microphone that picks up the sound, the noise component contained in the audio signal can be removed.

  Patent Document 1 discloses a technique for preventing reduction of desired speech and reducing only unnecessary sound to be reduced. Patent Document 2 discloses a technique for improving the intelligibility of speech or the like lowered by an adaptive filter for noise removal. Patent Document 3 discloses a technology relating to a noise canceller that can obtain an optimal noise reduction effect in a timely manner according to the noise situation.

JP-A-6-67692 JP-A-8-102644 JP-A-9-36763

  When noise reduction processing is performed using an audio signal mainly including an audio component and a reference signal mainly including a noise component, the audio component may be mixed into the reference signal depending on the use state of the noise reduction apparatus. As described above, when the sound component is mixed in the reference signal, the sound component included in the sound signal is canceled when the noise reduction process is performed, and the sound pressure level of the signal after the noise reduction process is lowered. there were.

  In view of the above problems, an object of the present invention is to provide a noise reduction device, a voice input device, a wireless communication device, a noise reduction method, and a noise reduction program that can suppress a decrease in sound pressure level.

  The noise reduction device according to the present invention includes a speech section determination unit that determines a speech section based on a first sound collection signal, and a noise component included in the first sound collection signal using the second sound collection signal. The noise reduction processing unit to be reduced, and the first sound collection signal and the signal after noise reduction processing output from the noise reduction processing unit in the voice section, the first sound collection signal with respect to the first sound collection signal A sound pressure level change amount calculating unit for calculating a change amount of the sound pressure level of the signal after the noise reduction process; and a signal of the signal after the noise reduction process according to the change amount calculated by the sound pressure level change amount calculating unit. A sound pressure level compensator for compensating the sound pressure level.

  The sound pressure level compensation unit has an absolute value of a sound pressure level difference, which is a difference between a sound pressure level of the first collected sound signal and a sound pressure level of the signal after the noise reduction processing, equal to or greater than a predetermined threshold. The sound pressure level of the signal after the noise reduction processing may be compensated.

  The sound pressure level compensation unit may amplify the signal after the noise reduction processing at an amplification factor corresponding to the sound pressure level difference.

  The sound pressure level compensation unit may gradually reduce the amplification factor after amplifying the signal after the noise reduction processing with an amplification factor corresponding to the sound pressure level difference.

  When the sound pressure level difference exceeds a predetermined upper limit value, the sound pressure level compensation unit may amplify the signal after the noise reduction processing with an amplification factor corresponding to the upper limit value.

  The speech segment determination unit may determine that the speech segment is a speech segment when a probability that a speech component is included in the first sound collection signal is equal to or greater than a predetermined value.

  The voice section determination unit has a ratio between a peak of a vowel frequency component of a voice component included in the first sound pickup signal and a noise level set for each band being equal to or greater than a predetermined value, and If the number of peaks greater than or equal to the value is equal to or greater than a predetermined number, it may be determined that the voice segment is present.

  The speech section determination unit measures a consonant spectrum pattern of a speech component included in the first collected sound signal for each predetermined frequency band, and a speech section when the consonant spectrum pattern increases as the frequency band increases It may be determined that

  The noise reduction processing unit may include an adaptive filter that generates a pseudo noise signal corresponding to a noise component included in the first sound collection signal using the second sound collection signal.

  The voice input device according to the present invention includes the noise reduction device. In the voice input device, the first microphone is provided on the first surface of the voice input device, and the second microphone is opposed to the first surface with a predetermined distance from the second surface. May be provided.

  A wireless communication device according to the present invention includes the noise reduction device. In the wireless communication device, the first microphone is provided on the first surface of the wireless communication device, and the second microphone is opposed to the first surface with a predetermined distance from the second surface. May be provided.

  The noise reduction method according to the present invention determines a speech section based on a first sound collection signal, reduces a noise component included in the first sound collection signal using a second sound collection signal, and In the section, using the first collected sound signal and the signal after the noise reduction process, a change amount of a sound pressure level of the signal after the noise reduction process with respect to the first sound collection signal is calculated, and the calculation The sound pressure level of the signal after the noise reduction processing is compensated according to the amount of change.

  A noise reduction program according to the present invention causes a computer to determine a speech section based on a first sound collection signal, and to reduce a noise component included in the first sound collection signal using a second sound collection signal. In the voice section, the amount of change in the sound pressure level of the signal after the noise reduction processing with respect to the first sound collection signal is calculated using the first sound collection signal and the signal after noise reduction processing. A noise reduction program for compensating a sound pressure level of the signal after the noise reduction processing according to the calculated change amount.

  According to the present invention, it is possible to provide a noise reduction device, a voice input device, a wireless communication device, a noise reduction method, and a noise reduction program that can suppress a decrease in sound pressure level.

It is a block diagram which shows the noise reduction apparatus concerning embodiment. It is a block diagram which shows an example of the audio | voice area determination part with which the noise reduction apparatus concerning embodiment is provided. It is a block diagram which shows the other example of the audio | voice area determination part with which the noise reduction apparatus concerning embodiment is provided. It is a block diagram which shows an example of the noise reduction process part with which the noise reduction apparatus concerning embodiment is provided. It is a figure for demonstrating in detail the noise reduction process part shown in FIG. It is a block diagram which shows an example of the sound pressure level variation | change_quantity calculation part with which the noise reduction apparatus concerning embodiment is provided. It is a figure for demonstrating an example of operation | movement of the noise reduction apparatus concerning embodiment. It is a block diagram which shows the other example of the noise reduction apparatus concerning embodiment. It is a figure which shows an example of the audio | voice input apparatus using the noise reduction apparatus concerning embodiment. It is a figure which shows an example of the radio | wireless communication apparatus using the noise reduction apparatus concerning embodiment.

Embodiments of the present invention will be described below with reference to the drawings.
FIG. 1 is a block diagram illustrating a noise reduction device according to an embodiment. As illustrated in FIG. 1, the noise reduction device 1 according to the present embodiment includes a speech section determination unit 11, a noise reduction processing unit 12, a sound pressure level change amount calculation unit 13, and a sound pressure level compensation unit 14.

  The noise reduction apparatus 1 according to the present embodiment receives a first sound pickup signal (speech signal) 21 mainly including a sound component and a second sound pickup signal (reference signal) 22 mainly including a noise component, and collects the sound. Noise reduction processing is performed using the sound signal 21 and the collected sound signal 22, and the signal after the noise reduction processing is output as an output signal 27. For example, the sound collection signal 21 and the sound collection signal 22 can be collected by using the sound microphone 16 and the reference sound microphone 17 as in the noise reduction device 1 ′ shown in FIG.

  The voice microphone 16 shown in FIG. 8 picks up sound mainly including a voice component, converts it into an analog signal, and outputs the converted analog signal to the AD converter 18. The reference sound microphone 17 collects a sound mainly including a noise component, converts it into an analog signal, and outputs the converted analog signal to the AD converter 19. The noise component included in the sound collected by the reference sound microphone 17 is used to reduce the noise component contained in the sound collected by the sound microphone 16.

  In addition, although noise reduction apparatus 1 'shown in FIG. 8 has shown the structure provided with two microphones, you may provide the microphone for reference sounds further, and may provide three or more microphones, for example. That is, you may comprise so that three or more sound collection signals may be input into the noise reduction apparatus 1 shown in FIG.

  The AD converter 18 samples the analog signal output from the audio microphone 16 at a predetermined sampling rate and converts it into a digital signal, thereby generating a sound collection signal 21. The AD converter 19 samples the analog signal output from the reference sound microphone 17 at a predetermined sampling rate and converts it into a digital signal, thereby generating a sound collection signal 22.

  For example, the frequency band of the sound input to the sound microphone 16 and the reference sound microphone 17 is approximately 100 Hz to 4000 Hz. Therefore, by setting the sampling frequency in the AD converters 18 and 19 to about 8 kHz to 12 kHz, an analog signal including an audio component can be handled as a digital signal.

  As shown in FIG. 1, the collected sound signal 21 is supplied to the speech section determination unit 11, the noise reduction processing unit 12, and the sound pressure level change amount calculation unit 13. The collected sound signal 22 is supplied to the noise reduction processing unit 12. In the present specification, the collected sound signal 21 mainly including a sound component is also referred to as a sound signal, and the collected sound signal 22 mainly including a noise component is also referred to as a reference signal (noise signal).

  The voice segment determination unit 11 determines a voice segment based on the supplied sound collection signal 21. Then, the speech segment determination unit 11 outputs speech segment information 23 and 24 indicating the speech segment to the noise reduction processing unit 12 and the sound pressure level change amount calculation unit 13, respectively.

  An arbitrary technique can be used for the speech segment determination processing in the speech segment determination unit 11. When the noise reduction device is used in an environment where the noise level is high, it is preferable to determine the speech section and the noise section with high accuracy. For example, the speech noise section detection technique A or the speech noise section detection technique described later is used. By using B, it is possible to detect the voice section and the noise section with high accuracy. The sound includes sounds other than human voices, but in these examples, human voices are mainly detected. Note that the speech noise section detection technique A is also described as an example in Japanese Patent Application No. 2011-254578, which is an application claiming priority based on Japanese Patent Application No. 2010-260798. The voice noise section detection technique B is also described in Japanese Patent Application No. 2011-020659 as an example.

  First, the speech segment determination technique A will be described. In the speech section determination technique A, the speech section is determined by paying attention to the frequency spectrum of the vowel component that is the main part of the speech. In the speech section determination technique A, an appropriate noise level is set for each band, a signal-to-noise level ratio with a peak of the vowel frequency component is obtained, and the signal-to-noise level ratio is a predetermined level ratio and a predetermined number of peaks. The voice section is determined by observing whether or not.

  FIG. 2 is a block diagram illustrating an example of a speech segment determination unit 11 ′ using the speech segment determination technique A. 2 includes a framing unit 31, a spectrum generating unit 32, a band dividing unit 33, a frequency averaging unit 34, a holding unit 35, a time averaging unit 36, a peak detecting unit 37, and a voice determining unit. 38.

  The framing unit 31 sequentially cuts the sound pickup signal 21 in frame units (predetermined number of samples) having a predetermined time width, and generates an input signal in frame units (hereinafter referred to as a framed input signal).

  The spectrum generation unit 32 performs frequency analysis of the framing input signal output from the framing unit 31, converts the time-domain framing input signal into the frequency-domain framing input signal, and collects the spectrum. Is generated. The spectrum pattern is a collection of spectra for each frequency in which a frequency and energy at the frequency are associated with each other over a predetermined frequency band. The frequency transform method used here is not limited to a specific means, but requires a frequency resolution necessary for recognizing the spectrum of speech, and therefore has a relatively high resolution such as FFT (Fast Fourier Transform) or DCT (Discrete). It is recommended to use an orthogonal transformation method such as Cosine Transform. In the present embodiment, the spectrum generation unit 32 generates a spectrum pattern of at least 200 Hz to 700 Hz.

  A spectrum (hereinafter referred to as a formant) that indicates a feature of a voice, which is a target to be detected when a voice determination unit 38 to be described later determines a voice section, usually includes a harmonic part from a first formant corresponding to a fundamental tone. There are a plurality of nth formants (where n is a natural number). Of these, the first formant and the second formant often exist in a frequency band of less than 200 Hz. However, since this band contains a low-frequency noise component with relatively high energy, formants are easily buried. Also, a formant of 700 Hz or more is easily buried in a noise component because the formant itself has low energy. Therefore, by using a spectrum pattern of 200 Hz to 700 Hz that is difficult to be buried in the noise component for the determination of the voice section, the determination target can be narrowed down and the voice section can be determined efficiently.

  In order to detect a spectrum characteristic of speech in an appropriate frequency band unit, the band dividing unit 33 divides each spectrum of the spectrum pattern into a plurality of divided frequency bands that are frequency bands divided by a predetermined bandwidth. To divide. In the present embodiment, the predetermined bandwidth is about 100 Hz to 150 Hz.

  The frequency averaging unit 34 calculates average energy for each divided frequency band. In the present embodiment, the frequency averaging unit 34 averages the energy of all spectra in the divided frequency band for each divided frequency band. However, the maximum or average amplitude value of the spectrum is used instead of the spectrum energy in order to reduce the calculation load. (Absolute value) may be substituted.

  The holding unit 35 is configured by a storage medium such as a RAM (Random Access Memory), an EEPROM (Electrically Erasable and Programmable Read Only Memory), and a flash memory, and the average energy for each band is set to a predetermined number in the past (this embodiment). N frames in the form) are held.

  The time averaging unit 36 derives, for each divided frequency band, band-specific energy that is an average over a plurality of frames in the time direction of the average energy derived by the frequency averaging unit 34. That is, the band-specific energy is an average value over a plurality of frames in the time direction of the average energy for each divided frequency band. In addition, the time averaging unit 36 may obtain a substitute value of the band-specific energy by performing a process according to averaging using the weighting coefficient and the time constant on the average energy for each divided frequency band of the immediately preceding frame.

  The peak detector 37 derives an energy ratio (SNR: Signal to Noise ratio) between each spectrum of the spectrum pattern and the band-specific energy in the divided frequency band in which the spectrum is included. Then, the peak detection unit 37 compares the SNR for each spectrum with a predetermined first threshold value, and determines whether or not the first threshold value is exceeded. If there is a spectrum whose SNR exceeds the first threshold value, this spectrum is regarded as a formant, and information indicating that a formant has been detected is output to the voice determination unit 38.

  When receiving information from the peak detection unit 37 that the formant has been detected, the audio determination unit 38 determines whether the framed input signal of the corresponding frame is audio based on the determination result of the peak detection unit 37. When it is determined that the framed input signal is speech, the speech determination unit 38 outputs the speech section information 23 and 24 to the noise reduction processing unit 12 and the sound pressure level change amount calculation unit 13, respectively.

  The speech section determination unit 11 ′ illustrated in FIG. 2 sets energy for each divided frequency band for each divided frequency band. Therefore, the voice determination unit 38 can accurately determine the presence / absence of a formant for each divided frequency band without being affected by noise components in other divided frequency bands.

  As described above, there are a plurality of formants from the first formant to the n-th formant, which is a harmonic part thereof. Therefore, even if the energy (noise level) of any divided frequency band is increased and a part of the formant is buried in noise, a plurality of other formants may be detected. In particular, since ambient noise is concentrated in the low range, even if the first formant corresponding to the fundamental tone and the second formant corresponding to the second overtone are buried in the low-frequency noise, the possibility of detecting a formant with a third or higher harmonic is possible. There is. Therefore, when the spectrum whose SNR exceeds the first threshold is greater than or equal to the predetermined number, the speech determination unit 38 can determine a speech section that is more resistant to noise by determining that the framed input signal is speech. it can.

  As described above, the speech segment determination unit 11 ′ using the speech segment determination technique A cuts out the input signal in units of frames having a predetermined time width, and generates a framed input signal. A spectrum generation unit 32 for converting the framing input signal from the time domain to the frequency domain to generate a spectrum pattern in which spectra for each frequency are collected, each spectrum of the spectrum pattern, and a predetermined bandwidth A peak detector 37 that determines whether or not the energy ratio of the divided frequency bands including the spectrum among the plurality of divided frequency bands that are the divided frequency bands to the energy by band exceeds a predetermined first threshold value. And a voice determination unit 38 that determines whether or not the framed input signal is voice based on the determination result of the peak detection unit, A frequency averaging unit 34 for deriving an average energy in the frequency direction of the spectrum in each divided frequency band of the spectrum pattern, and a time averaging unit 36 for deriving the energy by band that is an average of the average energy in the time direction for each divided frequency band. And comprising.

  For example, the speech determination unit 38 can determine that the framed input signal is speech when the spectrum in which the energy ratio exceeds the first threshold is equal to or greater than a predetermined number.

  Next, the speech section determination technique B will be described. In the speech section determination technique B, the speech section is determined by paying attention to the property that the spectrum pattern that is a feature of the consonant tends to rise to the right. In the speech segment determination technique B, the spectrum pattern of the consonant is measured in the mid-high frequency band, and the characteristics of the frequency distribution of the consonant that is partially buried by the noise component are set in a band where there is not much influence of noise. By specializing and extracting, it is possible to determine the speech section with high accuracy.

  FIG. 3 is a block diagram illustrating an example of a speech segment determination unit 11 ″ using the speech segment determination technique B. The speech section determination unit 11 ″ includes a framing unit 41, a spectrum generation unit 42, a band division unit 43, an average derivation unit 44, a noise level derivation unit 45, a determination selection unit 46, and a consonant determination unit 47.

  The framing unit 41 sequentially extracts the sound pickup signal 21 in units of frames having a predetermined time width, and generates a framing input signal that is an input signal in units of frames.

  The spectrum generation unit 42 performs frequency analysis of the framing input signal output from the framing unit 41, converts the time-domain framing input signal into the frequency-domain framing input signal, and collects the spectrum. Is generated. The spectrum pattern is a collection of spectra for each frequency in which a frequency and energy at the frequency are associated with each other over a predetermined frequency band. The frequency conversion method used here is not limited to a specific means, but a frequency resolution necessary for recognizing a speech spectrum is necessary, and therefore, an orthogonal transformation method such as FFT or DCT having a relatively high resolution is used. Good.

  The band dividing unit 43 divides each spectrum of the spectrum pattern generated by the spectrum generating unit 42 for each predetermined bandwidth, and generates a plurality of divided frequency bands. In the present embodiment, the band dividing unit 43 divides the frequency range of, for example, 800 Hz to 3.5 kHz for each bandwidth of about 100 Hz to 300 Hz, for example.

  The average deriving unit 44 derives average energy for each band, which is an average energy for each divided frequency band (band) divided by the band dividing unit 43 in the spectrum pattern.

  The consonant determination unit 47 compares the band-by-band average energies derived by the average deriving unit 44. If the band-by-band average energy of the higher frequency band is higher, the consonant is included in the framed input signal. It is determined that

  In general, consonants tend to have a spectral pattern that rises to the right. Therefore, the speech segment determination unit 11 ″ using the speech segment determination technique B derives the average energy for each band in the spectrum pattern and compares the energy for each band to the right in the spectrum pattern characteristic of the consonant. Detect upward trend. Therefore, the speech segment determination unit 11 '' can accurately detect a consonant segment in which a consonant is included in the input signal.

  The consonant determination unit 47 counts a combination in which the average energy for each band between adjacent bands is higher in the high frequency band than in the adjacent low frequency band, and the counted value is a predetermined first threshold value. If it is above, the 1st judgment means which judges that a consonant is contained is provided. In addition, the consonant determination unit 47 measures a combination in which the average energy for each band between adjacent bands is higher in the high frequency band than in the adjacent low frequency band, and when this combination continues across the bands And a second determination means for determining that a consonant is included when the counted value is equal to or greater than a predetermined second threshold value. The consonant determination unit 47 uses the first determination unit and the second determination unit in accordance with the noise level.

  Here, the noise level deriving unit 45 derives the noise level of the framed input signal so as to select the first determination unit and the second determination unit as appropriate. For example, the noise level can be an average value of average energy for each frequency band of the framed input signal. Further, the noise level deriving unit 45 may derive a noise level for each framed input signal, or may use an average value of noise levels of the framed input signal for a predetermined time. The determination selection unit 46 selects the first determination unit when the derived noise level is less than the predetermined threshold, and selects the second determination unit when the derived noise level is equal to or higher than the predetermined threshold.

  As described above, the speech segment determination unit 11 ″ using the speech segment determination technique B includes the framing unit 41 that cuts out the input signal in units of predetermined frames and generates a framed input signal, The spectrum generation unit 42 that converts the input signal from the time domain to the frequency domain and generates a spectrum pattern in which the spectrum for each frequency is collected, and the average energy for each predetermined bandwidth to be connected in the spectrum pattern The average deriving unit 44 for deriving the average energy for each band and the derived average energy for each band are compared. If the average energy for each band in the higher frequency band is higher, the framed input signal A consonant determination unit 47 that determines that a consonant is included.

  For example, the consonant determination unit 47 counts combinations in which the average energy for each band between adjacent bands of the spectrum pattern is larger in the higher frequency band than in the adjacent lower frequency band, and the counted value is determined in advance. It is possible to determine that a consonant is included if it is equal to or greater than the threshold value.

  In addition, when applying said audio | voice area determination technique A and B to the noise reduction apparatus concerning this Embodiment, a parameter can be set for every product. That is, when the speech segment determination techniques A and B are applied to a product that requires more reliable speech segment determination, a stricter threshold can be set as a parameter for speech segment determination.

  The noise reduction processing unit 12 included in the noise reduction device 1 illustrated in FIG. 1 performs noise reduction processing using at least two collected sound signals 21 and 22. That is, the noise reduction processing unit 12 uses the sound collection signal 22 mainly including the noise component to reduce the noise component included in the sound collection signal 21 mainly including the sound component. Thus, by reducing the noise component included in the collected sound signal 21, it is possible to improve the ease of listening to the voice.

  FIG. 4 is a block diagram illustrating an example of the noise reduction processing unit 12 included in the noise reduction device 1 according to the present embodiment. The noise reduction processing unit 12 illustrated in FIG. 4 includes an adaptive filter 51, an adaptive coefficient adjustment unit 52, and an adder 53.

  The adaptive filter 51 receives a sound collection signal 22 mainly including a noise component, and uses this sound collection signal 22 to artificially generate a noise component that may be included in the sound collection signal 21. The noise signal 55 is output. Here, the pseudo noise signal 55 is a signal whose phase is inverted with respect to the collected sound signal 21.

  The adder 53 adds the sound pickup signal 21 and the phase-inverted pseudo noise signal 55 to generate a signal 25 after noise reduction processing. Further, the adder 53 adds the sound pickup signal 21 and the pseudo-noise signal 55 whose phase has been inverted to generate a feedback signal 56 and outputs it to the adaptive coefficient adjustment unit 52.

  The adaptive coefficient adjustment unit 52 adjusts the coefficient of the adaptive filter 51 in accordance with the speech segment information 23. That is, the adaptive coefficient adjustment unit 52 adjusts the coefficient so that the adaptive error is reduced when the speech section information 23 does not indicate a speech section (in the case of a noise section). On the other hand, when the speech section information 23 indicates a speech section, the coefficient of the adaptive filter 51 is maintained or only the coefficient is finely adjusted.

  FIG. 5 is a diagram for explaining the noise reduction processing unit 12 illustrated in FIG. 4 in detail. FIG. 5 shows an example in which the adaptive filter 51 is configured by an FIR (Finite Impulse Response) filter. The adaptive filter 51 illustrated in FIG. 5 includes delay elements 61_1 to 61_n, multipliers 62_1 to 62_n + 1, and adders 63_1 to 63_n. The pseudo noise signal 55 is generated by processing the sound collection signal 22 using the delay elements 61_1 to 61_n, the multipliers 62_1 to 62_n + 1, and the adders 63_1 to 63_n.

  The adaptive coefficient adjustment unit 52 adjusts the coefficients of the multipliers 62_1 to 62_n + 1. That is, the adaptive coefficient adjustment unit 52 minimizes the difference (feedback signal 56) between the pseudo noise signal 55 and the collected sound signal 21 when the speech section information 23 does not indicate a speech section (in the case of a noise section). The coefficient of the adaptive filter 51 is adjusted. Thereby, the pseudo noise signal 55 output from the adaptive filter 51 can be brought close to the noise component included in the sound pickup signal 21 picked up by the sound microphone.

  On the other hand, when the voice section information 23 indicates a voice section, the collected sound signal 21 includes a voice component. In this case, the coefficient of the adaptive filter 51 may not be adapted to the noise component and may not converge due to the influence of the audio component. Therefore, in order to stably update the coefficient of the adaptive filter 51, when the speech section information 23 indicates a speech section, the coefficient of the adaptive filter 51 is maintained or only the coefficient is finely adjusted. It is desirable.

  The sound pressure level change amount calculation unit 13 included in the noise reduction device 1 illustrated in FIG. 1 performs a sound collection signal 21 and noise reduction processing when the speech section information 24 output from the speech section determination unit 11 indicates a speech section. Using the noise-reduced signal 25 output from the unit 12, the amount of change in the sound pressure level of the noise-reduced signal 25 with respect to the collected sound signal 21 is calculated. The sound pressure level change amount 26 calculated by the sound pressure level change amount calculation unit 13 is output to the sound pressure level compensation unit 14.

  That is, the sound pressure level change amount calculation unit 13 determines the sound pressure level of the collected sound signal 21 to determine whether the quality of the signal 25 after noise reduction processing output from the noise reduction processing unit 12 is appropriate. The sound pressure level of the signal 25 after the noise reduction processing is compared. Then, the sound pressure level change amount calculation unit 13 calculates a sound pressure level difference that is a difference between the sound pressure level of the collected sound signal 21 and the sound pressure level of the signal 25 after the noise reduction processing, and this sound pressure level difference. Can be output as the sound pressure level change amount 26.

  The adaptive filter used in the noise reduction processing unit 12 derives an acoustic spatial characteristic filter of a noise component mixed in the voice microphone as seen from the reference sound microphone. The adaptive filter has a function of attenuating a signal component coming from a direction where a main noise source exists. The direction of noise arrival is in all three-dimensional directions centered on the position of the voice microphone, and the action of the adaptive filter is the same in all directions. Therefore, when a noise component comes from behind the speaker when the speaker is speaking into the voice microphone, the signal component (including the voice component and the noise component) in the noise arrival direction is canceled. For this reason, the audio component is also canceled. Further, even when the noise arrival directions are different, if the approach angle of the speaker's voice to the speech microphone is close to the approach angle of the noise component to the speech microphone, the noise reduction processing unit 12 may reduce the noise component. Since the sound component is also canceled when canceling, the sound pressure level of the signal 25 after the noise reduction processing is lowered.

  In the noise reduction processing unit 12 shown in FIGS. 4 and 5, the noise reduction processing is performed by adding the pseudo noise signal 55 whose phase is inverted with respect to the sound collection signal 21 to the sound collection signal 21. Here, since the pseudo noise signal 55 depends on the accuracy of the coefficient of the adaptive filter 51 and the like, the sound pressure level of the signal 25 after noise reduction processing is the sound pressure level of the collected sound signal 21 that is the original sound signal. It does not match. However, when the canceling action for the sound component is slight, a large decrease in the sound pressure level does not occur in the signal 25 after the noise reduction processing. That is, a difference in sound pressure level between the collected sound signal 21 and the signal 25 after the noise reduction processing does not occur or is minimal even if it occurs. In the noise reduction apparatus according to the present embodiment, the sound level change amount calculation unit 13 calculates the sound pressure level difference between the collected sound signal 21 before the noise reduction process and the signal 25 after the noise reduction process, and this sound pressure level. By comparing the difference with a predetermined threshold, it is possible to monitor the situation in which the audio component is canceled in the noise reduction processing unit 12.

  At this time, in order for the sound level change amount calculation unit 13 to accurately calculate the sound pressure level change amount (sound pressure level difference), it is necessary to calculate the sound pressure level difference only in the section where the sound is being emitted. . Therefore, when a voice segment is output from the voice segment determination unit 11 as a voice segment, a voice segment is output from the voice segment determination unit 11 as a voice segment. The level change amount calculation unit 13 calculates the sound pressure level difference. Here, the sound pressure level difference calculated by the sound level change amount calculation unit 13 is a reference value (sound level) when the sound pressure level compensation unit 14 compensates (adjusts) the sound pressure level of the signal 25 after the noise reduction processing. Pressure compensation level reference value).

  Usually, when a speaker utters a voice, it becomes intermittent because there is a break of a word or timing of breathing. In such a case, the timing at which the speech segment determination unit 11 determines that the segment is a speech segment is intermittent, and the speech segment information 23 and 24 indicating the speech segment is also discrete. In the case of voice, the level of the sound pressure level increases when viewed locally (single tone unit). However, when viewed globally (between phrase units), it is natural to think that a constant sound pressure level is maintained. This situation is the same for environmental noise. Therefore, since the characteristics of the noise reduction processing unit (adaptive filter) 12 can also be regarded as a gradual change, the sound pressure compensation level reference value (corresponding to the sound pressure level difference) obtained discretely is It is preferable to hold (do not update) until it is determined by the speech section determination unit 11 that the speech section is a speech section.

  The speech segment determination in the speech segment determination unit 11 detects whether or not the speech segment is a speech using a signal having a certain time width. Therefore, the information related to the sound pressure level difference can also be calculated in units of time width similar to that in the case of voice segment determination. For example, the sound pressure level difference between the collected sound signal 21 and the noise-reduced signal 25 can be calculated using the power amount in the unit time width.

  FIG. 6 is a block diagram illustrating an example of the sound pressure level change amount calculation unit 13 provided in the noise reduction device according to the present embodiment. The sound pressure level change amount calculation unit 13 illustrated in FIG. 6 includes a signal buffer 71, a signal power calculation unit 72, a signal buffer 73, a signal power calculation unit 74, and a sound pressure level difference calculation unit 75. The sound pressure level change amount calculation unit 13 illustrated in FIG. 6 can calculate the sound pressure level difference between the sound collection signal 21 and the signal 25 after noise reduction processing in a certain unit time. In addition, the sound pressure level change amount calculation unit 13 calculates the sound pressure level difference at a timing at which the speech segment information 24 output from the speech segment determination unit 11 indicates a speech segment.

  The signal buffer 71 temporarily accumulates the supplied sound collection signal 21 in order to accumulate the sound collection signal 21 for a unit time. The signal buffer 73 temporarily accumulates the supplied signal 25 in order to accumulate the signal 25 for a unit time.

  The signal power calculation unit 72 calculates a power value per unit time using the sound collection signals for unit time accumulated in the signal buffer 71. In addition, the signal power calculation unit 74 calculates a power value per unit time using the signals for the unit time accumulated in the signal buffer 73.

  Here, the power value per unit time is the magnitude of the sound collection signal 21 and the signal 25 in unit time. For example, the maximum value of the amplitude (absolute value) of the sound collection signal 21 and the signal 25 in unit time is An average value, an integrated value of the amplitude (absolute value) of the sound pickup signal 21 and the signal 25 in unit time, or the like can be used. In the present embodiment, as long as the values indicate the magnitudes of the sound pickup signal 21 and the signal 25, values other than the maximum value and the integral value may be used as the power value.

  The sound pressure level difference calculation unit 75 calculates a sound pressure level difference that is a difference between the power value of the collected sound signal 21 obtained by the signal power calculation unit 72 and the power value of the signal 25 obtained by the signal power calculation unit 74. The calculated sound pressure level difference is output to the sound pressure level compensation unit 14 as the sound pressure level change amount 26.

  The sound pressure level compensation unit 14 compensates (adjusts) the sound pressure level of the signal 25 after the noise reduction processing according to the sound pressure level change amount 26 calculated by the sound pressure level change amount calculation unit 13. For example, the sound pressure level compensator 14 has an absolute value of the sound pressure level difference, which is the difference between the sound pressure level of the collected sound signal 21 and the sound pressure level of the signal 25 after noise reduction processing, equals or exceeds a predetermined threshold value. In this case, the sound pressure level of the signal 25 after the noise reduction process is compensated. At this time, for example, the sound pressure level compensation unit 14 may amplify the signal 25 after the noise reduction process with an amplification factor corresponding to the sound pressure level difference. Further, in order to suppress a sudden gain adjustment, the sound pressure level compensator 14 is a signal after noise reduction processing at an amplification factor corresponding to the upper limit value when the sound pressure level difference exceeds a predetermined upper limit value. 25 may be amplified.

  In addition, the sound pressure level difference output as the sound pressure level change amount 26 from the sound pressure level change amount calculation unit 13 changes stepwise in the time direction. Therefore, when the sound pressure level difference is used as a reference value (sound pressure compensation level reference value) when the sound pressure level compensation unit 14 compensates (adjusts) the sound pressure level of the signal 25 after the noise reduction processing, the adjustment is performed. The variation of the later output signal 27 becomes large. In addition, since sudden fluctuations in the noise component give an audible impression to the listener, it is preferable to perform a relaxation process that smoothes fluctuations in the sound pressure level. That is, the sound pressure level compensator 14 may amplify the signal after the noise reduction process with an amplification factor corresponding to the difference in sound pressure level, and then perform a relaxation process (smoothing process) for gradually reducing the amplification factor. Good. In the relaxation processing, for example, a low-pass filter process is performed on the stepped waveform based on the sound pressure compensation level reference value 92 as shown in FIG. 7, or a sound pressure level adjustment value 93 to be described later is gradually reduced. It can be realized by doing.

  Next, operation | movement of the noise reduction apparatus 1 concerning this Embodiment is demonstrated. FIG. 7 is a diagram for explaining an example of the operation of the noise reduction apparatus 1 according to the present embodiment. In FIG. 7, since an unfavorable pseudo noise signal 55 is generated in the adaptive filter of the noise reduction processing unit 12, the timing when the sound pressure level of the signal 25 after the noise reduction processing is reduced is determined as a voice section ( A sound pressure compensation level reference value (respectively indicated by a black circle 92) and a sound pressure level adjustment value 93 actually used are indicated in the time axis direction.

  Here, the timing determined to be a voice segment (respectively indicated by an arrow 91) is a timing determined to be a voice segment by the voice segment determination unit 11, and more preferably a timing determined to be a voice with high probability. is there. For example, by adjusting a threshold value for determining whether or not a speech section is present and making it difficult to determine that a sound component is included in the collected sound signal 21, a more speech-like section (a section having a high probability of being a voice) is selected. It can be detected. The sound pressure compensation level reference value (represented by each black circle 92) is the sound pressure level difference output as the sound pressure level change amount 26 from the sound pressure level change amount calculation unit 13. That is, the sound pressure compensation level reference value is a reference value for determining an amplification factor when the sound pressure level compensation unit 14 performs sound pressure level compensation. The sound pressure level adjustment value 93 corresponds to the amplification factor when the sound pressure level compensation unit 14 compensates the sound pressure level, that is, the amplification factor when the signal 25 after noise reduction processing is amplified.

  FIG. 7 shows a case where the sound pressure level compensation unit 14 performs a relaxation process for smoothing fluctuations in the sound pressure level. That is, in order to smooth the fluctuation of the sound pressure level adjustment value 93, the sound pressure level adjustment value 93 is set to a value corresponding to the sound pressure level difference (sound pressure compensation level reference value), and then the set sound is adjusted. The pressure level adjustment value 93 is gradually reduced. In this embodiment, when the sound pressure level adjustment value 93 is set to a value corresponding to the sound pressure level difference, it is gradually changed.

  Furthermore, in FIG. 7, the threshold value of the sound pressure level difference when compensating the sound pressure level is set to +6 dB. That is, when the sound pressure level difference is smaller than +6 dB (including 0 dB), the sound pressure level is not compensated. However, even when the absolute value of the sound pressure level difference is smaller than +6 dB, when the sound pressure level adjustment value 93 is being gradually reduced, as in the timings G and H shown in FIG. The sound pressure level is compensated using the sound pressure level adjustment value 93 that is being reduced.

  In FIG. 7, the upper limit value of the sound pressure level adjustment value 93 is +12 dB. That is, even if a sound pressure level difference exceeding +12 dB is detected, the actually used sound pressure level adjustment value is suppressed to +12 dB. Thus, the reason why the sound pressure level adjustment value is provided with the upper limit value is to prevent the signal 25 after noise reduction processing from being amplified at an excessive gain in the sound pressure level compensation unit 14. The sound pressure level difference threshold value and the upper limit value of the sound pressure level adjustment value 93 described above are examples, and these values can be set arbitrarily.

  Next, the operation shown in FIG. 7 will be specifically described. At the timing A determined as the voice section, the sound pressure level compensation is not performed because the sound pressure compensation level reference value (sound pressure level difference) is smaller than the threshold value +6 dB.

  Since the sound pressure compensation level reference value is equal to or greater than the threshold value of +6 dB at the timing B determined as the voice section, the sound pressure level compensation unit 14 compensates the sound pressure level. At this time, the sound pressure level adjustment value 93 is adjusted from 0 dB to the same value as the sound pressure compensation level reference value. Since the timing B is an audio section, the signal 25 after the noise reduction processing is a signal including many audio components. Therefore, even if the sound pressure level adjustment value 93 is raised relatively steeply, the sound quality is unlikely to be uncomfortable. In addition, since the noise component included in the signal 25 after the noise reduction processing is small, even if the sound pressure level adjustment value 93 is sharply raised, there is a low possibility of giving a sense of discomfort such as discontinuity due to the noise component.

  At timing B, after adjusting the sound pressure level adjustment value 93 to the same value as the sound pressure compensation level reference value, the adjusted sound pressure level adjustment value 93 is held for a certain period, and then gradually The pressure level adjustment value 93 is reduced. Thus, by gradually reducing the sound pressure level adjustment value 93, the fluctuation of the sound pressure level of the output signal 27 can be smoothed. Therefore, it is possible to achieve both the reduction of the uncomfortable feeling due to the fluctuation of the noise component and the effect of the noise reduction processing.

  At the timings C, D, and E determined as the voice section, the sound pressure level is compensated in the sound pressure level compensator 14 because the sound pressure compensation level reference value (sound pressure level difference) is equal to or greater than the threshold value +6 dB. . Also in this case, the sound pressure level adjustment value 93 is adjusted to be the same value as each sound pressure compensation level reference value. At timings C, D, and E, after adjusting the sound pressure level adjustment value 93 to the same value as each sound pressure compensation level reference value, the adjusted sound pressure level adjustment value 93 is held for a certain period. Thereafter, the sound pressure level adjustment value 93 is gradually reduced.

  At timing F determined to be a voice section, the sound pressure compensation level reference value (sound pressure level difference) is equal to or greater than the threshold value +6 dB, and further exceeds the upper limit value +12 dB of the sound pressure level adjustment value 93. In this case, the sound pressure level adjustment value 93 is suppressed to +12 dB which is the upper limit value. Under a situation where a desired noise reduction effect cannot be obtained even if the noise reduction processing unit 12 is used, due to the influence of the pseudo noise signal 55 generated by the adaptive filter 51, the audio signal included in the signal 25 after the noise reduction processing is The sound pressure level is likely to be unstable. Therefore, by providing an upper limit value for the sound pressure level adjustment value 93, it is possible to suppress the signal 25 after noise reduction processing from being amplified at an excessive amplification factor in the sound pressure level compensation unit 14.

  At timing F, after adjusting the sound pressure level adjustment value 93 to +12 dB, the adjusted sound pressure level adjustment value 93 is held for a certain period, and then the sound pressure level adjustment value 93 is gradually reduced. Yes.

  At timing G determined to be a speech section, the sound pressure compensation level reference value (sound pressure level difference) is smaller than the threshold value +6 dB. Since the timing G is in the middle of gradually reducing the sound pressure level adjustment value 93, the sound pressure level adjustment value 93 exceeds the sound pressure compensation level reference value. At this time, if the sound pressure level adjustment value 93 is the same as the sound pressure compensation level reference value, the sound pressure level adjustment value 93 is lowered more than necessary, and the sound pressure level fluctuates rapidly. Therefore, in this case, the sound pressure level compensation unit 14 compensates the sound pressure level using the sound pressure level adjustment value 93 that is being reduced.

  At timing H determined as a voice section, the sound pressure compensation level reference value (sound pressure level difference) is smaller than the threshold value +6 dB. However, since the timing H is in the process of gradually reducing the sound pressure level adjustment value 93, the sound pressure level is compensated using the sound pressure level adjustment value 93 that is being reduced.

  At the timings I, J, and K determined as the voice section, the sound pressure compensation level reference value (sound pressure level difference) is smaller than the threshold value +6 dB. Since the sound pressure level adjustment value 93 is also 0 dB, the sound pressure level compensation unit 14 does not perform sound pressure level compensation.

  The speech section that is determined to be speech with high probability is preferably a section in which speech such as strong vowels can be detected in the phrase. In such a voice section, it is less susceptible to noise than other parts, and it can be said that it is a good time zone for acquiring a sound pressure level difference. Further, the sound pressure level compensation process shown in FIG. 7 is a trajectory of change from several seconds to several tens of seconds over the whole phrase. At this time, in the voice section (strong voice part), it is raised to the original sound pressure level, and in other parts, sudden fluctuations in the sound pressure level can be suppressed by relaxation processing that gradually reduces the amplification factor. Therefore, the output signal 27 after the sound pressure level compensation processing is performed in the sound pressure level compensation unit 14 is a good sound signal.

  As described in the background art, in the noise reduction processing technique, for example, a noise signal collected by a microphone that mainly collects noise from a voice signal collected by a microphone that mainly collects sound (see The noise component contained in the audio signal is removed by subtracting (signal).

  However, when noise reduction processing is performed using an audio signal mainly including an audio component and a reference signal mainly including a noise component, the audio component may be mixed into the reference signal depending on the use state of the noise reduction apparatus. . As described above, when the sound component is mixed in the reference signal, the sound component included in the sound signal is canceled when the noise reduction process is performed, and the sound pressure level of the signal after the noise reduction process is lowered. there were.

  That is, for example, in a portable wireless communication device such as a transceiver (see FIG. 10), which is often used in a factory where a loud noise such as an operation sound of a work machine is generated, or in a crowded place or an intersection. Therefore, it is necessary to reduce noise components mixed in the microphone. Unlike a mobile phone, a wireless communication device that is used to listen to sound transmitted from a speaker on the main body side away from the ear is generally carried away from the body. There are also various styles of ways to carry wireless communication devices.

  Furthermore, a speaker microphone device (see the voice input device shown in FIG. 9) in which the sound collection unit and the reproduction unit are separated from the wireless communication device main body to improve portability can provide a convenient usage pattern. For example, it is close to the back of the voice input device rather than the front side of the voice input device, such as hanging the voice input device from the neck or placing it on the shoulder, etc. There is also a usage form that speaks from the direction. In such a case, the voice arrival direction is not an ideal arrival direction (for example, the front direction of the voice microphone).

  Therefore, when noise reduction processing using an adaptive filter is performed on a device such as a transceiver (speech input device or wireless communication device), it must be assumed that the reference signal also includes a voice component. Therefore, a technique for suppressing a decrease in the sound pressure level of the audio signal is required.

  Patent Document 1 described above discloses a method of maintaining speech clarity by observing filter coefficients in an adaptive filter and detecting a state in which speech components are canceled. According to this method, an audio microphone that mainly collects sound and a reference sound microphone that mainly collects noise that is low in sensitivity to the direction of arrival of the sound are arranged. Then, when processing with the adaptive filter, when a situation close to the voice arrival direction is generated as a noise cancellation signal, the gain factor applied to the entire adaptive filter coefficient is adjusted to limit the adaptive filter processing. As a result, a decrease in the sound pressure level of the sound component is prevented.

  However, in the technique according to Patent Document 1, it is assumed that a sound source exists on the voice microphone side. In addition, since directivity is given to the reference sound microphone, it is difficult to use the reference sound microphone in a transceiver in which an audio component may be mixed.

  In the technique according to Patent Document 2 described above, the sound pressure level of the audio signal is prevented from being lowered by adjusting the sound pressure level of the error signal or the sound pressure level of the input signal. However, whether to control the sound pressure level of the error signal, which is a noise signal, in order to maintain the sound pressure level of the sound, or to control the sound pressure level of the input signal (including the delay signal) mixed with the noise signal Since either one is performed, the sound pressure level of the audio signal is maintained, but a noise reduction effect cannot be obtained.

  Furthermore, in the noise reduction processing using the adaptive filter disclosed in Patent Document 2, noise cancellation processing by filtering processing is performed using its own signal. For this reason, it is strongly influenced by the mixed audio signal, and the noise component in the audio signal section cannot be reduced. Further, due to the system configuration, an error signal is added to the adaptive filter output signal to obtain a system output signal. However, even if the adaptive filter output signal or the input signal and the error signal in the speech signal section are added as they are, the noise reduction effect cannot be obtained, and the addition of the sound pressure level control does not improve the clarity of the speech.

  Thus, even if the techniques disclosed in Patent Document 1 and Patent Document 2 are used, there is a problem that the sound pressure level of the sound cannot be maintained sufficiently.

  Therefore, in the noise reduction device according to the present embodiment, the amount of change (sound pressure level difference) in the sound pressure level of the signal 25 after the noise reduction processing with respect to the sound collection signal 21 in the speech section determined as speech with high probability. Is calculated by the sound pressure level change amount calculation unit 13, and the signal after noise reduction processing is performed in the sound pressure level compensation unit 14 according to the change amount (sound pressure level difference) calculated by the sound pressure level change amount calculation unit 13. 25 sound pressure levels are compensated.

  Therefore, in the noise reduction device according to the present embodiment, when the sound pressure level of the signal 25 after the noise reduction process is reduced, the amount of change (sound pressure level difference) calculated by the sound pressure level change amount calculation unit 13. Since the signal 25 after the noise reduction processing can be amplified with an amplification factor corresponding to the above, a decrease in the sound pressure level of the output signal 27 can be suppressed.

  Moreover, in the noise reduction apparatus according to the present embodiment, a relaxation process for smoothing the fluctuation of the sound pressure level may be performed. That is, the sound pressure level compensation unit 14 may amplify the signal 25 after the noise reduction process with an amplification factor corresponding to the difference in sound pressure level, and then perform a relaxation process for gradually reducing the amplification factor. By carrying out such processing, the output signal 27 is greatly fluctuated even when the sound pressure level difference output from the sound pressure level change amount calculation unit 13 changes stepwise in the time direction. Can be suppressed. Thereby, rapid fluctuations in the noise component can be suppressed, and a sense of incongruity in hearing can be suppressed. Therefore, it is possible to provide a noise reduction device that can improve the intelligibility of voice while exhibiting a sufficient noise reduction effect under various environments.

  Next, a voice input device using the noise reduction device according to the present embodiment will be described. FIG. 9 is a diagram illustrating an example of a voice input device 500 using the noise reduction device according to the present embodiment. FIG. 9A is a front view of the voice input device 500, and FIG. 9B is a rear view of the voice input device 500. As shown in FIG. 9, the voice input device 500 is configured to be connectable to the wireless communication device 510 via a connector 503. The wireless communication device 510 is a general wireless device, and is configured to be able to communicate with other wireless communication devices at a predetermined frequency. The voice of the speaker is input to the wireless communication device 510 via the voice input device 500.

  The voice input device 500 includes a main body 501, a code 502, and a connector 503. The main body 501 is configured to have a size and shape suitable for being held by a speaker's hand, and includes a microphone, a speaker, an electronic circuit, and a noise reduction device. As shown in FIG. 9A, a speaker 506 and an audio microphone 505 are provided on the front surface of the main body 501. As shown in FIG. 9B, a reference sound microphone 508 and a belt clip 507 are provided on the back surface of the main body 501. An LED 509 is provided on the top surface of the main body 501. A PTT (Push To Talk) 504 is provided on a side surface of the main body 501. The LED 509 notifies the speaker of the detection state of the speaker's voice by the voice input device 500. The PTT 504 is a switch for setting the wireless communication device 510 in a voice transmission state, and detects that the protruding portion is pushed into the housing.

  The noise reduction device 1 ′ (see FIG. 8) according to the present embodiment is built in the voice input device 500, and the voice microphone 16 included in the noise reduction device 1 ′ corresponds to the voice microphone 505 of the voice input device 500. The reference sound microphone 17 included in the noise reduction device 1 ′ corresponds to the reference sound microphone 508 of the sound input device 500. Further, the output signal 27 output from the noise reduction device 1 ′ is supplied to the wireless communication device 510 via the code 502 of the voice input device 500. That is, the voice input device 500 supplies the output signal 27 after the noise reduction processing by the noise reduction device 1 ′ to the wireless communication device 510. Therefore, the sound transmitted from the wireless communication apparatus 510 to another wireless communication apparatus is a sound subjected to noise reduction processing. In the embodiment as shown in FIG. 8, the noise reduction device 1 may be built in the wireless communication device 510.

  Next, a radio communication apparatus (transceiver) 600 using the noise reduction apparatus according to this embodiment will be described. FIG. 10 is a diagram illustrating an example of a wireless communication device 600 using the noise reduction device according to the present embodiment. FIG. 10A is a front view of the wireless communication apparatus 600, and FIG. 10B is a rear view of the wireless communication apparatus 600. As shown in FIG. 10, the wireless communication apparatus 600 includes an input button 601, a display unit 602, a speaker 603, an audio microphone 604, a PTT (Push To Talk) 605, a switch 606, an antenna 607, a reference sound microphone 608, and A lid 609 is provided.

  The noise reduction device 1 ′ (see FIG. 8) according to the present embodiment is built in the wireless communication device 600, and the voice microphone 16 included in the noise reduction device 1 ′ corresponds to the voice microphone 604 of the wireless communication device 600. The reference sound microphone 17 included in the noise reduction device 1 ′ corresponds to the reference sound microphone 608 of the wireless communication device 600. Further, the output signal 27 output from the noise reduction device 1 ′ is subjected to high frequency processing in an internal circuit of the wireless communication device 600, and wirelessly transmitted from the antenna 607 to another wireless communication device. Here, since the output signal 27 output from the noise reduction device 1 ′ is a signal on which noise reduction processing has been performed, the sound transmitted to the other wireless communication device is the sound on which noise reduction processing has been performed. Even if sound transmission is started by the user pressing the PTT 605, the noise reduction process is started. When the user stops pressing the PTT 608 and the sound transmission is ended, the noise reduction process is ended. good.

  Although the present invention has been described with reference to the above embodiment, the present invention is not limited to the configuration of the above embodiment, and can be made by those skilled in the art within the scope of the invention of the claims of the claims of the present application. It goes without saying that various modifications, corrections, and combinations are included. For example, the sound microphone 11 and the reference sound microphone 12 may be provided at substantially the same position in the upper part (or lower part) of the device, and these microphones may be arranged so as to have different directivities. For example, it is preferable to arrange the sound microphone 11 and the reference sound microphone 12 so that the directivities thereof are different by 180 °.

DESCRIPTION OF SYMBOLS 1 Noise reduction apparatus 11 Voice area determination part 12 Noise reduction process part 13 Sound pressure level variation | change_quantity calculation part 14 Sound pressure level compensation part 16 Microphone for sound 17 Microphone for reference sound 21 Sound collection signal (voice signal)
22 Collected sound signal (reference signal)
23, 24 Voice section information 25 Signal 26 after noise reduction processing Sound pressure level change (sound pressure level difference)
27 Output signal

Claims (15)

A speech segment determination unit that determines a speech segment based on the first collected sound signal;
A noise reduction processing unit that reduces a noise component included in the first sound collection signal using a second sound collection signal;
In the speech section, the sound of the signal after the noise reduction processing with respect to the first sound collection signal is performed using the first sound collection signal and the signal after noise reduction processing output from the noise reduction processing unit. A sound pressure level change amount calculating unit for calculating a change amount of the pressure level;
A sound pressure level compensation unit that compensates a sound pressure level of the signal after the noise reduction processing according to a change amount calculated by the sound pressure level change amount calculation unit,
Noise reduction device.   The sound pressure level compensation unit has an absolute value of a sound pressure level difference, which is a difference between a sound pressure level of the first collected sound signal and a sound pressure level of the signal after the noise reduction processing, equal to or greater than a predetermined threshold. 2. The noise reduction device according to claim 1, wherein a sound pressure level of the signal after the noise reduction processing is compensated when the noise reduction processing is performed.   The noise reduction device according to claim 2, wherein the sound pressure level compensation unit amplifies the signal after the noise reduction processing at an amplification factor corresponding to the sound pressure level difference.   The noise reduction device according to claim 3, wherein the sound pressure level compensation unit amplifies the signal after the noise reduction processing with an amplification factor corresponding to the sound pressure level difference, and then gradually reduces the amplification factor.   The sound pressure level compensation unit amplifies the signal after the noise reduction processing at an amplification factor corresponding to the upper limit value when the sound pressure level difference exceeds a predetermined upper limit value. The noise reduction device according to claim 1.   6. The voice section determination unit according to claim 1, wherein the voice section determination unit determines that the voice section is a voice section when a probability that a voice component is included in the first sound collection signal is equal to or higher than a predetermined value. The noise reduction device described.   The voice section determination unit has a ratio between a peak of a vowel frequency component of a voice component included in the first sound pickup signal and a noise level set for each band being equal to or greater than a predetermined value, and The noise reduction device according to any one of claims 1 to 5, wherein a noise section is determined when the number of peaks equal to or greater than a value is equal to or greater than a predetermined number.   The speech section determination unit measures a consonant spectrum pattern of a speech component included in the first collected sound signal for each predetermined frequency band, and a speech section when the consonant spectrum pattern increases as the frequency band increases The noise reduction device according to claim 1, wherein the noise reduction device is determined to be.   The said noise reduction process part is provided with the adaptive filter which produces | generates the pseudo noise signal corresponding to the noise component contained in the said 1st sound collection signal using the said 2nd sound collection signal. The noise reduction apparatus as described in any one of.   A voice input device comprising the noise reduction device according to claim 1. A first microphone is provided on a first surface of the voice input device;
The voice input device according to claim 10, wherein the second microphone is provided on a second surface facing the first surface with a predetermined distance.   A wireless communication device comprising the noise reduction device according to claim 1. A first microphone is provided on a first surface of the wireless communication device;
The wireless communication device according to claim 12, wherein the second microphone is provided on a second surface facing the first surface with a predetermined distance. Determining a voice interval based on the first sound pickup signal;
Reducing a noise component included in the first sound collection signal using the second sound collection signal;
In the voice section, using the first collected sound signal and the signal after the noise reduction process, the amount of change in the sound pressure level of the signal after the noise reduction process with respect to the first collected sound signal is calculated,
Compensating a sound pressure level of the signal after the noise reduction processing according to the calculated change amount,
Noise reduction method. On the computer,
The voice section is determined based on the first sound collection signal,
Reducing a noise component included in the first sound collection signal using the second sound collection signal;
In the voice section, using the first collected sound signal and the signal after the noise reduction process, the amount of change in the sound pressure level of the signal after the noise reduction process with respect to the first collected sound signal is calculated,
Compensating the sound pressure level of the signal after the noise reduction processing according to the calculated change amount,
Noise reduction program.

Images