JPH07101853B2 - Noise reduction method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a noise reduction method suitable for use in a communication system in a high noise environment.

[0002]

2. Description of the Related Art Generally, in a communication system used in a noisy environment such as an aircraft or an engine room of a ship, even if a voice is usually detected by a microphone, noise due to the high noise is also detected as it is. The voice may be disturbed by noise, making the call extremely difficult.

Therefore, conventionally, a bone conduction microphone (generally an acceleration type pickup) having a high noise reduction effect is used to detect a voice. By attaching the bone conduction microphone to the bone part of the human head, for example, the forehead,
A slight vibration of the bone portion generated when uttering is converted into an electric signal to detect the voice in a pseudo manner, and the influence of a high noise environment in the surroundings is eliminated.

[0004]

However, the conventional method of detecting the sound by the bone conduction microphone has a problem in sound quality because it detects the original sound artificially through the bone vibration. In reality, it is not established as a practical means of communication, such as the intelligibility and naturalness of the communication voice being significantly inferior to that of a microphone.

The present invention solves the problems existing in the prior art as described above, and can improve the S / N ratio of a communication voice in a communication system under a high noise environment and at the same time dramatically improve the sound quality. The purpose is to provide a noise reduction method.

[0006]

A noise reduction method according to the present invention is a method in which an accelerometer such as a bone conduction microphone (hereinafter referred to as a bone conduction microphone) 2 and a normal microphone (hereinafter referred to as a normal microphone) 3 are used for voice Simultaneously, the voice signal Ss obtained from the normal microphone 3 is discriminated into a voiced sound component (vowel component) Sa and an unvoiced sound component (consonant component) Sb, and the voiced sound component Sa is converted into a frequency power spectrum Ps. The pitch frequency Pf of the voiced sound component is obtained based on the voice signal Sf obtained from the bone conduction microphone 2,
A frequency component corresponding to the pitch frequency Pf is extracted from the frequency power spectrum Ps, the extracted frequency power spectrum Po is converted into a voiced sound processing component Sc in a voice signal, and an unvoiced sound component with respect to the voiced sound processing component Sc. Sb, preferably the unvoiced sound processing component Sb obtained by attenuating the unvoiced sound component Sb to a predetermined magnitude is added to obtain the speech processed signal So. In this case, it is preferable that the voiced sound component Sa is converted to the frequency power spectrum Ps by the Fourier transform, and the frequency power spectrum Po is converted to the voiced sound processing component Sc by the inverse Fourier transform. Also, the frequency power spectrum P
The frequency component extracted from s is preset so as to be a fundamental frequency or a predetermined nth harmonic component. The unvoiced sound processing component Sd can be delayed for a predetermined time, and this delay time can be varied. Further, the voiced sound component Sa and the unvoiced sound component Sb are separated from the voice signal Ss by discriminating the voiced sound component and the unvoiced sound component based on the voice signal Sf obtained from the bone conduction microphone 2, and based on the result of this discrimination. Furthermore, the comb filter 4 can be used to extract the frequency component from the frequency power spectrum Ps.

[0007]

According to the noise reduction method of the present invention, the voice in the high noise environment is detected by the bone conduction microphone 2 and the normal microphone 3 at the same time. Then, the voiced sound component and the unvoiced sound component are discriminated based on the voice signal Sf obtained from the bone conduction microphone 2. Based on the discrimination result, the voice signal Ss obtained from the normal microphone 3 is the voiced sound component Sa and the unvoiced sound component Sb.
Is determined.

As a result, the unvoiced sound component Sb is attenuated to a predetermined size to reduce the noise component, and further delayed for a predetermined time to become the unvoiced sound processing component Sd.

On the other hand, the voiced sound component Sa of the voice signal Ss is
For example, it is converted into a frequency power spectrum Ps by the Fourier transform method, and this frequency power spectrum Ps is supplied to the input side of the comb filter 4. On the other hand, in the signal processing system on the bone conduction microphone 2 side, the pitch frequency Pf of the voiced sound component is obtained from the audio signal Sf, and this pitch frequency Pf is given to the control terminal of the comb filter 4. As a result, a frequency component corresponding to the pitch frequency Pf is extracted from the frequency power spectrum Ps at the output of the comb filter 4, and the frequency power spectrum Po based on this frequency component is subjected to, for example, an inverse Fourier transform, whereby a noise component is generated. A voiced sound processing component Sc from which is removed is obtained. It should be noted that the frequency component corresponding to the pitch frequency Pf extracted from the frequency power spectrum Ps is sufficient from the fundamental frequency to a predetermined n-th harmonic component (for example, about the third harmonic component) set in advance.

Therefore, if the unvoiced sound processing component Sd and the component voiced sound processing component Sc are added, the target voice processing signal So can be obtained. The audio processing signal Sd is a signal obtained by subjecting the audio signal Ss obtained from the normal microphone 3 to noise reduction processing.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, preferred embodiments according to the present invention will be described in detail with reference to the drawings.

First, a signal processing apparatus 1 capable of implementing the noise reduction method according to the present invention will be described with reference to FIG.

In FIG. 1, reference numeral 2 denotes a bone conduction microphone (acceleration type pickup) attached to the forehead of the human body, and this bone conduction microphone 2 is connected to the input side of a low pass filter 11. The output side of the low-pass filter 11 is connected to the input side of the analog-digital converter 12, and the converter 1
The output side of 2 is connected to the utterance state recognition unit 13. further,
The output side of the recognition unit 13 is connected to the input side of a voice component determination unit 14 that determines a voiced sound component and an unvoiced sound component, and
The discrimination unit 14 is connected to a pitch frequency detection unit 15 that detects the pitch frequency. Then, the detection unit 15 is connected to the control terminal of the comb filter 4.

On the other hand, 3 is a normal microphone for directly detecting a vocal sound represented by a dynamic microphone, a condenser microphone, etc. The normal microphone 3 is connected to the input side of the low-pass filter 16. The output side of the filter 16 is connected to the input side of the analog-digital converter 17, and the output side of the converter 17 is connected to the first changeover switch 1.
8 is connected to one contact portion. The first changeover switch 18 is changed over based on the recognition result of the utterance state recognition unit 13. The other contact of the changeover switch 18 is connected to the movable contact of the second changeover switch 19. The second changeover switch 19 is changed over via the control circuit 20 according to the discrimination result of the voice component discrimination section 14. Furthermore, one fixed contact portion of the second changeover switch 19 is connected to the input side of the high-speed Fourier converter 21, and the output side of the converter 21 is connected to the input side of the comb filter 4. The output side of the comb filter 4 is connected to the input side of the inverse Fourier transformer 22.

On the other hand, the other fixed contact portion of the second changeover switch 19 is connected to the input side of the attenuation circuit 23, and the output side of the circuit 23 is connected to the input side of the delay circuit 24. A coefficient unit 25 is connected to the attenuation circuit 23.

The output side of the delay circuit 24 and the output side of the inverse Fourier converter 22 are both connected to the input side of the digital-analog converter 26 and the converter 26
The output side of is connected to the input side of the output unit 28 via the low-pass filter 27, and the output side of the output unit 28 is connected to the speaker 29 that outputs sound.

Next, a noise reduction method according to the present invention using the signal processing device 1 will be described with reference to FIGS. 1 and 2.

First, the voice is simultaneously detected by the bone conduction microphone 2 and the ordinary microphone 3.

The high-frequency component of the audio signal Sf obtained from the bone conduction microphone 2 is removed by the low-pass filter 11,
It is converted into a digital signal by the analog-digital converter 12. Then, the uttered state recognition unit 13 recognizes whether or not the uttered state. Speech signal S
It can be recognized by whether or not the voltage based on f is generated above a predetermined threshold value.

On the other hand, the voice signal Ss obtained from the ordinary microphone 3
The high-pass frequency component is removed by the low-pass filter 16, converted into a digital signal by the analog-digital converter 17, and applied to one contact portion of the first changeover switch 18.

As a result, when the utterance state recognition unit 13 recognizes the voice signal Sf, the first changeover switch 18 is controlled to be turned on, and the voice signal Ss is passed through the first changeover switch 18 to the movable contact of the second changeover switch 19. Given to the department.
If the utterance state recognition unit 13 does not recognize the utterance, the first changeover switch 18 is controlled to be turned off, and the input of the voice signal Ss is cut off.

On the other hand, the voice signal Sf that has passed through the utterance recognition unit 13 is given to the voice component discriminating unit 14 to discriminate a voiced sound component and an unvoiced sound component. In this case, the discrimination is performed by the highly accurate autocorrelation method. That is, the voiced sound component is composed of harmonics of the fundamental frequency (pitch frequency) of the vocal cord and is characterized as a vowel component, and the unvoiced sound component is composed of random noise components and characterized as a consonant component. Also, the autocorrelation function Φ for a finite discrete signal
When the autocorrelation of the audio signal Sf is obtained based on the general expression of (k), it becomes a periodic function which takes a maximum value at k = 0, and by finding the maximum value Φ (t ₁ ) in the part of k ≠ 0, , The most dominant cycle of the audio signal Sf is obtained.
Therefore, the voiced sound component and the unvoiced sound component can be discriminated based on the dominant cycle and the value of Φ (t ₁ ). Specifically, if a threshold value is set, the value of Φ (t ₁ ) is larger than the threshold value, and the dominant cycle is a value within a range of frequencies, the dominant cycle is the voice signal. The voice signal is determined to be the fundamental period, and the voice signal is regarded as the voiced sound component, and otherwise, it is regarded as the unvoiced sound component.

When the voice signal Sf is the unvoiced sound component, the second changeover switch 19 is switched to the attenuation circuit 23 side, and the input voice signal Ss, that is, the unvoiced sound component Sb.
Is supplied to the attenuation circuit 23. The unvoiced sound component Sb is reduced by the attenuation circuit 23 by a coefficient of "1" or less (preferably 1/5 to 1).
/ 10) and is further delayed via the delay circuit 24. As a result, the unvoiced sound processing component Sd is obtained, and the same component Sd is added to the input side of the digital-analog converter 26. The attenuation amount of the attenuation circuit 23 can be arbitrarily set by the coefficient unit 25, and preferably, the attenuation amount that can reduce the noise component is selected. The reason for processing the unvoiced sound component Sb in this way is as follows. If a voiced component is a voiced component even in a high-noise environment, a pitch frequency is detected by the autocorrelation method, and spectral subtraction is performed according to the detected pitch frequency, whereby a voice with a high S / N ratio can be obtained. However, when a non-periodic noise component such as unvoiced sound component random noise overlaps, the unvoiced sound component and the noise component are very similar in nature, so that it is not easy to improve the S / N ratio. Therefore, when the voice signal is determined to be an unvoiced sound component, the unvoiced sound component Sb is multiplied by the above-described appropriate coefficient (which can be changed arbitrarily) to be attenuated and output as the unvoiced sound processing component Sd which is a pseudo consonant component. The delay circuit 24 adjusts the timing with respect to the processing speed on the side of the voiced sound component Sa described later.

On the other hand, when the voice signal Sf is a voiced sound component, the second changeover switch 19 is switched to the high speed Fourier converter 21 side, and the input voice signal Ss, that is, the voiced sound component Sa, is a high speed Fourier converter. 21. The switching control of the second changeover switch 19 is performed via the control unit 20 to realize smooth high speed switching. Then, the voiced sound component Sa is subjected to Fourier transform by the high-speed Fourier transform unit 21, and the frequency power spectrum Ps obtained by the Fourier transform is supplied to the input side of the comb filter 4.

On the other hand, when the discrimination result of the voice component discriminating unit 14 is the voiced sound component, the pitch frequency detecting unit 15 obtains the pitch frequency Pf, and the pitch frequency Pf is given to the control terminal of the comb filter 4.

The comb filter 4 is controlled to the pitch frequency Pf and constitutes a filter having a constant pass band width with the pitch frequency Pf and the harmonic component of the pitch frequency as the center frequency. Further, the comb filter 4 has an arbitrary pass band width corresponding to a change in the pitch frequency Pf.

As a result, as shown in FIG. 2, the comb filter 4 extracts the frequency component corresponding to the pitch frequency Pf from the frequency power spectrum Ps, and the extracted frequency power spectrum Po is the inverse Fourier transformer 22.
By the inverse Fourier transform. As a result, the voiced sound processing component Sc from which the noise component has been removed is obtained and applied to the input side of the digital-analog converter 26. The frequency components extracted from the frequency power spectrum Ps are from the fundamental frequency to a predetermined n-th order harmonic component set in advance, and normally the third-order harmonic component is sufficient. Figure 2
Where fp is the fundamental frequency component, 2fp is the second harmonic component,
3fp shows a third harmonic component.

Then, the unvoiced sound processing component Sd and the voiced sound processing component Sc delayed by a predetermined time are added at the input side of the digital-analog converter 26, whereby noise is added to the voice signal Ss obtained from the ordinary microphone 3. The audio processing signal So that has been subjected to the reduction processing and the time adjustment processing is obtained, and this audio processing signal So is converted into an analog signal by the converter 26, and is output as audio from the speaker 29 via the low-pass filter 27 and the output unit 28. Is output.

The embodiment has been described in detail above.
The present invention is not limited to such an embodiment.
For example, the voice signal is separated into the changeover switch to obtain the voiced sound component and the unvoiced sound component, but it is not necessary to separate them. In addition, it is of course possible to use a method other than the Fourier transform for the conversion into the frequency power spectrum. In addition, the detailed circuit configuration and method may be arbitrarily changed without departing from the scope of the present invention.

[0030]

As described above, according to the noise reduction method of the present invention, the voice is simultaneously detected by the acceleration type pickup and the ordinary microphone, and the voice signal obtained from the ordinary microphone is discriminated into the voiced sound component and the unvoiced sound component. The component is converted to a frequency power spectrum, while the pitch frequency of the voiced sound component is obtained based on the voice signal obtained from the acceleration pickup, and the frequency component corresponding to the pitch frequency is extracted from the frequency power spectrum, and the extracted frequency Since the power spectrum is converted to the voiced sound processing component in the voice signal, and the voiced sound processing component and the unvoiced sound processing component obtained by attenuating the unvoiced sound component or the unvoiced sound component are added to obtain the voice processing signal, it is possible to operate in a high noise environment. Can improve the S / N ratio of the call voice in the communication system of Voice of sound quality, that is, can dramatically improve the clarity and naturalness of sound.

[Brief description of drawings]

FIG. 1 is a block circuit diagram of a signal processing device capable of implementing a noise reduction method according to the present invention.

FIG. 2 is an operation explanatory view showing a frequency power spectrum,

[Explanation of symbols]

 2 Bone conduction microphone 3 Normal microphone 4 Comb filter Ss Voice signal Sf Voice signal Sa Voiced sound component Sb Unvoiced sound component Sc Voiced sound processing component Sd Unvoiced sound processing component So Voice processing signal

Claims (9)

[Claims] 1. A voice is simultaneously detected by an acceleration type pickup and an ordinary microphone, a voice signal obtained from the ordinary microphone is discriminated into a voiced sound component and an unvoiced sound component, and the voiced sound component is converted into a frequency power spectrum. Form the pitch frequency of the voiced sound component based on the voice signal obtained from the pickup, and extract the frequency component corresponding to the pitch frequency from the frequency power spectrum, the extracted frequency power spectrum in the voice signal voiced sound processing component And a voiceless signal is obtained by adding a voiced sound processing component and an unvoiced sound component or an unvoiced sound processing component obtained by processing the unvoiced sound component. 2. The noise reduction method according to claim 1, wherein the voiced sound component is subjected to a Fourier transform to obtain a frequency power spectrum, and the frequency power spectrum is subjected to an inverse Fourier transform to obtain a voiced sound processing component. 3. The noise reduction method according to claim 1, wherein the frequency component extracted from the frequency power spectrum is a fundamental frequency or a predetermined n-th harmonic component set in advance. 4. The noise reduction method according to claim 1, wherein the unvoiced sound processing component is obtained by attenuating the unvoiced sound component to a predetermined size. 5. The noise reduction method according to claim 4, wherein the unvoiced sound processing component is delayed for a predetermined time. 6. The noise reduction method according to claim 5, wherein the delay time is variable. 7. A voiced sound component and an unvoiced sound component are discriminated based on a voice signal obtained from an acceleration pickup, and a voice signal obtained from a normal microphone is selectively separated into a voiced sound component and an unvoiced sound component based on the discrimination result. The noise reduction method according to claim 1, which is characterized in that. 8. The noise reduction method according to claim 1, wherein the acceleration type pickup uses a bone conduction microphone. 9. The noise reduction method according to claim 1, wherein a comb filter is used to extract the frequency component from the frequency power spectrum.