JP3306784B2 - Bone conduction microphone output signal reproduction device - Google Patents

Abstract

PURPOSE: To provide a bone conduction microphone output signal reproduction device in which a voice signal is corrected in the unit of phoneme so as to obtain an output voice signal with high quality. CONSTITUTION: A voice signal is collected simultaneously from a bone conduction microphone 1 and an air conduction microphone 2 and sets of voice signal waveform patterns divided in the unit of short phoneme time are stored in a conversion rule decision device 9, and when the signal waveform pattern of a collected voice is received from the bone conduction microphone 1, a pattern closest to the signal waveform pattern stored in the conversion rule decision device 9 is selected and the signal waveform pattern related to the selected signal is obtained and the patterns are combined and the result is outputted from an output terminal 16.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a bone conduction microphone output signal reproducing apparatus.

[0002]

2. Description of the Related Art When using communication devices such as mobile phones and transceivers and audio recording devices such as tape recorders,
In order to pick up the voice of the speaker, a microphone that picks up vibrations in the air, a so-called air conduction microphone, is often used. However, when attempting to record a speaker's voice in a place where high-level noise is radiated, such as a construction site, using an air-conducting microphone, the noise is superimposed on the voice, and S / S is sufficient for transmitting. N (voice signal to noise ratio)
Can not get. Therefore, in such a transmission environment, a bone conduction microphone is used.

[0003] The bone conduction microphone is also called a bone conduction microphone. The bone conduction microphone is used to record the vibration of the bone caused by the vibration of the vocal cords at the time of voice generation in a forehead, a chin, a cheek, an ear hole or the like, and to use the signal as a substitute for the actual voice. It is one of the vibration pickups. Under high-level noise, human bones vibrate due to the noise, and it is observed that this is superimposed on the bone-conducted microphone output audio signal.However, since the sound can be recorded at a position close to the sound source, that is, the vocal cord vibration, A higher S / N can be obtained as compared with the air conduction microphone, and it is effective as a voice input unit under high noise.

However, bone conduction microphones have some problems in frequency characteristics as compared with air conduction microphones. The first problem is that a flat frequency characteristic cannot be obtained for the received signal, and the sound is likely to be a sound in which the low frequency band is emphasized. The second problem is that the sound generated by vocal cord vibration
The power of the bone-conducted microphone output voice signal is different from that of the air-conducted voice in the voice generated via the generated feedback other than the vocal cords. For example, when a bone-conducting microphone is placed in the ear canal, the sound-repelling “n” that vibrates the nasal cavity is detected as a higher-level signal than the vowel generated by vocal cord vibration because the position of the nasal cavity and the ear canal are close to each other. , The sound is more uncomfortable than the air-conducted sound. The third problem is that the material of the bone conduction microphone
Depending on the shape and the wearing state, unnecessary sound generated by friction between the microphone unit and the skin is easily picked up, and this unnecessary sound is superimposed on the bone-conducted microphone transmission voice as noise constantly generated by opening and closing of the palate.

[0005] Therefore, when an attempt is made to obtain a clear voice equivalent to that of a normal microphone picked up by a bone conduction microphone, the frequency characteristics of each phoneme (or a short time period equivalent to the phoneme) of the bone conduction microphone output voice signal are determined. It is necessary to use a signal processing technique for flattening, adjusting audio power, removing unnecessary noise, and synthesizing speech again. As such a signal processing technique, conventionally, an attempt has been made to improve the voice quality by performing correction using an active filter on a bone conduction microphone output voice signal.

FIG. 6 shows an example of this filter correction. Here, a voice recorded by a speaker using a bone conduction microphone is called a bone conduction voice, and a voice recorded by a normal air conduction microphone is called an air conduction voice. First, the bone-conducted voice and the air-conducted voice of the speaker are simultaneously recorded using the bone-conducted microphone 1 and the air-conducted microphone 2, respectively, and these are temporarily stored in the tape recorder 3 or the like. A long-term average spectrum is observed for each of the stored voice waveforms, and a long-term spectrum calculation unit 4 obtains a difference between the characteristics of the sound-collecting waveform of the air-conducting microphone 2 and the sound-collecting waveform of the bone-conducting microphone 1. Therefore, if a filter that realizes this difference characteristic is realized by the filter unit 5, the sound collected by the bone conduction microphone 1 will be passed through the filter unit 5 and the pseudo air-conducted sound equivalent to the sound collected by the air-conducted microphone 2. And is obtained from the output terminal 6.

[0007]

However, the above-described conventional improvement method improves the sound characteristics as a long-term average value, and does not correct each syllable correctly. In order to perform more accurate correction, the bone-conducted speech is decomposed in phonemes (or in a short time at the same level as the phoneme), and then the bone-conducted speech is calculated from the bone-conducted speech obtained in advance for each phoneme. There is a demand for a method of performing audio correction filter processing on the audio data and regenerating the audio.

However, the friction sound between the bone conduction microphone and the skin is superimposed on the bone conduction sound as described above. Therefore, in the above-described method in which the bone conduction voice is corrected and used in signal processing, there is a problem that even if the bone conduction voice is divided in units of phonemes, unpleasant noise remains. The present invention has been made in view of the above circumstances, and has as its object to provide a bone-conducted microphone output signal reproducing device capable of obtaining a high-quality output sound by correcting the sound in units of phonemes.

[0009]

According to a first aspect of the present invention, there is provided a bone conduction microphone output signal reproducing apparatus comprising: a bone conduction microphone;
Means for dividing the bone conduction microphone output signal every predetermined short time; air conduction microphone; means for dividing the air conduction microphone output signal every predetermined short time; and the bone conduction microphone output for the predetermined short time. Means for determining a correspondence between a signal and the predetermined short-time air-conducting microphone output signal, and determining a signal conversion rule for the predetermined short-time unit from the bone-conducting microphone output signal to the air-conducting microphone output signal; and Means for storing a conversion rule; and generating and outputting the predetermined short-time pseudo air-conducting microphone output signal from the predetermined short-time bone conduction microphone output signal based on the signal conversion rule stored in the means. Means,
Means for joining each of the predetermined short-time pseudo air-conducting microphone output signals to obtain a long-time pseudo air-conducting microphone output signal, wherein the bone-conducting microphone and the air-conducting microphone record simultaneously. A one-to-one correspondence is obtained for each of the predetermined short-time divided audio signal waveforms, and the predetermined short-time bone conduction microphone output signal is converted into the predetermined short-time pseudo air conduction microphone output signal. By storing as a rule, joining the predetermined short-time pseudo air-conducting microphone output signal obtained based on the conversion rule and the predetermined short-time bone-conducting microphone output signal to obtain a long-term signal waveform The long-time pseudo air-conducting microphone output signal is reproduced.

A bone-conducting microphone output signal reproducing apparatus according to a second aspect of the present invention provides a bone-conducting microphone, a means for dividing the bone-conducting microphone output signal for each predetermined short time, and the bone-conducting microphone output for a predetermined short time. Means for extracting a signal to derive a fundamental frequency and a vocal tract characteristic parameter; an air-conducting microphone; a stage for dividing the air-conducting microphone output signal for each of the predetermined short periods; Means for extracting a vocal tract feature parameter by extracting a feature of the microphone output signal, and determining a correspondence between a vocal tract feature parameter of the bone-conducted microphone output signal and a vocal tract feature parameter of the air-conducted microphone output signal, Converting the vocal tract feature parameters of the guided microphone output signal into the vocal tract feature parameters of the air guided microphone output signal in the predetermined short time unit Vocal tract characteristics of the pseudo air-conducted microphone output signal from the vocal tract characteristic parameters of the bone-conducted microphone output signal based on the conversion rules stored in the means. Means for generating and outputting a parameter; and synthesizing the pseudo air-conducted microphone output signal for a predetermined short time from the vocal tract characteristic parameter of the pseudo air-conducted microphone output signal and a fundamental frequency component of the bone-conducted microphone output signal. Means, and a means for joining each of the predetermined short-time pseudo air-conducting microphone output signals to obtain a long-time pseudo air-conducting microphone output signal, wherein each of the bone-conducting microphone and the air-conducting microphone includes For each audio signal waveform simultaneously recorded and divided every predetermined short time, vocal tract feature parameters are extracted. A one-to-one correspondence is obtained and stored as a conversion rule from the vocal conduction feature parameter of the bone conduction microphone output signal to the vocal tract feature parameter of the pseudo air conduction microphone output signal, and the conversion rule and the bone conduction microphone output signal are stored. Vocal tract feature parameters obtained based on the parameters obtained using the voice conduction feature parameters and the predetermined short-time pseudo air conduction microphone output signal obtained from the pitch component of the bone conduction microphone output signal by joining The long-time pseudo air-conducting microphone output signal is reproduced by obtaining a long-time signal waveform.

According to a third aspect of the present invention, there is provided a bone conduction microphone output signal reproducing apparatus, comprising: a bone conduction microphone; a unit for dividing the bone conduction microphone output signal for each predetermined short time; Means for storing a signal conversion rule for obtaining a pseudo air-conducted microphone output signal corresponding to a signal, and the predetermined short-time bone-conducted microphone output signal based on the signal conversion rule stored in the means. Means for generating and outputting a short-time pseudo air-conducting microphone output signal; and means for joining each of the predetermined short-time pseudo air-conducting microphone output signals to obtain a long-time pseudo air-conducting microphone output signal. Have,
The long-term pseudo-conducted microphone output signal obtained based on the conversion rule and the predetermined short-time bone-conducted microphone output signal is joined to obtain a long-term signal waveform, thereby obtaining the long-term pseudo-conducted microphone. It is characterized by reproducing an air conduction microphone output signal.

According to a fourth aspect of the present invention, there is provided a bone conduction microphone output signal reproducing apparatus, comprising: a bone conduction microphone; a unit for dividing the bone conduction microphone output signal for each predetermined short time; Means for extracting a signal to derive a fundamental frequency and a vocal tract feature parameter, and means for storing a conversion rule for obtaining a pseudo air conduction microphone output signal corresponding to the predetermined short-time bone conduction microphone output signal. Means for storing a conversion rule for obtaining a vocal tract feature parameter of the pseudo air conduction microphone output signal corresponding to the vocal tract feature parameter of the bone conduction microphone output signal, and based on the conversion rule stored in the means. Generating and outputting a vocal tract feature parameter of the pseudo air-conducting microphone output signal from the vocal tract feature parameter of the bone conduction microphone output signal; Means for synthesizing the predetermined short-time pseudo air-conducted microphone output signal from the vocal tract characteristic parameter of the pseudo air-conducted microphone output signal and a fundamental frequency component of the bone-conducted microphone output signal; and Means for joining each of the short-time pseudo air-conducting microphone output signals to obtain a long-time pseudo air-conducting microphone output signal, wherein said conversion rule and said predetermined short-time bone-conducting microphone output signal are voiced. By combining the vocal tract characteristic parameter obtained using the characteristic parameter and the predetermined short-time pseudo air-conducting microphone output signal obtained from the pitch component of the bone-conducting microphone output signal to obtain a long-term signal waveform The long-time pseudo air-conducting microphone output signal is reproduced.

[0013]

In the bone conduction microphone output signal reproducing apparatus according to the first aspect, for a speaker wearing the bone conduction microphone, bone conduction sound from the bone conduction microphone and air conduction sound from the air conduction microphone are simultaneously recorded in advance. Then, the obtained bone-conducted voice signal and air-conducted voice signal are each divided in a short time, and the correspondence between the bone-conducted voice signal and the air-conducted voice signal in a short time is obtained and stored as a signal conversion rule. Next, when bone-conducted speech is input from the bone-conducted microphone, the obtained bone-conducted speech signal is divided for a short time, converted into a pseudo-bone-conducted speech signal according to the previously stored signal conversion rule, and these are joined. To reproduce a long pseudo air-conducted voice signal.

In the bone-conducting microphone output signal reproducing apparatus according to the second aspect, for a speaker wearing the bone-conducting microphone, bone-conducting voice is obtained from the bone-conducting microphone and air-conducting voice is obtained from the air-conducting microphone in advance. Then, the obtained bone-conducted speech signal and the air-conducted speech signal are each divided in a short time to perform signal analysis. As a result, the voice conduction feature parameters are obtained in a short time unit from the bone conduction and air conduction voice signals, and the correspondence between the voice conduction feature parameters of each voice signal is obtained and stored as a conversion rule. Next, when bone-conducted speech is input from the bone-conducted microphone, the obtained bone-conducted speech signal is divided in a short time and analyzed based on the voice-conducted feature parameters obtained based on the conversion rules stored previously. Derive the pseudo-conductivity feature parameters of the bone-conducted speech signal,
Using this parameter and the fundamental frequency obtained by signal analysis of the original bone-conducted speech signal, a short-time pseudo-bone-conducted speech signal is synthesized, and these are joined to reproduce a long-time pseudo-conducted speech signal. .

According to a third aspect of the present invention, when a bone conduction sound is input from the bone conduction microphone, the obtained bone conduction sound signal is divided every predetermined short time. By a universal signal conversion rule for obtaining a pseudo air-conducting microphone output signal corresponding to a predetermined short-time bone-conducting microphone output signal, the signal is converted into the predetermined short-time pseudo-bone-conducting voice signal, and these are joined together. Plays a long pseudo air-conducted audio signal.

According to a fourth aspect of the present invention, when a bone conduction sound is input from the bone conduction microphone, the obtained bone conduction sound signal is divided every predetermined short time. Extracted, synthesized by combining the vocal tract feature parameter obtained by this and the vocal tract feature parameter obtained by a preset universal conversion rule, and the pitch component of the predetermined short-time bone conduction microphone output signal. The predetermined short-time pseudo bone-conducted audio signals are generated, and these are joined to reproduce a long-time pseudo air-conducted audio signal.

[0017]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a schematic configuration of a bone conduction microphone output signal reproducing apparatus according to a first embodiment of the present invention. In FIG. 1, reference numeral 1 denotes a bone conduction microphone, which is attached to a face part, for example, a forehead, a chin, a cheek, an ear hole, or the like, and records a vocal cord vibration of a speaker transmitted to a bone or skin. Reference numeral 2 denotes an air-conducting microphone, which records a real voice signal of a speaker who propagates in the air, that is, a general microphone.

Numerals 3 and 4 denote low-pass filters for preventing aliasing distortion of the output signals of the bone conduction microphone 1 and the air conduction microphone 2, respectively. The cutoff frequency of the low-pass filters 3 and 4 is such that the frequency band of the pseudo air-conducted sound to be finally obtained is to be the same band as the original bone-conducted sound.
Each has the same value, for example, 4 kHz. When the frequency band of the pseudo air-conducted sound to be finally obtained is to be expanded beyond the band of the original bone-conducted sound, the cutoff frequency is, for example, 4 kHz for the low-pass filter 3 and the low-pass filter. 4 may be different values such as 7 kHz.

Reference numerals 5 and 6 denote A / D converters for performing A / D conversion on the outputs of the low-pass filters 3 and 4, respectively, in order to facilitate signal processing performed in the subsequent stage. Each of the A / D converters 5 and 6 includes a frequency band in which sound characteristics of the respective sampling frequency clearly appear, and satisfies the Nyquist sampling theorem with respect to the cutoff frequencies of the low-pass filters 3 and 4. Is fine. Further, the number of quantization bits of the A / D converters 5 and 6 may be any number as long as the characteristics of the voice clearly appear and the quantization distortion is small.

That is, for example, low-pass filters 3 and 4
The sampling frequency and the number of quantization bits of the A / D converters 5 and 6 when the respective cutoff frequencies are the same at 4 kHz are, for example, 8 kHz sampling,
The same is obtained by 12-bit linear quantization. When the cutoff frequency of each of the low-pass filters 3 and 4 is different, such as 4 kHz for the low-pass filter 3 and 7 kHz for the low-pass filter 4, the A / D converter 5
Is, for example, 8 kHz sampling and 12-bit linear quantization, and the A / D converter 6 is, for example, 16 kHz sampling and 16-bit linear quantization.

Reference numerals 7 and 8 denote short-time analyzers, which divide the bone-conducted speech signal and the air-conducted speech signal obtained from the A / D converters 5 and 6 into short-time sections. This division unit takes the same value in each of the short time analyzers 7 and 8, and sets the time length of the phoneme or phoneme level, for example, 32 msec. Further, for example, in the case where a loss of signal power occurs due to multiplication by a window function in a smoothing unit 13 described later, the short-time analyzers 7 and 8 divide the window while overlapping a part of the waveform to obtain a window. Perform processing to avoid loss in the function. FIG. 2 shows an example of a waveform divided by this processing. In the example shown in this figure, divided waveform patterns A, B, and C are generated while partially overlapping the original audio waveform.

Referring again to FIG. 1, reference numeral 9 denotes a conversion rule determiner, which learns and stores the correspondence between the short-time bone-conducted speech signal obtained by the short-time analyzers 7 and 8 and the air-conducted speech signal. It is. That is, the short-time bone conduction voice signal is represented by a (n), a
The short air conduction audio signal recorded simultaneously with (n)
Assuming that (n), the conversion rule determiner 9 calculates a (n) and b
The combination with (n) is determined as a signal conversion rule. In addition,
n indicates the number of the rule to be stored. Here, it is assumed that the number of rules is 1000 at the maximum.

A rule storage 10 stores the conversion rules determined by the conversion rule determiner 9 and supplies the conversion rules to a signal converter 12 at the subsequent stage. Reference numeral 11 denotes a switch for switching an output destination of the bone conduction voice signal from the short-time analyzer 7, and the switch 11 controls a learning mode for learning a signal conversion rule and a signal conversion of the bone conduction voice signal based on the signal conversion rule. And the playback mode for performing

Reference numeral 12 denotes a signal converter, which is a rule storage 1
Based on the signal conversion rule stored as 0, a short-time pseudo air-conducted voice signal is obtained from the short-time bone-conducted voice signal output from the short-time analyzer 7. The smoother 13
A short-time pseudo air-conducted audio signal output from the signal converter 12 is spliced according to the time axis of the original bone-conducted audio signal, and signal distortion occurs due to discontinuity of the signal at the spliced end. In this case, a smoothing process is performed so as not to cause the problem. As a smoothing method, for example, the amplitude value of the signal junction is set to a value close to 0 by a Hamming window function.

Reference numeral 14 denotes a D / A converter, which is a smoother 13
Is to convert a digital signal output from the device into an analog signal. Reference numeral 15 denotes a low-pass filter.
The aliasing distortion of the output signal of the A converter 14 is prevented. Here, the D / A converter 14 is the A / D converter 6
And the low-pass filter 15 has the same sampling frequency and the same number of quantization bits as the low-pass filter 4.
And has the same cutoff frequency. Reference numeral 16 denotes an output terminal for finally outputting a pseudo air-conducted voice signal.

With respect to the operation of the apparatus having the above-described configuration,
The description will be made separately for the learning mode and the reproduction mode. The learning mode is a mode in which a signal conversion rule is determined by obtaining a correspondence between the bone-conducted voice and the air-conducted voice. The playback mode is a mode in which a pseudo air-conducted voice output is obtained from the bone-conducted voice based on the signal conversion rule. It is.

(1) Learning Mode Operation In the learning mode, the switch 11 is connected to the learning mode. In such a state, first, the utterer gives a vocabulary or sentence in which all features are expressed as a voice signal, for example, a document, written by Itabashi, "Common voice data for voice recognition", a symposium of the ASJ Standardization, Proceedings, 1985, 10
Say zero Japanese city names. Here, the environment used by the speaker needs to be a room with a low ambient noise level that does not adversely affect the feature extraction of the voice.

The uttered voice is simultaneously input to the bone conduction microphone 1 and the air conduction microphone 2, respectively, and is converted into digital waveform data through the low pass filters 3 and 4 and the A / D converters 5 and 6. The digital waveform data is divided by the short time analyzers 7 and 8 into short time units as described above, and sent to the conversion rule decision unit 9. As described above, the conversion rule determiner 9 combines the short-time bone conduction audio signal a (n) and the short-time air conduction audio signal b (n) to convert a (n) to b (n).
To the conversion rule.

For a (n), when a large number of bone-conducted voice signals are observed, similar signal patterns are observed, but for similar signal patterns, the same a
(N). That is, assuming that the feature amount on the spectrum of a (n) already stored in the conversion rule determiner 9 such as the LPC cepstrum coefficient is A (n), the newly input short-time bone conduction speech signal a (N ')
Is the absolute value | A (n) −A of the distance of the feature amount on the spectrum, when the spectrum is A (n ′).
(N ′) | is smaller than a predetermined threshold value TH,
This a (n ') is classified as the same bone conduction voice signal pattern as a (n). In this way, the conversion rule determiner 9 determines a certain number of conversion rules. The conversion rules thus obtained are stored in the rule storage 10, and the learning mode ends when a sufficient number of conversion rules are obtained.

(2) Reproduction Mode The reproduction mode is a mode used after the learning mode is completed. In the reproduction mode, the switch 11 is connected to the reproduction mode, and the short-time analyzer 7 and the signal converter 12 are connected. In this mode, the air conduction microphone 2, the low-pass filter 4, the A / D
The converter 6, the short-time analyzer 8, and the conversion rule determiner 9 are not used.

In the reproduction mode, the voice of the speaker is converted into digital waveform data through the bone conduction microphone 1, the low-pass filter 3, and the A / D converter 5, and is divided by the short-time analyzer 7 into short-time units. After the signal converter 12
Sent to Here, the bone-conducted voice signal transmitted to the signal converter 12 is defined as x. Next, the signal converter 12 obtains the absolute value D (n) of the distance of the feature amount on the spectrum between the input x and each a (n) stored in the rule storage 10. Note that D (n) is D (n) = |, where X and a (n) are spectral feature amounts of X and a (n), respectively.
XA (n) |.

Here, when D (n) has the minimum value, a
(N) is a pseudo bone conduction audio signal of the input signal x, and a pseudo air conduction audio signal b (n) is derived. Derived b
(N) is sent to the smoother 13, where it is converted from a short-time split signal into a longer-time signal. Since this signal is digital waveform data, it is converted to an analog signal waveform via the D / A converter 14 and the low-pass filter 15 and output from the output terminal 16 as an original analog signal.

FIG. 3 is a block diagram showing a schematic configuration of a bone conduction microphone output signal reproducing apparatus according to a second embodiment of the present invention. In FIG. 3, portions common to FIG. 1 are denoted by the same reference numerals, and description thereof will be omitted. In FIG. 3, LPC analyzers 16 and 17 perform linear predictive analysis (LPC) on the outputs of the short-time analyzers 7 and 8, respectively, and convert the input speech into a pitch frequency and a parameter indicating a vocal tract feature, for example, LPC. It is separated into coefficients and the like.

The pitch frequency of the bone-conducted voice among the separated voices is sent to an LPC synthesizer 19 described later.
The characteristic parameters of the bone conduction voice and the air conduction voice are sent to the switch 11 and the coefficient conversion rule determiner 9 ', respectively. The coefficient conversion rule determiner 9 ′ determines a conversion rule from the characteristic parameter of the bone-conducted speech to the characteristic parameter of the air-conducted speech, and supplies the rule to the rule storage 10.

Reference numeral 12 'denotes a coefficient converter, which is determined by the coefficient conversion rule determiner 9', and is based on the conversion rules stored in the rule storage 10 and is used to input the characteristics of the bone conduction speech through the switch 11. The characteristic parameters of the pseudo air-conducted speech are derived from the parameters. LPC synthesizer 19
Performs linear prediction analysis (LPC) synthesis on the basis of the pitch frequency of the bone-conducted speech output from the LPC analyzer 17 and the characteristic parameters of the pseudo-conducted speech output from the coefficient converter 12 ', and performs pseudo-simulation in a short time. A signal waveform of the air-conducted voice is generated.

Next, the operation of the bone-conducted microphone output signal reproducing apparatus having the configuration shown in FIG. 3 will be described separately for the learning mode and the reproduction mode, as in the first embodiment. (3) In the learning mode In the learning mode, the switch 11 is connected to the learning mode. In such a state, first, the speaker utters a vocabulary or a sentence in which all features are expressed as an audio signal as in the case of the first embodiment. This sound is
The bone-conducting microphone 1 and the air-conducting microphone 2 are simultaneously input to the low-pass filters 3, 4 and A /
The data is converted into digital form waveform data through the D converters 5 and 6, divided by the short time analyzers 7 and 8 in short time units, and sent to the coefficient conversion rule determiner 9 ′.

The coefficient conversion rule determiner 9 'first extracts and stores speech characteristic parameters x (t) (where t is the input time) such as LPC coefficients for the bone-conducted speech divided into many parts. A certain number of representative parameters p (n) covering a wide range of voice features are derived. This method is well known as vector quantization,
Specific methods include, for example, the LBG algorithm and K
-What is known by the name such as average clustering is used. Here, the number of parameters finally classified, that is, the number n of codebooks is, for example, 25
There are six.

Next, the bone conduction speech feature parameter x
(T) is replaced with any of the 256 representative parameters p (n). That is, x (t) and p (n)
For example, when a feature on a spectrum such as an LPC cepstrum is X (t) and P (n), an absolute value of a distance of the feature on the spectrum D (n) = | X (t) −
For P (n) |, p when D (n) takes the minimum value
Replace (x) with (n).

Here, the air-conducted speech feature parameters recorded simultaneously with the bone-conducted speech feature parameters x are represented by y (t) and x.
Is replaced by p (n), y (t) is replaced by y (t, n)
And for all t, y (t, n) is p
Each (n), that is, 256 types are classified here. However, the classified y (t, n) are totaled, and an arithmetic average thereof is calculated to calculate an average value q (n). By the above-described operation, a conversion rule of the pseudo air-conducted speech feature parameter q (n) with respect to the bone-conducted speech feature parameter p (n) is derived. In the coefficient conversion rule determiner 9 ', this p
A set of (n) and q (n) is sent to the rule storage 10,
The data is stored in the rule storage 10.

(4) Reproduction Mode In the reproduction mode, the switch 11 is connected to the reproduction mode, and the LPC analyzer 17 and the coefficient converter 12 'are connected. In such a state, the voice of the speaker is converted into digital waveform data through the bone conduction microphone 1, the low-pass filter 3, and the A / D converter 5,
It is divided into short time units by the short time analyzer 7 and separated into pitch frequency data v and bone conduction speech feature parameters x by the LPC analyzer 17.

The bone-conducted voice feature parameter x is sent to the signal converter 12 and replaced by a representative parameter p (n) calculated in advance by the coefficient conversion rule determiner 9 ′ and stored in the rule storage 10. That is, x and p
When the spectrum of (n) is X and P (n), respectively, the absolute value of the spectrum distance D (n) = | X−P
For (n) |, p when D (n) takes the minimum value
X is replaced by (n).

Here, based on the rules stored in the rule storage 10, the pseudo air-conducted speech feature parameter q (n) is derived from p (n) and sent to the LPC synthesizer 19.
The LPC synthesizer 19 generates a signal waveform of the pseudo air-conducted voice in short time units based on q (n) and the pitch frequency data v output from the LPC analyzer 17. The generated pseudo air-conducted voice signal waveform is sent to the smoother 13,
The short-time split signal is converted into a long-time signal. Since the output signal of the smoother 13 is digital waveform data, the output signal is converted into an analog waveform via the D / A converter 14 and the low-pass filter 15 and output from the output terminal 16 as an original analog signal.

Here, if each of the conversion rules stored in the respective rule storages 10 of the first and second embodiments described above is a rule that gives a universal conversion result to any speaker, The air-conducting microphone, the part for processing the air-conducted voice recorded by the air-conducting microphone, and the part for calculating and determining the conversion rule become unnecessary, and the configuration of the bone-conducting microphone output signal reproducing device becomes simpler. .
The bone-conducting microphone output signal reproducing device having such a configuration will be described below.

FIG. 4 is a block diagram showing a schematic configuration of a bone conduction microphone output signal reproducing apparatus according to a third embodiment of the present invention. In FIG.
The description is omitted. The device shown in this figure is different from the device shown in FIG. 1 in that an air conducting microphone 2, a low-pass filter 4,
The configuration is such that the A / D converter 6, the short-time analyzer 8, the conversion rule determiner 9, and the switch 11 are removed. However, the rule storage unit 10 of FIG. 4 stores in advance conversion rules that provide universal conversion results for any speaker.

FIG. 5 is a block diagram showing a schematic configuration of a bone conduction microphone output signal reproducing apparatus according to a fourth embodiment of the present invention. In FIG. 5, parts common to those in FIG. Description is omitted. The device shown in FIG.
, An air conduction microphone 2, a low-pass filter 4, an A / D converter 6, a short-time analyzer 8, an LPC analyzer 18, a coefficient conversion rule determiner 9 ', and a switch 11
Has been removed. However, the rule storage unit 10 in FIG. 5 stores in advance conversion rules that provide universal conversion results for any speaker.

In the bone conduction microphone output signal reproducing device according to the third and fourth embodiments described above, since the conversion rule is stored in the rule storage unit 10 in advance, the learning mode in the first and second embodiments is not changed. not exist. Therefore, only the same operation as in the reproduction mode in the first and second embodiments is performed.

As described above, FIG. 1, FIG. 3, FIG.
With the configuration shown in FIG. 5, the speech recorded by the bone conduction microphone 1 is subjected to conversion processing from bone conduction speech to pseudo air conduction speech in short time units of phoneme level based on a conversion rule created in advance. . Therefore, it is possible to remove the influence of the noise included in the bone-conducted speech on the pseudo-conducted speech, and to simulate speech having superior characteristics compared to the conventional correction method using a fixed filter characteristic based on a time average value. It can be obtained as air conduction voice.

Further, in the configurations shown in FIGS. 1 and 3, since the bone-conducting microphone 1 and the microphone 2 can record sound simultaneously, the signals recorded by the respective microphones can correspond to each other. Since this correspondence can be obtained according to the speaker, the conversion process from the bone-conducted speech to the pseudo-conducted speech can be performed with extremely high accuracy.

In the configuration shown in FIGS. 3 and 5, the bone-conducted speech and the air-conducted speech are respectively separated into a fundamental frequency and a voice-conducting feature parameter by various signal analysis techniques, and are converted by using the voice-conducting feature parameter. It is configured to generate rules. Therefore, the rule storage unit 10
Can be reduced, and voice quality can be improved when voices divided for a short time in phoneme units are joined.
Furthermore, in the configuration shown in FIGS. 4 and 5, since universal conversion rules are created in advance for an unspecified number of speakers, the device configuration can be simplified, and the user of the device can also be used. The effort for creating (learning) the conversion rule can be saved.

[0050]

According to the present invention, it is possible to accurately reproduce even a high-frequency signal component which cannot be collected by a conventional bone conduction microphone. Further, by performing the conversion in a short time (phoneme) unit, there is an effect that an optimum voice can be reproduced for each phoneme, as compared with the conventional method in which the correction is made based on the difference of the average spectrum of the voice. Furthermore, since the voice recorded by the bone conduction microphone is not used as the current signal of the voice for correction, unnecessary noise superimposed on the voice collected by the bone conduction microphone does not remain in the converted voice. That is, there is an effect that the influence of unnecessary noise can be removed from the output signal. Also, by storing in advance the universal signal conversion rules for the speaker, there is an effect that the learning operation for each user can be eliminated.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a schematic configuration of a bone conduction microphone output signal reproducing device according to a first embodiment of the present invention.

FIG. 2 is a diagram for explaining short-time analysis of an audio signal.

FIG. 3 is a block diagram showing a schematic configuration of a bone conduction microphone output signal reproducing device according to a second embodiment of the present invention.

FIG. 4 is a block diagram illustrating a schematic configuration of a bone conduction microphone output signal reproducing device according to a third embodiment of the present invention.

FIG. 5 is a block diagram showing a schematic configuration of a bone conduction microphone output signal reproducing device according to a fourth embodiment of the present invention.

FIG. 6 is a block diagram illustrating a conventional bone conduction microphone output signal reproducing device.

[Explanation of symbols]

1 ... bone conduction microphone, 2 ... air conduction microphone, 3,
4,15 ... low-pass filter, 5,6 ... A / D converter,
7, 8: short-time analyzer, 9: conversion rule determiner, 9 '...
Coefficient conversion rule determiner, 10 ... rule storage, 11 ... switch, 12 ... signal converter, 12 '... coefficient converter, 13
... Smoothing device, 14 ... D / A converter, 16 ... Output terminal, 1
7, 18 ... LPC analyzer, 19 ... LPC synthesizer.

Continuation of the front page (56) References JP-A-56-65200 (JP, A) JP-A-63-278100 (JP, A) JP-A-4-245720 (JP, A) JP-A-4-271398 (JP) JP-A-4-276799 (JP, A) JP-A-8-70344 (JP, A) JP-A-3-125400 (JP, U) (58) Fields investigated (Int. Cl. ⁷ , DB Name) G10L 21/00-21/04 H04R 1 / 00,3 / 00

Claims (4)

(57) [Claims] 1. A bone conduction microphone, means for dividing an output signal of a bone conduction microphone every predetermined short time, an air conduction microphone, and means for dividing an output signal of the air conduction microphone every said predetermined time, A signal in the predetermined short time unit from the bone conduction microphone output signal to the air conduction microphone output signal is obtained by determining a correspondence between the predetermined short time bone conduction microphone output signal and the predetermined short time air conduction microphone output signal. Means for determining a conversion rule; means for storing the signal conversion rule; and a method for determining the predetermined short time from the bone-conducting microphone output signal for a predetermined time based on the signal conversion rule stored in the means. Means for generating and outputting a conducting microphone output signal; and joining the predetermined short-time pseudo-conducting microphone output signal to each other for a long-time pseudo-conducting microphone. Means for obtaining a crophone output signal, wherein a one-to-one correspondence is obtained for each of the audio signal waveforms that have been simultaneously recorded from the bone conduction microphone and the air conduction microphone and subjected to the predetermined short-time division, A conversion rule for converting a predetermined short-time bone conduction microphone output signal into the predetermined short-time pseudo air conduction microphone output signal is stored, and is obtained based on the conversion rule and the predetermined short-time bone conduction microphone output signal. Reproducing the long-time pseudo air-conducting microphone output signal by joining the predetermined short-time pseudo air-conducting microphone output signal to obtain a long-time signal waveform. apparatus. 2. A bone conduction microphone, means for dividing an output signal of the bone conduction microphone every predetermined short time, and extracting a characteristic of the bone conduction microphone output signal of the predetermined short time to obtain a fundamental frequency and a vocal tract feature parameter. Means for deriving an air-conducting microphone, a stage for dividing the air-conducting microphone output signal for each of the predetermined short time periods, and vocal tract feature parameters by extracting features of the predetermined short-time air-conducting microphone output signal. Means for deriving a vocal tract characteristic parameter of the bone conduction microphone output signal and a vocal tract characteristic parameter of the air conduction microphone output signal, and calculating the air conduction from the vocal tract characteristic parameter of the bone conduction microphone output signal. Means for determining a conversion rule for the vocal tract feature parameter of the microphone output signal in the predetermined short time unit, and means for storing the conversion rule Means for generating and outputting a vocal tract characteristic parameter of a pseudo air conduction microphone output signal from a vocal tract characteristic parameter of the bone conduction microphone output signal based on the conversion rule stored in the means; Means for synthesizing the predetermined short-time pseudo air-conducting microphone output signal from the vocal tract characteristic parameter of the output signal and the fundamental frequency component of the bone-conducting microphone output signal; and the predetermined short-time pseudo air-conducting microphone output signal. Means for obtaining a pseudo air-conducting microphone output signal for a long period of time by combining each of the above, wherein each of the voices recorded simultaneously from each of the bone-conducting microphone and the air-conducting microphone and divided every predetermined short time The vocal tract feature parameters are extracted for the signal waveform, and a one-to-one correspondence is obtained. It is stored as a conversion rule from the voice conduction feature parameter to the vocal tract feature parameter of the pseudo air conduction microphone output signal, and is obtained based on the conversion rule and the parameter obtained using the voice conduction feature parameter of the bone conduction microphone output signal. The predetermined short-time pseudo air-conducting microphone output signal obtained from the vocal tract feature parameter obtained and the pitch component of the bone-conducting microphone output signal is joined to obtain a long-time signal waveform, thereby obtaining the long-time pseudo air-conducting signal. A bone-conducted microphone output signal reproducing device for reproducing a conductive microphone output signal. 3. A bone conduction microphone, means for dividing the bone conduction microphone output signal for each predetermined short time, and obtaining a pseudo air conduction microphone output signal corresponding to the predetermined short time bone conduction microphone output signal. Means for storing the signal conversion rules, and generating the predetermined short-time pseudo air-conducting microphone output signal from the predetermined short-time bone conduction microphone output signal based on the signal conversion rules stored in the means. Output means, and means for joining each of the predetermined short-time pseudo air-conducting microphone output signal to obtain a long-time pseudo air-conducting microphone output signal, wherein the conversion rule and the predetermined short-time By joining the predetermined short-time pseudo air-conducting microphone output signal obtained based on the bone-conducting microphone output signal to obtain a long-term signal waveform Bone conduction microphone output signal reproducing apparatus characterized by playing the long pseudo air conduction microphone output signal. 4. A bone conduction microphone, means for dividing the bone conduction microphone output signal for each predetermined short period, and extracting a characteristic of the bone conduction microphone output signal for the predetermined short period to obtain a fundamental frequency and a vocal tract characteristic parameter. Means for deriving a pseudo-conducted microphone output signal corresponding to the predetermined short-time bone-conducted microphone output signal; and a vocal tract characteristic parameter of the bone-conducted microphone output signal. Means for storing a conversion rule for obtaining a vocal tract feature parameter of a pseudo air-conducted microphone output signal corresponding to the following; and a pseudo-conduction technique based on the vocal tract feature parameter of the bone-conducted microphone output signal based on the conversion rule stored in the means. Means for generating and outputting a vocal tract feature parameter of the air conduction microphone output signal; and vocal tract characteristics of the pseudo air conduction microphone output signal. Means for synthesizing the predetermined short-time pseudo air-conducting microphone output signal from the characteristic parameter and the fundamental frequency component of the bone-conducting microphone output signal, and joining each of the predetermined short-time pseudo air-conducting microphone output signal. Means for obtaining a pseudo air-conducting microphone output signal for a long time by using the conversion rule and a voice-conducting characteristic parameter of the predetermined short-time bone-conducting microphone output signal. Reproducing the long-time pseudo air-conducted microphone output signal by joining the predetermined short-time pseudo air-conducted microphone output signal obtained from the pitch component of the conductive microphone output signal and obtaining a long-time signal waveform. A bone conduction microphone output signal reproduction device characterized by the above-mentioned.