JP2000148172A - Operating characteristic detecting device and detecting method for voice - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect the voice operating characteristic showing whether a conversation is present in an input signal or not. SOLUTION: In order to form an output signal whether a conversation is present in an input signal or not, this device is provided with first voice operating characteristic detectors 3-6, 14 operating so as to form a spectral similarity value between the input signal component and the component of an input signal judged to have no conversation, a memory 15 for storing the data derived from the part having no conversation, and an auxiliary voice operating characteristic detector 20. The auxiliary voice operating characteristic detector 20 controls updating of the memory 15, and operates so as to form the spectral similarity value between the latest component of the input signal and the earlier component of the input signal.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention
An ice activity detector is a device to which a signal having the purpose of detecting a period of a conversation or a period including only noise is supplied. The invention is not limited to these applications, and a particular embodiment of the invention for such a detector is a mobile radiotelephone system, in which the conversation is performed by a conversation coder (co).
der) to improve the efficient use of the radio spectrum, and in those systems the noise level (from the on-board units) is generally high.

[0002]

The essence of speech motion characteristic detection is to look for different quantities between speech and non-speech periods. In an apparatus that includes a conversation coder, many parameters can be readily used from one coder or from another stage, and by utilizing such parameters, It is desirable to simplify the necessary processing economically. In many situations, dominant noise occurs within a limited region of a frequency spectrum. For example, the noise (eg, engine noise) of a moving car (vehicle) is a low frequency band spectrum. If knowledge of such locations in the noise spectrum is available, it is desirable to base the determination of whether speech is present on the measurand obtained from the relatively noisy portions of the spectrum. Of course, it is actually possible to filter the signal before detecting and analyzing the speech behavior characteristics, but if the speech behavior characteristic detector relies on the output of the speech coder, this precedent stage can be used. Filtering interferes with the coded audio signal.

[0003]

According to the present invention, an embodiment of which is shown in FIG. 3, an input signal component and an output signal indicating whether or not a conversation exists in the input signal are generated. A first operative to form a value of spectral similarity between components of the input signal determined to be free of speech;
A voice operation characteristic detector, a memory for storing data derived from the part where there is no conversation, and an auxiliary voice operation characteristic detector, wherein the auxiliary voice operation characteristic detector updates the memory. Wherein the auxiliary operating characteristic detector operates to form a value of spectral similarity between a current component of the input signal and an earlier component of the input signal. Provided.

A method for detecting a conversational operation characteristic of an input signal, comprising the steps of: receiving an input signal; estimating a noise signal portion of the input signal; storing data representing the noise signal portion; Form a value M of spectral similarity between the portion and the noise signal portion, and define a threshold value (thr
The parameter derived from the value M is compared with a first threshold value T to create a first voice behavior characteristic indication that indicates whether a conversation is present or absent according to whether it is higher than the threshold value. The step of approximating comprises generating an auxiliary voice behavior indicator, and generating the auxiliary voice activity indicator comprises determining the similarity between the latest component of the input signal and the earlier component of the input signal. And comparing said spectral distortion value with a second threshold value to create an indication of the presence or absence of speech in response to exceeding or not exceeding that value, Updating the stored data with an input signal only during a period in which an appropriate voice behavior characteristic indication indicates that no conversation is present.

It is desirable that the value M is a distortion value by Itakura and Saito.

[0006] Other aspects of the invention are within the scope of the claims.

[0007]

BRIEF DESCRIPTION OF THE DRAWINGS Several embodiments of the present invention will now be described with reference to the accompanying drawings.

The general principle characterizing the first embodiment of the speech behavior characteristic detector according to the present invention is as follows.

The n signal samples (s ₀ , s ₁ , s ₂ , s ₃ , s ₄ ... S _n-1 ) are:
When passing through a conceptual fourth-order finite pulse response (FIR) digital filter of pulse response (1, h ₀ , h ₂ , h ₃ ), it becomes a filtered signal (ignoring samples from previous frames). ),

[0010]

(Equation 1)

The zero-order autocorrelation coefficient is the sum of the squares of each term, which is normalized, that is, divided by the total number of terms (for a fixed frame length, the division is omitted. And therefore the sum of the filtered signals is

[0012]

(Equation 2)

This therefore corresponds to the logically filtered signal s
', The signal s in the pass band of the conceptual filter
Is the electric energy of the portion.

When the expansion is performed ignoring the first four terms,

[0015]

(Equation 3)

[0016] Thus, _R'0 is obtained by coupling of the autocorrelation coefficients R _i, weighted by constants in parentheses to determine the frequency band to a value _R'0 responds. In fact, the term in parentheses is the autocorrelation coefficient of the pulse response of the logic filter, so the above expression can be expressed simply as:

[0017]

(Equation 4)

[0018] Here, N is the order of the filter, H _i is (not normalized) of the pulse response of the filter autocorrelation coefficients.

That is, the effect of signal filtering on the signal autocorrelation coefficient is simulated by using the pulse response of the required filter to produce the sum of the (unfiltered) signal autocorrelation coefficients. (Simult
e) Yes.

Thus, a relatively simple algorithm involving a small number of multiplication operations can simulate a digital filter that generally requires this number of 100 multiplication operations.

On the other hand, in this filtering operation, the signal spectrum is matched with the reference spectrum (matche).
d) In state (negative phase response of logic filter), it can be seen as a form of spectral comparison. Since the logical filter in this application is chosen to approximate the inverse of the noise spectrum, this operation is based on the speech and noise spectra, such as values indicating dissimilarities between the spectra, and the generated 0th order. It can be seen as a spectral comparison with the autocorrelation coefficient (ie, the energy of the defiltered signal). The distortion value by Itakura and Saito is calculated by the prediction filter (pre
Used in LPC to evaluate the match between the distor filter and the input spectrum, one format is shown as follows:

[0022]

(Equation 5)

[0023] In this case, such as _{A 0} is LPC parameters - it is a self-correlation coefficient of the data set. This turns out to be very similar to the relationship obtained above, where the LPC coefficients have the inverse spectral response of the input signal.
In fact, given that the LPC coefficient set is the pulse response of the inverse LPC filter,
Saito's distortion value is simply a form of equation 1, where the filter response H is the all-port model (all-p
ole model).

In fact, the LPC coefficients of the test spectrum and the autocorrelation coefficients of the reference spectrum can be used to convert to obtain different values of spectral similarity.

The distortion value due to IS is calculated as “coding speech based on vector quantization” (“Speech Coden”).
g based up Vector Quant
isation "by A Buzo, AH Gra
y, RM Gray and JD Marke
1, IEEE Trans on ASSP, Vol A
SSP-28, No. 5, October 1980).

The above results are only approximate, since the signal frame has only a finite value length and the number of terms (N, where N is the filter order) is ignored. However, it shows very well whether there is a conversation and is therefore used as the value M of the conversation report. The noise spectrum is known,
If it is static noise, it is quite possible to apply fixed h ₀ , h ₁ etc. coefficients to the inverse noise filter.

However, a device that can adapt to different noise situations would be more beneficial.

FIG. 1 shows a first embodiment of the present invention, in which a signal s from a microphone (not shown) has an input 1
At the appropriate sampling rate by the analog / digital converter 2.
ate) to convert to digital samples. LPC analysis unit 3 (general LPC coder)
Is frame sequential samples of n (eg 160) - for arm, to obtain a set of LPC filter coefficients L _i of N that are sent to indicate conversation input (eg 8 or 12). The speech signal s is also a correlation unit (corre
later unit 4) (usually this is LPC coder 3)
Is a part of Because, here, the separation correlator [se
It can be evaluated to provide a [automatic correlation correlation], but the autocorrelation vector R _i of the conversation is usually LP
C is generated as one step of the C analysis). The correlator 4 generates an autocorrelation vector R _i ,
The vector R _i includes a zero-order correlation coefficient R ₀ and at least two further auto-correlation coefficients R 1, R 2, R 3. These are multiplier units.
it) 5.

The second input 11 is connected to a second microphone located away from the speaker, so that only background noise is received. The input from the microphone is converted into a digital input sample sequence by the AD converter 12, and
The LPC is analyzed by the LPC analyzer 13. "Noise" LPC coefficients produced from analyzer 13 passes through the correlation unit 14, the self-correlation vector generated by it is multiplied term by term with the autocorrelation coefficients R _i of the input signal from the conversation microphone of the multiplier 5 , The weighting factors generated thereby are added by adder 6 according to equation 1, thereby providing a filter having the inverse phase shape of the noise spectrum from the noise-only microphone (actually at the signal-pulse-noise microphone). It has the same shape as the noise spectrum), thus filtering out most of the noise. The resulting measured value M is compared to a threshold value by a thresholder 7 to generate a logic output 8 indicating whether a conversation is present. Here, when M is large, it is considered that a conversation exists.

Although this embodiment uses two microphones and two LPC analyzers, the cost and complexity increase, but they can be increased if necessary.

On the other hand, another embodiment uses the autocorrelation from the noise microphone 11 and the corresponding value formed using the LPC coefficients from the main microphone 1. In that case, not the LPC analyzer but another autocorrelator is required.

Thus, these embodiments can operate in different situations with noise of different frequencies, or in a given situation, where there is a changing noise spectrum.

In the preferred embodiment of FIG. 2, a buffer 15 is provided for storing a set of LPC coefficients (or a set of autocorrelation vectors), the values of which are "non-speech". ) (Ie, noise only) "from microphone input 1 during a period defined as"". These values are used to obtain the values according to Equation 1, and of course this measurement corresponds to the distortion measurement method by Itakura and Saito, but not the current frame of LPC coefficients, but an estimate of the inverse noise spectrum. The difference is that a single frame with stored LPC coefficients is used, which matches the value.

The LPC coefficient vector _{L i} output by analyzer 3 is also directed to a correlator 14, thereby generating an autocorrelation vector of the LPC coefficient vector.
The buffer memory 15 is controlled by the speech / non-speech output of the thresholder 7 and provides a "speech" frame.
During the "noise" frame, the buffer retains the "noise" autocorrelation coefficient, but during the "noise" frame, a new set of LPC coefficients is used to update the buffer, e.g. The output of the correlator 14 for transmitting each autocorrelation coefficient is connected to the buffer 15 via the switch 16. The correlator 14 may be arranged after the buffer 15. Further, the speech / non-speech determination for updating the coefficients need not be from output 8;
It can (preferably) be obtained in other ways.

[0035] Since a period without a conversation often occurs,
The LPC coefficients stored in the buffer are updated from time to time, allowing the device to follow changes in the noise spectrum. If the noise spectrum is relatively stable in time (as is often the case), then such a buffer update may be needed very rarely or only for the initial operation of the detector. Though conceivable, it is often desirable to update in situations such as a moving (car) radio.

As a variation on this embodiment, the system applies Equation 1 with the coefficient terms matching a simple fixed high-pass filter, and then switches using the "noise period" LPC coefficients, thereby reducing the system. Start adaptation. If speech detection fails for several reasons, the system can again use a simple high-pass filter.

The above value can be normalized by dividing by R ₀ , and the expression compared to the threshold is

[0038]

(Equation 6)

This value is independent of the frame's total signal power and is therefore compensated for in terms of total signal level changes, but does not provide a significant contrast between "noise" and "talk" levels and therefore It is not suitable for use in noisy environments.

When the noise spectrum changes gradually (as described below) (obtained from noise microphones or noise-only periods in the various embodiments described above).
Instead of using LPC analysis to obtain the inverse filter coefficients of the noise signal, a general adaptive filter (adapt
An ive filter can be used to generate a prototype of the inverse noise spectrum, and a relatively slow precision factor common to such filters can be obtained. In an embodiment consistent with FIG. 1, the LPC analysis unit 13 is easily adapted to a compatible filter (eg, transversal (t
(transversal) FIR or lattice filter
(Lattice filter), whose filter is connected to the system to convert the noise input to white noise by generating an inverse filter prototype, whose coefficients are autocorrelated as described above. Is supplied to the vessel 14.

In the second embodiment shown in FIG.
The PC analysis means 3 is replaced with such a compatible filter, and the buffer means 15 is omitted. However, switch 16 operates to prevent the adaptive filter from adapting its coefficients during the talk period.

A second audio performance detector used in another embodiment of the present invention will now be described.

In the following description, it is clear that the LPC coefficient vector is simply the pulse response of the FIR filter, and that the FIR filter is the inverse phase spectral shape of the input signal. When a distortion value between Itakura and Saito is formed between adjacent frames, the value is actually equal to the power of the signal because it is filtered by the LPC filter of the previous frame. Thus, if there is little difference in the spectra of adjacent frames, the corresponding slight spectral power of the frames will escape filtering and its value will be small.
At the same time, large spectral differences between the frames generate large Itakura-Saito distortion values, whose values reflect the similarity of the spectra of adjacent frames. Speed
For the chicoder, it is desirable to maximize the frame length by minimizing the data rate.
That is, if the frame length is long enough, the speech signal will show significant spectral changes from frame to frame (otherwise the coding is redundant). On the other hand, noise has a spectral shape that gradually changes from frame to frame, and during periods when no speech is present in the signal, applying an anti-phase LPC filter from the previous frame to reduce most of the noise power. "Filter out"
Therefore, the distortion value by Itakura and Saito is correspondingly small.

The Itakura-Saito distortion value between adjacent frames of a noisy signal, including intermittent conversations, is generally greater during speech periods than during noise periods, and the degree of change (indicated by a standard deviation) ), And intermittent changes are small.

Here, the standard deviation of M (standard)
The deviation is also a reliable value, and the effect of taking each standard deviation is essentially to smooth the value.

In this second version of the speech behavior characteristic detector, the measured parameter used to determine whether speech is present is preferably the standard deviation of the Itakura-Saito distortion value. However, other methods of measuring change and measuring spectral distortion (eg, based on FFT analysis) can be applied.

There is also an advantage to using adaptive thresholds for detecting speech behavior characteristics. Such a threshold should not be adjusted during the duration of the conversation, at which point the speech signal will be thresholded out.
t). Therefore, it is necessary to control the threshold adapter using a speech / non-speech control signal,
This control signal is preferably independent of the output of the threshold adapter. The threshold T is adjusted so that it is kept at a level higher than the level of the value M when only noise is present. Since the value generally varies randomly in the presence of noise, the threshold is varied by determining the average level for many blocks and setting the threshold to a level proportional to this average level. However, this is generally not sufficient in noisy situations and the parameters for some blocks
An assessment of the degree of change in the data is taken into account.

Therefore, the threshold value T is calculated according to the following equation.

[0049]

(Equation 7)

Where M is the average of the measured values for many consecutive frames, d is the standard deviation of the measured values for those frames, and K is a constant (typically Is 2.)

In practice, the adaptation operation should not be started again immediately after the indication that no conversation is present, and the descent is stable (to avoid repeated rapid switching during adaptation and non-adaptation conditions). You should wait until you confirm that you have done so.

FIG. 3 illustrates a preferred embodiment of the present invention having the foregoing.
In the preferred embodiment, input 1 is an analog / digital converter
Sampled and digitized by the ADC 2
Received by the input signal of the negative-phase filter analyzer 3.
And the inverted-phase filter analyzer 3 actually
Part of the speech coder on which the operating characteristic detector operates
Filter that matches the opposite phase of the input signal spectrum
Coefficient L_i(Typically 8). Digital signal
Also supplied to autocorrelator 4 (which is part of analyzer 3)
And the autocorrelator 4 receives the input signal (or at least
Is as many low order terms as the LPC coefficient).
Kutor R_iOccurs. The operation of these parts of the device is
This is shown in FIGS. 1 and 2. Autocorrelation coefficient R_iIs suitable
Next, several consecutive speech frames (typically 5 to 5)
20ms) and their reliability is taken
Be improved. This averaging is performed by the autocorrelation in the buffer 4a.
Storing each set of autocorrelation coefficients output by the detector 4;
Using the averager 4b, the current self
Correlation coefficient R_iAnd buffer 4 stored in buffer 4a.
weighting of coefficients from previous frames supplied from a
This is achieved by generating a summed value. It
Averaged autocorrelation coefficient Ra obtained by _iIs heavy
Supplied to the locating and adding means 5, 6, which also comprises
The buffer stored from the autocorrelator 14 via the buffer 15
Negative phase filter coefficient L during the noise period_iAutocorrelation vector A
_iAnd receive Ra_iAnd A_iIs defined by
Form the value M.

[0053]

(Equation 8)

This value is compared by a thresholder 7 to a threshold value and a logical result is generated at output 8 indicating whether a conversation exists or not.

[0055] In order to reverse phase filter coefficients L _i matches the good estimate of the reverse-phase noise spectrum, it is desirable to update these coefficients during the noise (of course, does not update the duration of the conversation). However, speed based on that update
The chi / non-speech decision is unaffected by the result of the update, or a single frame of misidentified signal may result in the speech performance detector eventually "out of lock". And the next frame is erroneously recognized. Thus, a control signal generating circuit 20 is provided, i.e., an auxiliary operating characteristic detector of the separated voice, which forms an independent control signal indicating whether speech is present or not, and provides the antiphase filter analyzer 3 (or buffer 8). ), Whereby the anti-phase filter autocorrelation coefficients A _i used to form the value M are updated only during the “noise” period. The control signal generation circuit 20 is an LPC analyzer 21
(Which is again part of the speech coder and is performed in particular by the analyzer 3), which matches the input signal and the autocorrelator 21a (which can be performed by the autocorrelator 3a). Set of LPC coefficients M
The _i occurs, the autocorrelation coefficients of the autocorrelator 21a is _{M i} B
_{Get i} . When the analyzer 21 is executed by the analyzer 3, M _i = L _i and B _i = A _i . These autocorrelation coefficients are calculated by weighting and adding means 22 and 23.
(Equivalent to 5 and 6), and this means is also applied to the autocorrelator 4
Receive the autocorrelation vector R _i of the input signal from. Therefore, the spectral similarity between the input speech frame and the previous speech frame is calculated. This is because, as mentioned above, current frame - beam of R _i and the previous frame - beam of B
Itakura-Saito distortion value between the _i, also the current frame - obtained by calculating the Itakura-Saito distortion values for beam of Ri and B _i, or the corresponding stored values in the buffer 24 the previous frame - To generate a spectrally different signal (in each case,
The value is preferably energy-normalized by dividing by Ro). Of course, the buffer 24 is updated here. This spectrally different signal, when compared to the threshold by the thresholder 26, indicates whether speech is present, as described above. This method is excellent for discriminating noise from non-speech conversations (a possible task in conventional systems (ta
sk)), it has been discovered that the ability to distinguish noise from spoken conversation is generally low. Therefore, the circuit 20 includes a pitch analyzer.
r) 27 (actually capable of operating as part of a speech coder, in particular capable of measuring long delay values of a predictor generated in a multi-pulse LPC coder) It is desirable to provide a speech detection circuit for the audible speech. The pitch analyzer 27 generates a logic signal that is "true" when a spoken conversation is detected, and this signal is
6 (which is generally "truth" when there is a non-speech conversation),
Provided at the input of NOR gate 28, it generates a signal that is "false" when speech is present and "truth" when noise is present. This signal is supplied to the buffer 8 (or the anti-phase filter analyzer 3), whereby the anti-phase filter coefficient Li is updated only during the noise period.

A threshold adapter 29 is also connected,
The non-speech signal control output of the control signal generation circuit 20 is received. The output of the threshold adapter 29 is supplied to the threshold 7. The output of the threshold adapter 29 is supplied to the threshold 7. The threshold adapter increments or increases the threshold in steps proportional to the instantaneous threshold level until the threshold approaches the noise power level (which is easily obtained, for example, by adding and weighting circuits 22, 23). Decrement
ment). Preferably, when the input signal is very small, the threshold is automatically set to low level. This is because, at small signal levels, the amount of signal generated by ADC 2 cannot produce reliable results.

Further, "Hangove"
r) "means for generating 30 is provided, which comprises
Is measured, and the output is short when the presence of a conversation is indicated for a period exceeding a predetermined time constant.
During a "hangover". In this way, the intermediate loss of the low-level conversation burst (clip).
ping) can be avoided and by choosing an appropriate time constant,
The short spike noise erroneously shown during the conversation can prevent the hang-over generator 30 from being activated. Of course, all of the functions described above could be implemented as a single digital processing means, suitably programmed, for example, as part of an LPC codec (which is the desired configuration) or associated memory device. It can be implemented by means such as a suitably programmed microcomputer or a digital signal processing chip (DSP) configured as a microcontroller chip.

As described above, the voice detection device can be easily configured as a part of the LPC codec.
On the other hand, if the autocorrelation coefficient of the signal, or a value associated therewith (partial correlation or "parcor" coefficient), is transmitted to a remote station, speech detection will be performed away from the codec. Will be

[Brief description of the drawings]

FIG. 1 is a block diagram showing a first embodiment of the present invention.

FIG. 2 is a block diagram showing a second embodiment of the present invention.

FIG. 3 shows a third preferred embodiment of the present invention.

[Explanation of symbols]

 Reference Signs List 1 ... input 2 ... AD converter 3 ... analyzer 4 ... AFC 5 ... multiplier 6 ... adder 7 ... thresholder 8 ... output 11 ... noise microphone 12 ... AD converter 13 ... analyzer 14 ... AFC 15 ... buffer memory 16 ... switch

 ──────────────────────────────────────────────────の Continuing on the front page (72) Inventor Even Boy United Kingdom IP9,2 ETXS, Suffolk, Ipswich, Kapel S.T.Marie, Homefield 5

Claims (8)

[Claims] (I) A spectral similarity between an input signal component and a component of an input signal determined to be free of speech to produce an output signal indicating whether speech is present or absent in the input signal. A first speech behavior characteristic detector (3-6, 14) operative to form a gender value; (ii) a memory (15) for storing data derived from the non-conversation-free part; ) An auxiliary voice performance detector (20), the auxiliary voice performance detector (20) controls updating of the memory (15), and the auxiliary voice performance detector (20). 20) An apparatus for detecting a speech behavior characteristic, which operates to form a value of spectral similarity between a latest component of the input signal and an earlier component of the input signal. 2. Apparatus according to claim 1, wherein the auxiliary voice behavior detector alone controls the updating of the memory. (I) means for receiving an input signal (1)
(Ii) a memory (15) for storing a noise expression signal representing the estimated noise portion of the input signal; and (iii) a part of the input signal and the noise from the input signal and the noise expression signal. Means (3-6, 1) for periodically forming a value of spectral similarity with the estimated noise portion;
(4); (iv) means (7) for comparing said value with a threshold value to produce an output signal indicative of the presence or absence of a conversation; and (v) an auxiliary voice behavior detector (20). And (vi) memory updating means for updating the memory with the input signal, wherein the auxiliary voice operation characteristic detector creates a control signal indicating whether or not a conversation exists. Therefore, it operates according to the value of the spectral similarity between the latest part of the input signal and the previous part of the input signal, and said memory updating means only when said control signal indicates that there is no conversation A voice operation characteristic detecting device operable to update a memory according to the input signal. 4. The apparatus according to claim 3, further comprising means for adjusting said threshold value during a period in which said control signal indicates that no conversation is present. 5. The auxiliary speech behavior characteristic detector further includes means for detecting spoken conversation including pitch analyzer means for generating a signal indicative of the presence of the spoken conversation; 5. The device according to claim 3, wherein the control signal generated by the auxiliary voice behavior detector (20) further obeys this. 6. An apparatus for encoding a speech signal, comprising an apparatus according to any one of claims 1 to 5. 7. A mobile telephone device comprising the device according to claim 1. 8. A method for detecting conversational behavior characteristics of an input signal, comprising: receiving the input signal; estimating a noise signal portion of the input signal; storing data representing the noise signal portion; Forming a value M of spectral similarity between a portion of the signal and the portion of the noise signal, and providing a first speech behavior characteristic indication indicating whether speech is present or absent according to whether it is above a threshold value. Comparing the parameter derived from the value M with a first threshold value T to produce, wherein the estimating step comprises creating an auxiliary voice behavior characteristic indication; and The creation of the 形成 forms a spectral distortion value of the similarity between the latest component of the input signal and the earlier component of the input signal, and the presence or absence of conversation in response to exceeding or not exceeding that value. Comparing the spectral distortion value with a second threshold value to create an absent indication, wherein the auxiliary voice behavior characteristic indication is stored by the input signal only during a period indicating that no speech is present. Updating the data.