JPH0522779A - Speech recognition remote controller - Google Patents

Abstract

PURPOSE:To enhance the precision of recognition of the speech recognition section of a speech recognition remote controller by taking out an output approximate to noise component from an adaptable noise reducer and by applying subtraction to the output. CONSTITUTION:Main input from a microphone 3 is sent to adder 15 via input terminal 10A of adaptable noise reducer 10. On the other hand, as a signal corresponding to the speech from speaker 8 which becomes a noise component, a speech signal from speech output circuit 6 is sent as a reference signal to adaptable filter 16 via input terminal 10C of adaptable noise reducer 10. This output is sent as a subtract signal to adder 15. With this, the noise component of main input is reduced, and is output to speech recognition unit 4 via output terminal 10B of adaptable noise reducer 10. Since speech recognition unit 4 executes speech recognition processing and command interpretation for the input for which noise is reduced, speech recognition processing can be executed highly precisely.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is provided so as to be integrally attached to an electric equipment body having a sound source such as a television receiver or as a single device for controlling the electric equipment. Such a voice recognition remote control device.

[0002]

2. Description of the Related Art Conventionally, an infrared remote control (so-called remote control) device or the like has been used to remotely control an electric device having a sound source such as a television receiver. This remote control device arranges some switches on the operation panel surface of a remote operation unit called a commander or remote control unit, and manually operates these switches to generate modulated infrared rays. Is received by an infrared sensor or the like of a receiving unit attached to the main body side of a television receiver or the like, and a control operation corresponding to the switch operation in the remote operation unit, such as power on / off, channel switching, and volume control. It is for performing operations such as adjustment.

[0003]

However, the above-mentioned remote control device has a drawback that the remote control unit is easily lost. This is because each individual home-use device comes with a separate remote control unit, which makes management complicated and leaves the remote control unit in an unexpected place due to its portability. Sometimes, for example, after being used by another person or a child, it may be necessary to search for it by placing it in an unexpected place. If the remote control unit is not found, some operations can be performed only by the remote control unit even if the minimum operation can be performed on the electric equipment main body side, which is inconvenient.

Further, as the functions of the electric devices to be controlled, such as recent television receivers, become more abundant, the operation of the remote control unit becomes complicated, and it is often impossible to use it without getting used to it. As a result, the dimensional and shape of the unit becomes large due to the increase in the number.

As a method for solving these drawbacks, a voice recognition remote control in which a user gives an instruction by voice and the voice is recognized by a receiving unit or the like of the electric equipment to be controlled and the equipment is controlled Although a device is conceivable, it is necessary to reduce ambient noise and noise as much as possible in order to improve the accuracy of voice recognition. In particular, in the case of an electric device having a sound source such as a television receiver, if a receiving unit is provided on the device body side and a microphone is arranged, the sound from the sound source is mixed with the sound from the user, The voice recognition accuracy is extremely reduced, which is not practical.

Here, it is conceivable that only the microphone can be detached and the voice recognition accuracy is improved by bringing the microphone close to the user's mouth to reduce unnecessary noise, but an operation such as holding the microphone by hand is required. Come on. In this case, when the wired microphone is used, the remote operability is worse than that of the infrared remote controller described above, and when the wireless microphone is used, the disadvantage that the unit is easily lost is not solved. Become.

The present invention has been made in view of the above circumstances, and it is possible to control an electric device only by voice without using an operation unit that is operated at hand, and moreover, it is provided in the electric device. It is an object of the present invention to provide a voice recognition remote control device that can perform voice recognition with high accuracy even if a voice or the like from a sound source is mixed with a voice for control operation.

[0008]

A voice recognition remote control device according to the present invention comprises a voice receiving means provided in the vicinity of a main body of an electric device having a sound source, and a main part obtained by receiving the voice receiving means. A subtracting means to which an input is supplied, and an adaptive filter for sending an output filtered by a signal from the sound source to the subtracting means and controlling a filter characteristic so as to minimize the power of the output from the subtracting means. By having a voice recognition unit to which the output from the subtraction unit is supplied and a control unit that controls the electric device according to the recognition output from the voice recognition unit, the above problems can be solved.

A microphone can be used as the voice receiving means, and as an electric device having a sound source, a sound receiver such as a television receiver, a radio receiver, a cassette tape recorder with a radio, a VTR (video tape recorder), or the like can be used. Examples include video equipment, air conditioners (air conditioners), coolers, fans, etc., where motor noise, wind noise, etc., tend to be noise noise. The voice receiving means may be integrally attached to the main body of the electric device, but may be provided in a control device independent from the dedicated electric device for receiving the voice and controlling the electric device. The subtracting means and the adaptive filter constitute a so-called adaptive noise reduction device.

As the voice recognition unit, a unit for identifying whether the voice uttered by the user or the like is one of the predefined commands by pattern matching can be used, and this pattern matching performs voice analysis. It is preferable to extract the characteristics of the voice and perform the normalization to remove the element due to the individual difference to perform the pattern. In voice analysis, frequency analysis is generally performed using a bandpass filter group, and the frequency analysis result may be obtained in the form of a power spectrum envelope or the like. From this speech analysis result, in order to remove elements due to individual differences such as sound intensity, duration, vocal cord characteristics, etc., normalization is performed, and the obtained normalized pattern is used for comparison in advance. And the identification result may be output.

[0011]

From the adaptive filter, an output similar to the noise component from the sound source mixed with the main input from the voice receiving means is taken out and sent to the subtracting means to be subtracted from the main input. It Therefore, substantially only the signal component such as voice is sent from the subtraction means to the voice recognition section, and the recognition accuracy in the voice recognition section is improved.

[0012]

1 is a block circuit diagram showing a schematic configuration of a voice recognition remote control apparatus according to a first embodiment of the present invention. In the first embodiment, the television receiver 2 is used as the electric device to be controlled. When the user 1 utters a predefined command, this is received (picked up) by the microphone 3 installed in the television receiver 2, the noise component is reduced by the adaptive noise reduction device 10, and the voice recognition device 4 receives it. Sent. The voice recognition device 4 interprets the command, and the control circuit 5 controls the television receiver 2 according to the interpreted command.

In the television receiver 2 which is the electric device to be controlled in FIG. 1, the audio output signal from the television audio output circuit 6 is amplified by the amplifier 7 and sent to the speaker 8 as a sound source. ing. Therefore, the microphone 3 which is the voice receiving means of the voice recognition remote control device provided in the main body of the television receiver 2 receives not only the command voice uttered by the user 1 but also the television from the speaker 8 which is the sound source. Audio (broadcast audio, etc.) will also be circulated and received (picked up). This speaker 8
Is a noise component unnecessary for the above command interpretation.

The main input received by the microphone 3 is sent to the adder 15 serving as subtraction means via the input terminal 10A of the adaptive noise reduction device 10. Adder 15
The output from the output terminal 10 of the adaptive noise reduction device 10 is
It is taken out via B and sent to the voice recognition device 4. An audio signal from the audio output circuit 6 is sent as a reference signal to the adaptive filter 16 via the input terminal 10C of the adaptive noise reduction device 10 as a signal corresponding to the audio from the speaker 8 which becomes the noise component. The output from the adaptive filter 16 is sent to the adder 15 as a subtraction signal. The filter characteristics of the adaptive filter 16 are controlled so as to minimize the power of the output (residual error) from the adder 15. As a result, the noise component (mainly the sound from the speaker 8) in the main input is reduced and extracted from the output terminal 10B of the adaptive noise reduction device 10, as described later.

The voice recognition device 4 performs voice recognition processing on the noise-reduced output (residual error) and interprets the command. The voice recognition in this case can be performed with high accuracy because noise components other than the command voice uttered by the user are reduced. According to the recognition result, the control circuit 5 causes the television receiver 2 to perform a control operation according to the above command, for example, turning on / off the power, switching the channel, adjusting the volume, and the like.

Next, the adaptive noise reduction apparatus 10 will be described with reference to FIGS. 2 and 3. First, Figure 2
FIG. 3 is a block circuit diagram for explaining the basic operation of the adaptive noise reduction device 10, and is provided to receive a signal source 11 corresponding to the command voice uttered by the user and a voice from the signal source 11. Microphone 12
A noise source 13 corresponding to the sound from the speaker 8;
A configuration having a microphone 14 provided to receive noise from this noise source is assumed. In this configuration, the microphone 12 for signal reception is the same as that shown in FIG.
The input from the microphone 14 for noise reception corresponds to the audio signal from the audio output circuit 6 in FIG.

In FIG. 2, the voice from the signal source 11 (the above-mentioned command voice) is received by the microphone 12, and the noise from the noise source 13 is received by the microphones 12 and 14, respectively. Therefore, the microphone 12, and the signal component s from the signal source 11 is a noise component n _a and superposition (sum) signal s + n _a is received from the noise source 13, adaptive noise reduction system which as the primary input 10 Is sent to the input terminal 10A. In the microphone 14,
A noise component n _b from the noise source 13 is obtained, and this is used as a reference input to the input terminal 10 C of the adaptive noise reduction device 10.
Have been sent to. A signal component s which corresponds to the command voice, the noise component n _a which corresponds to the speaker voice, n
_Although it is uncorrelated with _b , the noise components n _a and n _b are correlated.

[0018] The primary input s + n _a obtained by being received by the microphone 12 is fed to the adder 15 is a subtraction means. The reference input n _b received by the microphone 14 and obtained is sent to the adaptive filter 16 as the input x and filtered to become the output y. The filter output y is sent as a subtraction signal to the adder 15 is subtracted from the primary input s + n _a, the residuals e is taken out from an output terminal 17 with are sent to the adaptive filter 16. The adaptive filter 16 adaptively controls the filter characteristic so as to minimize the power of the residual e.

As shown in FIG. 3, the adaptive filter 16 has a filter section 21 and an adaptive algorithm section 22. Based on the reference input n _b , the adaptive filter 16 learns the noise component n _a . A pseudo output (pseudo noise) y is generated. The reference input n _b is the terminal 16a.
Is supplied as an input x to the filter unit 21 and the adaptive algorithm unit 22 via. Output y from the filter unit 21
Is taken out as an output of the adaptive filter 16 through a terminal 16b and sent to the adder 15 which is the subtracting means. The residual e from the adder 15 is supplied to the adaptive algorithm unit 22 via the terminal 16c. Adaptive algorithm section 22, by varying the filter coefficients of the filter unit 21 changes the filter characteristics, the output y obtained input x to filter main input s + n _a
The adaptive control is performed so as to minimize the power of the residual e obtained by subtracting from.

[0020] That is, the residual e is, e = s + n _a -
It can be represented as y, and the mean value of the squares of this is E [e ² ].
^{Is, e 2 = s 2 + (} n a -y) than _{^{2 + 2s (n a -y)}} , E [s ^2] + E [(n _a -y) ^2] + 2E [s (n
_a- y)]. Considering the expected value the average of long time which, s is an n _a, and since the y uncorrelated, three items 2E [s (n _a -y)] is Yuki becomes 0, expectation value of E [e ^2] is, E [s ^2] + E [(n _a -
y) ² ]. Given that the adaptive filter converges, this is adjusted so that E [e ² ] is minimized, while E [s ² ] is not affected, so E _min [e ² ] =
E a [s ^2] + E _min [(n _a -y) ^2]. That is, by minimizing E [e ² ], E [(n _a
-Y) ² ] is minimized, and the filter output y becomes an estimated value of n _a .

FIG. 3 shows a specific example in which a so-called FIR (finite impulse response) filter is used as the filter section 21. That is, the reference input x from the input terminal 16a is the delay element 23 ₁ according to the number of taps,
23 ₂ , ..., 23 _L sent to a series circuit. Input x ₀ from the input terminal 16a and each delay element 23 ₁ , 2
Each output from 3 ₂ , ..., 23 _L x ₁ , x ₂ , ...
., X _L are coefficient multipliers 24 ₀ , 24 ₁ , 24, respectively.
₂ , ..., 24 _L , each of which has a filter coefficient w
₀ , w ₁ , w ₂ , ..., W _L are multiplied to adder 25
Have been sent to. Each filter coefficient w ₀ , w ₁ , w ₂ , ...
.., w _L are corrected by the coefficient correction signal from the adaptive algorithm unit 22, and the output y from the adder 25 is taken out from the output terminal 16 b.

As the adaptive algorithm used in the adaptive algorithm unit 22, many methods have been proposed. One specific example is LMS (least mean square,
The Least Mean Square algorithm is described.

The input data at the k-th sample cycle time point (time k) of the data sequence of the input x and the delayed output data from the delay elements 23 ₁ , 23 ₂ , ..., 23 _L are respectively x _k0 , X _k1 , x _k2 , ..., X _kL , the input vector X to be FIR-filtered
a _k, put the X _k = _{[x k0 x k1 x k2 ··· x} kL ] ^T ··· (1). T in the equation (1) represents a transposed symbol. For this input vector X _k, each filter coefficient (weighting coefficient) and _{w k0, w k1, w k2} , ···, w kL, when the FIR filter output and y _k, the relationship between the input and output follows (2)
It becomes like a formula. y _k = w _k0 x _k0 + w _k1 x _k1 + ... + w _kL x _kL (2) Further, the filter coefficient vector (weight vector) W
a _k, W _k = _{[w k0 w k1 w k2 ··· w} kL ] ^T ··· (3) and if the definition, the input-output relationship ^{_{is, y k = X k T W}} k ··· (4) It is described as. If the desired response is d _k , the error ε _{k from the} output is expressed as ε _k = d _k −y _k = d _k −X _k ^T W _k (5). Using these, the LMS algorithm is expressed as W _{k + 1} = W _k +2 με _k X _k (6). In equation (6), μ is a gain factor that determines the speed and stability of adaptation.

The FIR filter approximates the transfer characteristic of the spatial path from the microphone 14 to the microphone 12 in FIG. 2 as a linear characteristic. Therefore, if the accuracy of the approximation is improved, the adaptive filter output y is the noise component n _a.
, The noise component n _a can be canceled by subtracting this from the main input s + n _a , and the residual e approaches the signal component s. The noise-reduced output (residual e) is taken out through the output terminal 17 and sent to the voice recognition device 4 in FIG. Various methods have been considered for the voice recognition in the voice recognition device 4, but FIG.
A specific example of voice recognition using a word-based recognition method will be described with reference to.

In other words, FIG. 4 shows an example of a concrete structure of a voice recognition device which performs voice analysis to extract features of a voice, and performs recognition in word units. In FIG. 4, the voice signal supplied to the input terminal 31 is sent to an acoustic analysis or voice analysis block 32 to extract acoustic features, and then a normalized pattern generation block 33 outputs a voice corresponding to each word. While the pattern (NAT pattern) is generated, the voice section determination block 34 performs section determination, and the determination result is the pattern generation block 33.
Sent to. The normalization pattern obtained from the pattern generation block 33 is sent to the pattern matching block 35 and collated with the comparison pattern from the comparison pattern block 36, whereby the recognition result is taken out from the output terminal 37.

That is, in the voice analysis (acoustic analysis), B
Frequency analysis is generally performed using a PF (band pass filter) group or the like. This is because such a process similar to frequency analysis is performed on the basilar membrane at the back of the human inner ear. The main purpose of acoustic analysis is to extract phonological information related to phonemes, which are language description units. There are several types of power spectrum envelopes, formant frequencies, etc., which are considered to be related to the frequency spectrum in the amount including this information. Of these, the power spectrum envelope is suitable for practical use because it can be handled directly. The filter bank method, F
The FT (Fast Fourier Transform) method, the cepstrum method, and the like are available, and the filter bank method is most easily realized in terms of the amount of calculation. From this, in the example of FIG. 4, the power spectrum envelope obtained by the filter bank (for example, BPF group) is adopted as the feature amount.

Next, the pattern generation block 33 carries out normalization in order to remove elements due to individual differences such as sound intensity, duration, vocal cord characteristics, etc. from the voice analysis result. That is, some acoustic characteristics of voice strongly depend on individual differences such as sound intensity, duration, vocal cord sound source characteristics, etc. in addition to phonological information. These are elements that obviously interfere with the recognition and must be removed before the voice pattern is generated.

Power detection is effective for the voice section determination in the voice section determination block 34. However, since the power of the unvoiced sound rises poorly, it often fails to make a decision. In order to compensate for this, zero cross, that is, the number of times the signal crosses the 0 level within a fixed time is also detected.

In this way, the normalized pattern obtained from the normalized pattern generation block 33 is sent to the pattern matching block 35 and collated with the previously obtained comparison pattern from the comparison pattern block 36. , Give recognition result.

If another voice is superimposed on the voice to be recognized, the obtained power spectrum envelope obviously has a distribution different from that of the original voice, and the normalization pattern also changes. Therefore, the recognition result may be different from the command uttered by the user. In particular, when trying to recognize the voice received by the microphone 3 provided in the television receiver 2 as shown in FIG. 1, the speaker 8 which is a sound source other than the user 2 is used.
Since the voices from are mixed in at a relatively large level, misrecognition often occurs. Therefore, in the embodiment of the present invention,
By reducing the voice other than the command voice uttered by the user, that is, mainly the voice (noise component) from the speaker 8 by the adaptive noise reduction device 10 in FIG. 1 described above, and sending the noise-reduced signal to the voice recognition device 4. , Prevents false recognition.

Therefore, even an electric device having a sound source such as the speaker 8 such as the television receiver 2 can be controlled only by the sound. That is, the user 2 can control the operation of the television receiver 2 only by uttering a predefined command to (the microphone 3 of) the television receiver 2 and the conventional remote control unit can be operated. It becomes unnecessary. This makes it possible to completely avoid the drawbacks of losing sight of the remote control unit, the complexity of operation, the size of the unit, and the like.

Next, FIG. 5 shows a second embodiment of the present invention. The parts corresponding to those of the first embodiment of FIG. 1 are designated by the same reference numerals and the description thereof will be omitted. The second embodiment shown in FIG. 5 shows a concrete example in which the present invention is applied to a stereo or audio multiplex type television receiver. That is, the audio signal from the left (L) channel audio output circuit 6L is transmitted to the speaker 8 via the amplifier 7L.
Audio output circuit 6R for the right (R) channel sent to L
Is sent to the speaker 8R via the amplifier 7R. Considering that the voices uttered from these speakers 8L and 8R also wrap around to the microphone 3, the audio signal from the audio output circuit 6L for the left (L) channel is supplied to the adaptive filter 16L and the audio signal for the right (R) channel is supplied. The audio signal from the audio output circuit 6R is supplied to the adaptive filter 16R,
These adaptive filters 1 are sent as reference signals, respectively.
The outputs from 6L and 16R are sent to the adder 15 as subtraction signals. The outputs (residuals) from the adder 15 as the subtraction means are sent to the adaptive filters 16L and 16R, respectively.

In the configuration of FIG. 5, the adaptive filter 1
In 6L, since only the audio signal of the left (L) channel is input as the reference signal, the wraparound component estimated by this is only that of the left (L) channel. Similarly,
The wrap-around component of the right (R) channel is estimated by the adaptive filter 16R. As a result of subtracting these two estimated signals from the main input from the microphone 3, substantially only the command voice uttered by the user 2 is extracted and sent to the voice recognition device 4. Therefore, the same effect as that of the first embodiment can be obtained.

The present invention is not limited to the above embodiment, and for example, the electric equipment to be controlled is
Any device that has a sound source may be used, and in addition to television receivers, not only audio receivers such as radio receivers, cassette tape recorders with radios, VTRs (video tape recorders), etc., but also motors can be used. Similar effects can be obtained even when the present invention is applied to an air conditioner (air conditioner), a cooler, a fan, etc., in which noises, wind noises, and the like tend to be noises. In addition, the sound receiving means such as a microphone is not limited to the case where it is provided in the main body of the electric device,
You may make it arrange | position in the control exclusive device separately provided in order to control the electric equipment used as a control object. Furthermore, the voice recognition device is not limited to the example of FIG. 4, and various other types of voice recognition devices can be used.

[0035]

As is apparent from the above description, according to the voice recognition remote control device of the present invention, the voice recognition remote control device receives the voice by the voice receiving means provided in the vicinity of the electric equipment body having the sound source. The main input is supplied to the subtracting means, and the output from the subtracting means and the output from the adaptive filter to which the signal from the sound source is input are sent to the subtracting means. In the adaptive filter, the output from the subtracting means is supplied. The filter characteristic is controlled so as to minimize the power, the output from the subtraction unit is recognized by the voice recognition unit, and the electric device is controlled according to the recognition output. An output that is similar to the noise component from the sound source mixed with the main input from the voice receiving means is taken out, and this output is sent to the subtracting means and subtracted from the main input. , Only the signal component, such as substantially the voice is sent to the speech recognition unit, the recognition accuracy of the speech recognition unit is enhanced.

Therefore, when the remote control of the electric device having the sound source is performed by the command voice uttered by the user using the voice recognition, the sound generated by the sound source of the electric device is circulated and received by the voice receiving means. It is possible to prevent erroneous operation (erroneous recognition) caused by (sound pickup), and to realize remote control without the conventional remote operation unit (remote control unit or commander). Since this remote control unit is not needed, problems due to losing sight of the remote control unit, complicated operation of the remote control unit, being unfamiliar without using it, increasing the number of switches and making the unit larger, etc. Of course, the disadvantages of will also disappear.

[Brief description of drawings]

FIG. 1 is a block circuit diagram showing a schematic configuration of a first embodiment of a voice recognition remote control device according to the present invention.

FIG. 2 is a block circuit diagram for explaining the configuration and operation of the adaptive noise reduction device in FIG.

FIG. 3 is a block circuit diagram showing a specific configuration example of an adaptive filter in FIG.

FIG. 4 is a block diagram showing a specific example of an internal configuration of the voice recognition device in FIG.

FIG. 5 is a block circuit diagram showing a schematic configuration of a second embodiment of a voice recognition remote control device according to the present invention.

[Explanation of symbols]

 1 ... User 2 ... Television receiver 3 ... Microphone (voice receiving means) 4 ... Voice recognition device 5 ... Control device 6 ... ..Voice output circuit 8 ... Speaker (sound source) 10 ... Adaptive noise reduction device 15 ... Adder (subtraction means) 16 ... Adaptive filter 21 ... ..Filter unit 22 ... Adaptive algorithm unit 32 ... Speech analysis block 33 ... Normalized pattern generation block 35 ... Pattern matching block

Claims (1)

Claim: What is claimed is: 1. A voice receiving means provided in the vicinity of an electric equipment body having a sound source, and a subtracting means to which a main input obtained by the voice receiving means is supplied. The output from the subtraction means and the adaptive filter for controlling the filter characteristics so as to minimize the power of the output from the subtraction means while sending the output filtered by the sound source to the subtraction means. A voice recognition control device, comprising: a voice recognition unit supplied and a control unit for controlling the electric device according to a recognition output from the voice recognition unit.