JPH0228878B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a voice recognition circuit, and more particularly to a voice recognition circuit suitable for use in voice remote control devices such as television receivers.

Speech recognition circuits are used in devices whose operations are controlled by humans using voice. Here, the speech recognition circuit cuts out the input speech, extracts the characteristics of the input speech,
This characteristic parameter is compared with the standard voice characteristic parameter to identify which of the operations of the device the input voice is a command for controlling. However, when switching the receiving channel by conveying a command word in voice on a television receiver, etc., the sound reproduced by the television receiver itself is used as the command voice given from the outside (for example, when switching the receiving channel from channel 2 to channel 4). "Channel 4"
This makes it difficult to input commands accurately.

Therefore, conventionally, as shown in Japanese Patent Application No. 54-91913, the command voice input section has two types: directional and non-directional.
The microphones are connected in opposite phases to each other, and the microphones are connected in opposite phases to each other.
Increase the SN ratio of the command voice within the directional range,
A method that sets a large tolerance value for identifying important command words such as "channel", "dengen", and "volume", and when these are identified, the speaker output is temporarily cut off or greatly attenuated. is proposed. This method requires two microphones with different directivity, and because it has a large tolerance for identifying important command words, there are problems such as erroneous muting due to command-like words in the voice coming from the speaker. Ta.

In addition, as shown in Japanese Patent Application No. 54-67142, the signal from the microphone that collects the voice from the speaker and the command voice, and the signal that drives the speaker are adjusted by passing them through a level adjuster and a phase adjuster. A method has been proposed in which the command signal is applied to a signal addition circuit in opposite phase to cancel out the speaker sound, and only the command sound is taken into the recognition circuit. However, with this method, it is difficult to completely adjust the level and phase in all frequency bands, and the speaker sound cannot be completely canceled, leading to malfunctions in command voice recognition.

An object of the present invention is to eliminate the above-mentioned drawbacks of the prior art, and to provide a low-cost voice recognition circuit that enables voice command input without malfunction even when used in devices such as television receivers that have their own voice emitting means. It is about providing.

In order to make it possible to input voice commands to a device such as a television receiver that has its own sound emitting means, the present invention detects the presence of voice for the number of command voices + ms from the sound collection signal of a microphone attached to the device. The presence of a command voice other than the voice from the voice emitting means is detected based on the difference in the zero-crossing wave number of each voice, mutating is applied to the voice emitting means, and the voice after the mutating is input into the voice recognition unit as a command voice. , is a voice recognition circuit that identifies the command word. In particular, we propose a simple method for reliably muting sound emitting means such as speakers, thereby improving the recognition rate of speech recognition circuits.

Embodiments of the present invention will be described below based on the drawings. FIG. 1 is a block diagram showing one embodiment of the present invention, showing the main parts of a television receiver that allows functions such as switching reception channels, adjusting volume, and turning on the power to be performed by voice commands. It is also a block diagram. In FIG. 1, 1 is the speaker of the television receiver itself, 2 is a muting circuit that mutates the television audio, 3 is a television audio amplification circuit that amplifies the demodulated audio signal of the receiving station, and 4 is the audio command word. A microphone for collecting sound, 5 a microphone amplification circuit for amplifying the microphone signal, 6 and 7 perform infinite clipping of the demodulated audio signal of the receiving station and the microphone signal up and down around the point of zero sound pressure;
A zero-crossing wave circuit that converts this into a rectangular wave pulse train,
8 and 9 are counting circuits for counting the above-mentioned zero-crossing wave numbers, 1
0 is a comparison circuit that compares the count values of the above-mentioned counting circuit;
Reference numeral 11 denotes a voice recognition unit that identifies command voices; 12 outputs a muting signal to a muting circuit, a command voice input signal to the voice recognition unit, and a control signal for switching channels of a television receiver, etc. based on the recognition result of the command voice. It is a control circuit.

The operation shown in FIG. 1 will be explained by taking as an example a case where the current reception channel 2 is changed to channel 4 by a voice command. The demodulated audio signal of channel 2 is now amplified by the television audio amplification circuit 3, and is reproduced by the speaker 1 without being muted by the muting circuit 2. The microphone 4 collects the demodulated sound from the speaker 1 as ambient sound, converts it into an electrical signal, and supplies it to the microphone amplifier circuit 5. When there is no command voice, the signal input to the zero-crossing wave circuit 6 is a signal that drives the speaker, and the signal input to the zero-crossing wave circuit 7 is a signal obtained by collecting the signal from the speaker 1 with the microphone 4. be. Therefore, although there are some differences in amplitude and phase, the signals are almost the same, and the rectangular wave pulse trains output from the zero-crossing wave circuits 6 and 7 are not affected by the slight difference in amplitude and are almost the same. Counting circuits 8 and 9 count the number of rectangular wave pulses from zero-crossing wave circuits 6 and 7 at a certain period, for example, every 40 ms. In this case, since these count values have a counting period set to a certain time, a slight shift in the signal phase does not affect the count values. From the above, when there is no voice command, the count values of the counting circuits 8 and 9 are the same.

Next, consider the case where a command voice called "Channel 4" is uttered while the demodulated voice is being played back by speaker 1. Microphone 4 collects the superimposed voice of the demodulated voice from speaker 1 and the command voice as ambient sound. The signal is amplified via the microphone amplifier 5 and input to the zero crossing wave circuit 7. At this time, only demodulated audio is input to the zero crossing wave circuit 6. Therefore, the count values of each zero-crossing wave will differ greatly.

FIG. 2 shows the counts of the zero-crossing wave circuits 6 and 7 and the difference between them based on the results of experiments conducted by the inventor. In the experiment, the counting period was set to 40 milliseconds, and the command voice ``Channel 4'' was uttered while the television receiver was making an announcement voice saying ``What is the weather in the Kanto region tomorrow?''. As is clear from FIG. 2, at the time when the voice command is superimposed on the announcer's voice, the count values of the outputs of the zero crossing wave circuits 6 and 7, that is, the count values of the counting circuits 8 and 9, are significantly different.

Returning to FIG. 1, the comparison circuit 10 is the counting circuit 8, 9
The count values of are compared every counting cycle, and the comparison results are sent to the control circuit 12 every cycle.

The control circuit 12 constantly monitors the results of the comparator circuit 10, and can tell from the results of the comparator circuit 10 when the count values of the counting circuits 8 and 9 are significantly different, and when the count values are significantly different, the first number + ms
The mutating signal is transmitted between the mutating circuit 2 and the mutating circuit 2.
At the same time, a command voice capture signal is sent to the voice recognition unit 11. By doing this, the voice recognition unit 11 can take in the voice excluding the first number + ms of the command voice "channel 4" without being interfered with by the demodulated voice.

The speech recognition unit 11 includes a feature extraction circuit 1 that extracts features of a speech signal, as shown in FIG. 3, for example.
3. Standard pattern storage circuit 1 that is registered in advance and stores characteristic patterns of standard voices to be identified.
4. It consists of a recognition processing circuit 15 that compares the characteristic pattern extracted from the input voice with a standard pattern and identifies the input voice based on the degree of similarity. The command voice captured by the voice recognition unit 11 is converted into a feature pattern by the feature extractor 13, and the standard voice registered in the standard voice storage circuit 14 as a standard pattern, such as "Channel 1", "Channel 2", etc. and is compared by the recognition processing circuit 15, and depending on the degree of similarity, it is selected as "channel 4".
, and sends a symbol representing "channel 4" to the control circuit 12.

Returning to FIG. 1, the control circuit 12 includes the voice recognition section 11
Based on this, a control signal (not shown) for changing the local oscillation frequency of the television receiver tuner (not shown) is sent to the tuner in order to switch the channel to channel 4. Thereafter, the control circuit 12 sends a muting release signal to the muting circuit 2 to release the muting. This completes the channel 4 reception operation using the command voice.

The command voice input signal to the voice recognition unit 11 has a large difference in the count values of the counting circuits 8 and 9, and continues for an arbitrary fixed period of several hundred milliseconds to several seconds from the time when the muting signal is sent. Of course, its duration must match the time of the command voice used.

Please refer to FIG. 4, which shows an outline of the operation timing of each part in the embodiment of FIG. 1, based on the counting period, together with the counted values.

In FIG. 1, the zero crossing wave circuits 6 and 7 can be easily constructed as Schmitt circuits using operational amplifiers, for example. This Schmitt circuit creates a dead zone by setting the Schmitt level a little higher than the level of zero sound pressure.
It is also possible to more reliably detect the command voice by making it insensitive to background noise having a smaller amplitude than the demodulated voice.

The comparator circuit 10 may be composed of a digital comparator and may be configured to send only the magnitude relationship between the count values of the counting circuits 8 and 9 to the control circuit 12, or may be composed of a subtraction circuit and the result of the subtraction is sent to the control circuit 12. It may be a configuration.

Although the control circuit 12 may be constructed of random logic, it is clear that similar operations can be performed using a microprocessor. In this case, it is also clear that if the processing time of the microprocessor is allowed, the counting operations of the counting circuits 8 and 9 and the comparison operation of the comparator circuit 10 can be performed by the software of the microprocessor. .

Further, the microphone amplifier 5 may be provided with a variable gain control function so that the input amplitude levels to the zero-crossing wave circuits 6 and 7 are substantially the same. This is because although the present invention is essentially independent of amplitude levels, extreme differences in amplitude will result in differences in zero-crossing counts. This variable gain control was applied in a patent application filed in 1973.
A much coarser gain adjustment is required than that shown in No. 67142. It is clear that the microphone 4 can also be another electroacoustic transducer.

FIG. 5 shows a block diagram of another embodiment of the invention. In FIG. 5, the same symbols as in FIG. 1 indicate the same parts. The basic operation is the same as that shown in FIG. 1, so the explanation will be omitted. The difference between FIG. 1 and FIG. 5 is that the input of the zero-crossing wave circuit 6 is connected to the output of the muting circuit 2, and the demodulated audio that has passed through the muting circuit 2 is guided to the zero-crossing wave circuit. The output of the control circuit 12 is also connected to the control circuit 12. Fifth
In the figure, the command voice capture signal is sent to the control circuit 12.
Therefore, the signal is sent from the time when the count values of the counting circuits 8 and 9 become significantly different (the time when the mutating signal is sent out) until the time when the count value of the counting circuit 9 becomes zero or below a certain threshold value. This is because from the time when muting is applied, the only voice collected by the microphone 4 is the command voice, the signal input to the zero-crossing wave circuit 6 disappears, and the signal input to the zero-crossing wave circuit 7 is only the command voice. This is because the command voice detection can be determined based on the count value of the output of the zero-crossing wave circuit 7, that is, the count value of the counting circuit 9. In other words, the control circuit 12 makes it possible to automatically send out the command voice capture signal (ie, cutting out the command voice) of the voice recognition unit 11 in accordance with the time length of the command voice. As a result, compared to the case where the command voice capture signal is fixed as in the embodiment of FIG. 1, there is less risk of noise other than the command voice being introduced into the recognition unit 11, and the recognition rate can be improved.

FIG. 6 shows a block diagram of still another embodiment of the present invention. In FIG. 6, the same reference numerals as in FIG. 5 indicate the same parts. The basic operation is the same as that shown in FIG. 5, so the explanation will be omitted. In FIG. 6, 16 and 17 indicate band pass filters. This bandpass filter normally passes only the voice band 300Hz to 3000KHz, and thereby prevents noise other than voice from being converted into zero-crossing waves in the zero-crossing wave circuits 6 and 7, and also prevents noise from being converted into zero-crossing waves by the zero-crossing wave circuits 6 and 7. 11 is designed to input only audio signals. 18 and 19 are differential circuits. This outputs the difference between a signal and a signal delayed by a certain time as a signal.

FIG. 7 shows the input waveform and output waveform to the zero-crossing wave circuit. In FIG. 7, the waveform indicated by (1) is a demodulated voice signal with a long period, and the waveform (2) is a command voice signal with a short period. In this case, the input to the zero-crossing wave circuit 6 shown in FIGS. 1 and 5 is the waveform shown in (1), and the input to the zero-crossing wave circuit 7 is the waveform shown in (1) and (2). This is the waveform (3) that is a composite of the waveforms shown in FIG. The outputs of the zero crossing wave circuits 6 and 7 are shown in a and b, respectively. Counting circuits 8 and 9 count the a and b waveforms. The waveforms shown in (4) and (5) are the waveforms shown in (1) and (3) passed through a differential circuit, and the waveforms shown in c and d are the waveforms shown in (4) and (5) passed through the zero crossing. This is the waveform passed through the wave circuit.

In FIG. 6, the inputs to the zero crossing wave circuits 6 and 7 have waveforms (4) and (5), and the counting circuits 8 and 9 count the waveforms shown as c and d. It is clearly seen from the figure that the difference in the number of zero crossings between the waveforms c and d is greater than the difference in the number of zero crossings between the waveforms a and b. This is better than the circuit shown in Figures 1 and 5.
This shows that the circuit shown in the figure can more clearly detect the presence of a command voice. It is clear that the difference circuits 18 and 19 shown in FIG. 6 can obtain the same effect as a differentiation circuit from their characteristics. It is also possible to obtain the input to the speech recognition section 11 from the output of the difference circuit 19, and in this case there is no influence on the speech recognition section.

Although the above explanation has been made based on voice control of a television receiver, the present invention is applicable to any device to be controlled as long as it is equipped with voice emitting means.

According to the present invention, since the command voice is detected based on the difference in the number of zero crossings within a certain time interval, it is possible to reliably detect the command voice with a low-cost circuit configuration without being affected by amplitude and phase shifts. Become. Furthermore, by muting the voice emitting means of the device based on this detection signal, the command voice can be accurately input to the voice recognition section, thereby preventing erroneous recognition. It is also possible to use this detection signal as a command voice cutout signal. As described above, the present invention is extremely effective as a voice recognition circuit for equipment having a voice emitting means.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention;
Fig. 2 is an explanatory diagram showing the experimental results of the zero-crossing wave count used in the present invention, Fig. 3 is a detailed block diagram of the speech recognition section, and Fig. 4 is an outline of the operation timing of each part in the embodiment of Fig. 1. FIG. 5 is a block diagram showing another embodiment of the present invention, FIG. 6 is a block diagram showing yet another embodiment of the present invention, and FIG. 7 is an input waveform to the zero-crossing wave circuit. and a waveform diagram showing output waveforms. Explanation of symbols, 1... Speaker, 2... Muting circuit, 4... Microphone, 6, 7... Zero crossing wave circuit, 8, 9... Counting circuit, 10... Comparing circuit, 11...
Speech recognition unit, 12...control circuit, 16, 17...bandpass filter, 18, 19...difference circuit.

Claims (1)

[Claims] 1. A voice that can identify and recognize the second voice when a first voice generated from a specific sound emitting device and a second voice generated from another source are input. The recognition circuit includes an electro-acoustic transducer that collects ambient sounds including the first and second sounds, and an electric signal waveform from the transducer is inputted, and the waveform is inputted upward and downward around a point of zero sound pressure. A first zero-crossing wave circuit converts the pulse train into a rectangular wave pulse train obtained by infinite clipping and outputs the pulse train. A second pulse train that converts into a rectangular wave pulse train obtained by performing infinite clipping up and down around a point and outputs it.
a zero-crossing circuit; a first counting circuit that counts the rectangular wave pulse train from the first zero-crossing circuit;
It comprises a second counting circuit that counts the rectangular wave pulse train from the second zero-crossing circuit, and a comparison circuit that compares each counting result in both of the counting circuits, and the comparison results are different. The first sound generated from the voice emitting device is reduced based on the above, and the converted output signal of the second sound from the electroacoustic transducer after that time is taken into a speech recognition unit for speech recognition. A speech recognition circuit characterized by: 2. The speech recognition circuit according to claim 1, wherein a bandpass filter is connected in series to each of the first and second zero-crossing circuits. 3. The speech recognition circuit according to claim 1 or 2, characterized in that a difference circuit or a differential circuit is connected in series to the first and second zero-crossing circuits, respectively. circuit.