JPH0754434B2 - Voice recognizer - Google Patents

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device.

2. Description of the Related Art In recent years, with the development of voice recognition technology, voice recognition devices are about to be put to practical use in various fields. However, in order to put them into practical use, there are various problems in actually using the recognition devices. It needs to be resolved. One of the practical problems is that when the SN ratio at the time of voice input is low, noise is erroneously detected as a voice section, and as a result, erroneous recognition is performed.

In the conventional voice recognition device, in order to solve the above-mentioned problems, the background noise is input in advance just before the voice is input and the power thereof is checked, and the threshold for detecting the voice section is set to be equal to or higher than the power. Then, the voice section is detected using the set threshold. By this method, the false recognition rate is reduced without erroneously detecting noise as a voice section even in an environment with a low SN ratio.

Hereinafter, the conventional speech recognition apparatus as described above will be described with reference to the drawings. FIG. 3 is a block diagram of a conventional registered word voice recognition device. In the figure,
1 is a voice input terminal, 2 is an analysis unit, 16 is a threshold value setting unit, 17 is a section detection unit, 18 is a matching unit, 19 is a recognition result output terminal, 20 is a registered voice buffer, 21 is an input voice buffer, 22 , 23 are switches. The operation of the speech recognition apparatus configured as described above will be described below.

First, at the time of registration, just before voice input, a background noise signal for a predetermined time is input from the voice input terminal 1, the power of the signal per unit time is calculated by the analysis unit 2, and the calculation result is input to the threshold setting unit 16. To be done. The threshold setting unit 16 obtains the average value of the powers calculated above, and sets the value obtained by adding a predetermined value (for example, 6 dB) to the average value on the left as the section detection threshold.

When inputting the registered word voice, the analysis unit 2 calculates the power and characteristic parameter of the signal for each unit time based on the signal input from the voice input terminal 1, and the power calculation result is input to the section detection unit 17 and the characteristic parameter is input. Audio buffer 21
Entered in. If the LPC cepstrum method is used as the analysis method, a predetermined number of cepstrum coefficients are calculated as the characteristic parameters. Next, the section detection unit 17 compares the power of the signal for each unit time with the section detection threshold set previously, and determines the portion where the signal power is continuously the section detection threshold or more for 60 msec or more as the voice section. However, even if the signal power is below the section detection threshold, there are 60 sections below the threshold.
If it does not continue for more than msec, it is regarded as a voice section. Next, the feature parameters for the determined voice section are input to the input voice buffer 21.
It is input by the user and stored in the registered voice buffer 20. The above processing after the voice input is repeated for all the recognized words.

Next, at the time of recognition, as in the case of registration, the section detection threshold is set from the background noise, and then the input speech is analyzed to detect the speech section. The analysis method and the section detection method are the same as at the time of registration. After the voice section is detected, the collating unit 18 collates the registered voice with the input voice, and outputs the word indicating the shortest distance from the recognition result output terminal 19 as the recognition result. Note that the switch 22 is provided in the threshold setting unit 16 when noise is input immediately before voice input,
At the time of voice input, the section detection unit 17 and the input voice buffer 21 operate to input the calculation result. Switch 23
It operates so as to input the characteristic parameter to the registration buffer 20 at the time of registration and to the collating unit 18 at the time of recognition.

DISCLOSURE OF THE INVENTION Problems to be Solved by the Invention However, in the above-described configuration, although it is possible to remove noise irrespective of the change in noise power, the start and end positions of the voice section are changed in accordance with the change in noise or vocal power. Due to the deviation, when the registered voice or standard voice is uttered and the input voice is uttered differently, matching is performed in different voice sections, so that there is a problem in that erroneous recognition is likely to occur.

In view of the above-mentioned problems, the present invention provides a voice recognition device capable of preventing a deviation of a voice section between a registered voice or a standard voice and an input voice and reducing erroneous recognition due to a difference in situation.

Means for Solving the Problems In order to achieve the above object, the voice recognition device according to claim 1, an analysis unit that detects the power of an input signal per unit time, an SN ratio calculation unit that calculates the SN ratio, A threshold setting unit that determines a section detection threshold in consideration of the SN ratio, a section detection unit that detects a voice section of the input signal by the determined threshold, a section correction unit that corrects a registered voice or a standard voice section, It is composed of a collating unit that collates a registered voice or standard voice with an input voice and outputs a recognition result.

The speech recognition apparatus according to claim 2 further includes an analysis unit that detects the power of the input signal per unit time, and a threshold setting unit that sets the section detection threshold in consideration of the peak value of the signal power and the noise power value. , A section detection unit that detects the voice section of the input signal according to the set threshold value, a section correction unit that corrects the section of the registered voice or standard voice, and the recognition result by comparing the registered voice or standard voice with the input voice And a collating unit for outputting

According to the speech recognition apparatus of claim 1, the SN ratio calculation unit performs SN
After calculating the ratio, the threshold determination unit determines a value equal to or higher than the noise power in an environment where the SN ratio is low, and a value obtained by subtracting a predetermined value from the peak value in an environment where the SN ratio is high. Is used to detect the section of the input voice, and the section correction unit compares the registered voice or the standard voice with the input voice using the threshold value.

According to the speech recognition apparatus of claim 2, the threshold value setting unit compares a value obtained by adding a predetermined value to the noise power value with a value obtained by subtracting the predetermined value from the peak value, and sets the larger value as the threshold value. After setting, the section detection unit detects the section of the input voice using the threshold, and the section correction unit corrects the registered voice or the standard voice section with the threshold, and the matching unit detects the registered voice or the standard voice. Match with the input voice.

Embodiment 1 FIG. 1 is a block diagram of a registered word voice recognition device in a first embodiment of the present invention (corresponding to the invention described in claim 1).

In the figure, 1 is a voice input terminal, 2 is an analysis unit, 3 is a temporary threshold value setting unit, 4 is an SN ratio calculation unit, 5 is a threshold value setting unit, 6 is a section detection unit, 7 is a section correction unit, and 8 is a collation. Reference numeral 9 is a registered voice buffer, 10 is an input voice buffer, 11 is a recognition result output terminal, 12 and 13 are switches, and the same elements as in the conventional example (see FIG. 3) are given the same numbers. There is.

The operation of the speech recognition apparatus configured as described above will be described below.

First, at the time of registration, immediately before voice input, a background noise signal for a predetermined time is input from the voice input terminal 1, and the analysis unit 2 calculates the signal power for each unit time. The calculation result is input to the temporary threshold value setting unit 3. The temporary threshold setting unit 3 obtains the average value of the powers, and the average value on the left is a predetermined value (in the present embodiment,
The added value is used as the temporary section detection threshold.

When inputting the registered word voice, the analysis unit 2 calculates the power of the signal and the characteristic parameter for each unit time based on the signal input from the voice input terminal 1. The power calculation result is
In the SN ratio calculation unit 4, the feature parameter is the input voice buffer.
Entered in 10. The analysis method is the same as the conventional example.
In the SN ratio calculation unit 4, a signal portion equal to or higher than the previously set provisional section detection threshold is set as a provisional voice section, and the ratio of the average value of the peak value in the provisional voice section and the previously calculated noise power is calculated as SN. If the SN ratio is less than or equal to a predetermined value (24 dB in this example), the speaker is instructed to retry registration, and the above registration process is restarted from the beginning.

If the SN value is 24 dB or more, the threshold setting unit 5 determines a value obtained by subtracting a predetermined value (18 dB in this embodiment) from the peak value as the detection threshold. The section detection unit 6 compares the signal power for each unit time with the detection threshold to detect a voice section. The section detection method is the same as the conventional example. Next, the characteristic parameters for the voice section are input from the input voice buffer 10 and registered in the registered voice buffer 9. The above processing after the input of the registered voice is repeated for all the recognized words.

Next, at the time of recognition, as in the case of registration, the provisional section detection threshold is set from the background noise, then the input voice is analyzed, and the result is input to the SN ratio calculation unit 4 and the input voice buffer 10. The SN ratio calculation unit 4 calculates the SN ratio in the same manner as at the time of registration, and the result is the threshold setting unit 5
To enter. In the threshold value setting unit 5, if the SN ratio is 24 dB or more, the value obtained by subtracting 18 dB from the peak value is set as the threshold value, and if the SN ratio is 24 dB or less, the provisional section detection threshold value is set as the threshold value, and then the section detection unit is set. In step 6, the voice segment is detected using the threshold value on the left.
The section detection method is the same as that at the time of registration.

Next, the section correcting unit 7 does not correct the registered voice section registered when the SN ratio is 24 dB or more. Only when the SN ratio is 24 dB or less, the section detection of the registered voice is performed again with the threshold value. Start over. Next, the matching unit 8 matches the registered voice with the input voice, and outputs the word indicating the shortest distance as a recognition result.
Output from 11. It should be noted that the switch 12 is set to the temporary threshold value setting unit 3 when noise is input immediately before voice input, and when the voice is input.
It operates so as to input the calculation result to the SN ratio calculation unit 4 and the input voice buffer 10. The switch 13 operates to input the characteristic parameter to the registration buffer 9 at the time of registration and to the section correction unit 7 at the time of recognition.

As described above, according to the present embodiment, the SN ratio calculating unit 4 calculates the ratio between the peak value of the signal and the average power value of the noise, and the threshold setting unit 5 determines that the SN ratio is equal to or less than a certain value. A value obtained by adding a predetermined value to the noise power value is determined as a threshold value when the SN ratio is a predetermined value or more and a predetermined value is subtracted from the peak value, and the section detection unit 6 uses the threshold value to detect the input speech. The section is detected, the section correcting unit 7 corrects the registered voice with the threshold value, and the collating unit 8
By checking the registered voice and the input voice with
It is possible to prevent the voice section from deviating from the registered voice and the input voice, and reduce erroneous recognition due to the difference in the situation.

FIG. 2 is a block diagram of a registered word voice recognition device in a second embodiment of the present invention (corresponding to the invention described in claim 2).

In the figure, 1 is a voice input terminal, 2 is an analyzing unit, 3 is a temporary threshold setting unit, 14 is a threshold setting unit, 6 is a section detecting unit, 7 is a section correcting unit, 8 is a collating unit, and 9 is for registered voice. A buffer, 10 is an input voice buffer, 11 is a recognition result output terminal, 12 and 15 are switches, and the same elements as those in the above-mentioned embodiment are given the same numbers.

The operation of the speech recognition apparatus configured as described above will be described below.

First, at the time of registration, immediately before voice input, a background noise signal for a predetermined time is input from the voice input terminal 1, and the analysis unit 2 calculates the signal power for each unit time. The calculation result is input to the temporary threshold value setting unit 3. The temporary threshold value setting unit 3 calculates the average value of the powers, and the average value shown on the left is a constant value (in the present embodiment,
The added value is used as the temporary section detection threshold.

When inputting the registered word voice, the analysis unit 2 calculates the power of the signal and the characteristic parameter for each unit time based on the signal input from the voice input terminal 1. The power calculation result is stored in the threshold setting unit 14, and the characteristic parameter is stored in the input voice buffer.
Entered in 10. The analysis method is the same as in the above-mentioned example. In the threshold value setting unit 14, a signal portion equal to or greater than the previously set provisional section detection threshold is set as a provisional voice section, and a value obtained by adding a predetermined value (18 dB in this embodiment) from the peak value in the provisional voice section The temporary section detection threshold value calculated in step 1 is compared, and if the latter value is large, the speaker is instructed to retry registration, and the above registration process is restarted from the beginning.

If the former value is large, this former value (peak value −18d
B) is set as the detection threshold value, and the section detection unit 6 compares the signal power per unit time with the detection threshold value to detect the voice section. The section detection method is the same as the conventional example. Next, the characteristic parameters for the voice section are input from the input voice buffer 10 and registered in the registered voice buffer 9. The above processing after the registered voice is repeated for all the recognized words.

Next, at the time of recognition, as in the case of registration, the temporary interval detection threshold is set from the background noise, then the input voice is analyzed, and the result is input to the threshold setting unit 14 and the input voice buffer 10. In the threshold value setting unit 14, the signal portion equal to or more than the section detection threshold is a temporary voice section,
The value obtained by subtracting 18 dB from the peak value in the temporary voice section is compared with the previously calculated temporary section detection threshold value, and the larger one of the two values is set as the threshold value, and then the section detection unit 6 uses the threshold value on the left. To detect the voice section. The section detection method is the same as that at the time of registration.

Next, the section correction unit 7 does not correct the registered voice section registered when the threshold is set to a value obtained by subtracting 18 dB from the peak value, and only when the threshold is set to the temporary section detection threshold. , The registered voice section detection is performed again with the above threshold value. Next, the collating unit 8 collates the registered voice with the input voice, and outputs the word indicating the shortest distance from the output terminal 11 as a recognition result. The switch 15 operates to input the calculation result to the temporary threshold setting unit 3 when noise is input immediately before voice input, and to the threshold setting unit 14 and the input voice buffer 10 when voice is input. The switch 13 operates to input the characteristic parameter to the registration buffer 9 at the time of registration and to the section correction unit 7 at the time of recognition.

As described above, according to the present embodiment, after comparing the value obtained by adding 6 dB to the noise power value in the threshold setting unit 14 and the value obtained by subtracting 18 dB from the peak value, the larger one is determined as the threshold value, The section detector 6 detects the section of the input voice using the above threshold,
By correcting the registered voice section with the threshold value in the section correction unit, and by checking the registered voice and the input voice in the matching unit, it is possible to prevent the voice section from deviating from the registered voice and the input voice, depending on the situation. False recognition can be reduced. Further, the present embodiment can expect the same effect as the first embodiment without the trouble of calculating the SN ratio.

Advantageous Effects of Invention The speech recognition apparatus according to claim 1 calculates the SN ratio in the SN ratio calculation unit, and the threshold value determination unit outputs a value equal to or higher than noise power in an environment where the SN ratio is low, and a value higher than the peak value in an environment where the SN ratio is high. After setting the value obtained by subtracting the predetermined value as the threshold value, the section detection unit detects the section of the input voice using the threshold value, and the section correction unit corrects the registered voice or standard voice section with the threshold value and collates. By comparing the registered voice or the standard voice with the input voice by the section, it is possible to prevent a deviation of the voice section between the registered voice and the input voice, and to reduce erroneous recognition due to a difference in the situation.

Further, in the voice recognition apparatus according to claim 2, after comparing the value obtained by adding a predetermined value to the noise power value with the threshold setting unit and the value obtained by subtracting the predetermined value from the peak value, the larger one is determined as the threshold, The section detection unit detects the section of the input voice using the threshold value, the section correction unit further corrects the registered voice or standard voice section with the threshold value, and the matching unit compares the registered voice or standard voice with the input voice. By performing the collation, it is possible to prevent the deviation of the voice section between the registered voice or the standard voice and the input voice, and reduce the erroneous recognition due to the difference in the situation. Further, compared to the above invention, the same effect can be expected without the trouble of calculating the SN ratio.

[Brief description of drawings]

FIG. 1 is a block diagram of a voice recognition device in a first embodiment of the present invention, FIG. 2 is a block diagram of a voice recognition device in a second embodiment of the present invention, and FIG. 3 is a voice recognition device in a conventional example. It is a block diagram of. 2 ... Analysis unit, 4 ... SN ratio calculation unit, 5, 14 ... Threshold setting unit, 6 ... Section detection unit, 7 ... Section correction unit, 8 ... Collation unit.

Claims (2)

[Claims] 1. An analysis unit for detecting the power of an input signal per unit time, an SN ratio calculation unit for calculating the ratio of voice power and noise power (hereinafter referred to as SN ratio), and section detection in consideration of the SN ratio. A threshold value determining unit that determines a threshold value, a section detecting unit that detects a voice section of the input signal based on the determined threshold value, a section correcting unit that corrects a registered voice or standard voice section, and a registered voice or standard voice is input. The SN ratio calculator calculates the SN ratio, and the threshold value determiner calculates the SN ratio. In the environment, set a value obtained by subtracting a predetermined value from the maximum power value of the signal (hereinafter called the peak value) as the threshold,
The section detection unit detects the section of the input voice using the threshold value, the section correction unit corrects the registered voice or standard voice section using the threshold value, and the matching unit compares the registered voice or standard voice with the input voice. A voice recognition device characterized by being configured to perform matching. 2. An analysis unit for detecting power of an input signal per unit time, a threshold setting unit for setting a section detection threshold in consideration of a peak value of signal power and a noise power value, and a set threshold. A section detection unit that detects a voice section of the input signal, a section correction unit that corrects a section of registered voice or standard voice, and a collation unit that collates the registered voice or standard voice with the input voice and outputs a recognition result. The threshold value setting unit compares a value obtained by adding a predetermined value to the noise power with a value obtained by subtracting the predetermined value from the peak value of the signal, and sets the larger value as the threshold value, and the section detection unit sets the above value. The input voice section is detected using a threshold value, the section correction unit corrects the registered voice or standard voice section with the threshold value, and the matching unit performs matching between the registered voice or standard voice and the input voice. Voice recognition characterized by Apparatus.