JPS6250838B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a speech recognition device that can distinguish human speech, and is particularly characterized by a voice registration method of a speech recognition device for a specific speaker.

FIG. 1 is a block diagram showing an existing speech recognition device. In the figure, 1 is a microphone that converts voice into an electrical signal, 2 is a feature extractor that extracts a voice feature pattern from the voice signal input from the microphone 1, and a voice area detector 3.
, a spectrum extractor 4, and a zero-cross detector 5
and a pattern creation section 6. still,
The audio region detector 3 monitors changes in the amplitude of the input signal and detects the time region in which the audio signal appears. The above spectrum extractor 4
The method extracts a frequency spectrum from an input audio signal and obtains a spectrum value for each frequency that includes formant information where the characteristics of the audio appear. The zero-crossing detector 5 detects the number of zero-crossings in the high-frequency components of the input audio signal, and determines whether the signal is voiced or unvoiced based on the magnitude of this. The pattern creation section 6 creates a characteristic pattern including voice spectrum information and voiced/unvoiced information based on the respective detectors 3, 4, and 5. Reference numeral 7 denotes a registered pattern memory in which standard feature patterns of speech are stored in advance as registered patterns for a large number of words, and this registered pattern is used as a registered pattern for the voice signal inputted from the microphone 1 in the registration mode performed in advance. This is obtained by the feature extractor 2 based on the above. Reference numeral 8 denotes a recognition processing unit that compares and recognizes the characteristic pattern of the input voice obtained from the feature extractor 2 and a large number of registered patterns in the registered pattern memory in the recognition mode for recognizing voice. This is to obtain an output signal corresponding to the input audio.

Connect such a voice recognition device to your home phone,
Attempts have been made to have this voice recognition device recognize the voice of a person calling home from outside the home, and to use this recognition result to control the operation of desired electrical equipment in the home. ing.

In this case, as long as a voice recognition device for a specific speaker as described above is used, the registration mode will be performed at home, and will be based on the voice signal with a natural voice band (100Hz to 600Hz) obtained from the microphone. Registered patterns are stored. For this reason, in the recognition mode, there is a considerable difference between the input pattern based on a voice signal whose frequency band is limited to 300Hz to 3400Hz due to the transmission characteristics specific to telephone lines, and the above registered pattern of the same voice. , it was not possible to obtain a recognition rate worthy of practical use.

In other words, the difference between the input pattern and the registered pattern that occurs due to the difference in the frequency band of the audio signal is
The presence or absence of spectral values in the low range below 300 Hz and the high range above 3400 Hz in the spectrum extractor 4, and the state of change in the amplitude of the low range component with large audio energy in the audio region detector 3. This is caused by an error in the time domain of the voice due to the difference, and a voiced/unvoiced determination error due to the difference in the number of zero crosses of the high frequency component in the zero cross detector 5.

The present invention has been made in view of this point, and aims to equalize the frequency bands of the audio signal obtained in the registration mode and the audio signal obtained in the recognition mode.

FIG. 2 shows a block diagram of the speech recognition device of the present invention. In the same figure, 1, 2, 7, and 8 are the first
Like the existing equipment shown in the figure, Microphon,
Feature extractor, registered pattern memory, recognition processing unit,
It shows. 9 is a telephone outside the home; 10 is a telephone line including a telephone exchange to which the telephone 9 is connected; one of these lines is connected to the home. 11 is a bandpass filter that limits the frequency band of the audio signal from the microphone 1;
As shown in the frequency characteristic diagram of Fig. 3, the telephone line 10
It has a passband of 300Hz to 3400Hz, which is equivalent to the transmission characteristics of . S ₁ and S ₂ are first and second changeover switches for interlocking mode changeover, and in the registration mode, these switches S ₁ and S ₂ are both on the b terminal side, and the microphone The audio signal from 1 is frequency-limited by the bandpass filter 11 and inputted to the feature extractor 2 via the first changeover switch _S1 , and the audio obtained from the feature extractor 2 is registered. The pattern is the second changeover switch
It is stored in the registered pattern memory 7 via _S2 .
In addition, in the recognition mode, these switches S ₁ and S ₂ are both on the a terminal side, and the voice signal obtained from the telephone 9 outside the home via the telephone line 10 is switched to the first switch.
The voice input pattern is inputted to the feature extractor 2 via the changeover switch _S1 , and the input pattern of the voice at this time obtained from the feature extractor 2 is inputted to the recognition processing section ₈ via the second changeover switch S2.

According to the above-described configuration, the registered pattern memory 7 stores an audio signal in the natural audio frequency band (100Hz to 6000Hz) obtained from the microphone 1 by using the bandpass filter 11 to adjust the transmission characteristics of the telephone line 10. This means that registered patterns extracted based on the equivalent passband limited to 300Hz to 3400Hz are stored for a large number of words. The recognition processing unit 8 uses these registered patterns to
By comparing and recognizing characteristic patterns based on voice signals obtained from a telephone 9 outside the home via the telephone line 10, and outputting a control signal corresponding to the input voice at this time, for example, "Light On". The gate lights at home are turned on.

As is clear from the above description, the speech recognition device of the present invention is equipped with a filter circuit having a frequency pass band equivalent to the transmission frequency characteristics of a telephone line, and in the registration mode, the speech recognition device receives information from a microphone through the filter circuit. This system stores a large number of registered patterns based on the voice signals obtained from the telephone line, and in the recognition mode, compares these registered patterns with input patterns based on the voice signals obtained from the telephone line to recognize the input voice at this time. Therefore, it is possible to equalize the frequency bands of the audio signal obtained in the registration mode and the audio signal obtained in the recognition mode.

Therefore, it is possible to avoid a decrease in the recognition rate of this type of speech recognition device due to the transmission characteristics of the telephone line, and the remote control of various electrical devices using the telephone becomes reliable.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an existing speech recognition device, FIG. 2 is a block diagram showing a speech recognition device of the present invention, and FIG. 3 is a frequency characteristic diagram of a bandpass filter used in the device of the present invention. , 1 is a microphone, 2 is a feature extractor, 7 is a registered pattern memory, 8 is a recognition processing section, 9 is a telephone, and 10 is a bandpass filter.

Claims (1)

[Claims] 1. A feature extractor that extracts a voice feature pattern from an input voice signal, a registered pattern memory in which a large number of registered patterns are stored in advance, and a It consists of a recognition processing unit that compares and recognizes input patterns, and a filter circuit that has a frequency pass band equivalent to the transmission frequency characteristics of a telephone line. The registered pattern obtained from the feature extractor is transmitted to the feature extractor via the filter circuit, and the registered pattern obtained from the feature extractor is stored in the registered pattern memory, and in voice recognition mode, the feature is extracted from the voice signal obtained from the telephone line. Introduce it into a container,
A speech recognition device characterized in that the speech at this time is recognized by the recognition processing section.