JPH0517700U - Voice recognizer - Google Patents

Abstract

(57) [Summary] [Purpose] When performing pattern matching by comparing the registered voice and the input voice, it is intended to suppress the occurrence of mismatching due to the influence of noise. [Structure] A low-pass filter 2 for detecting a noise level of an input voice is connected, and a gain of a voice amplifier circuit 4 is controlled according to the detected noise level. If pattern matching between the output voice of the voice amplifier circuit 4 and the registered voice is performed,
Since the influence of noise is prevented and the output voice is always controlled and amplified with a constant gain, the influence of noise can be avoided and the mismatch can be reduced.

Description

[Detailed description of the device]

[0001]

[Industrial applications]

 The present invention relates to a voice recognition device that is less affected by ambient noise.

[0002]

[Prior Art]

 A voice recognition method is known in which a plurality of words to be recognized are registered in advance and the voice input at the time of recognition is compared with the registered voice to perform pattern matching to recognize a specific speaker. Recently, such a voice recognition method has been adopted to input data by voice and control the controlled device by voice.

When performing such voice registration, the input voice signal is amplified by an AGC (automatic gain control) circuit so as to control the gain, and then registered. Also during recognition, the input voice signal is compared with the registered voice after the gain is controlled by the AGC circuit. In the case of performing such voice recognition, it is desirable that the circumstances of the surrounding sounds at the time of voice registration and at the time of voice recognition are the same, and that the gain control for each voice is equally performed.

[0004]

[Problems to be solved by the device]

 By the way, in the conventional voice recognition device, the control of the gain of the AGC circuit for the voice is not performed at the same time during the voice registration and the voice recognition. There is a problem that it does not work well and mismatch occurs.

That is, voice registration is performed in a quiet, noise-free place, but since voice recognition is often performed in a noisy place in the surroundings, the gain control is affected by noise and the gain control is different from that during voice registration. Will be different. For example, when controlling an in-vehicle audio device by voice recognition, voice recognition is performed while the vehicle is running, so the influence of road noise cannot be avoided.

Since gain control for a voice signal is performed for a spoken word, a gain difference occurs between a low power level portion (consonant) and a high power level portion (vowel). However, since the road noise is superimposed during recognition, the input voice is masked by the load noise, and the level difference between the low power level part and the high power level part is reduced, so there is no gain difference. Come on. For this reason, a difference occurs between the registered voice pattern and the input voice pattern, and the pattern matching does not work well, resulting in a mismatch.

The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a voice recognition device that eliminates the influence of ambient noise and reduces mismatching during voice recognition. To do.

[0008]

[Means for Solving the Problems]

 In order to achieve the above object, the present invention provides a voice recognition device that recognizes a specific speaker by registering a plurality of voices to be recognized in advance and performing pattern matching between the input voice and the registered voice. Noise detection section that detects the ambient noise level when inputting, an audio amplification section that controls the gain of the input audio according to the detected noise level, and noise that removes the noise component of the output of the audio amplification section. It is characterized by having a component removal unit and a voice recognition unit that recognizes a specific speaker by comparing the output of the noise component removal unit and the registered voice.

[0009]

[Action]

 The noise level around the input voice during voice recognition is detected, and the input voice is amplified so that the gain is controlled according to the detected noise level. After removing the noise component from this amplified voice output, the registered voices are compared and pattern matching is performed. Since the gain of the input voice is controlled according to the noise level, amplification is always performed according to the surrounding noise. This can eliminate the influence of surrounding noise and reduce mismatching during voice recognition.

[0010]

【Example】

 An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of a voice recognition device of the present invention. 1 is a microphone for inputting voice, 2 is a low-pass filter for detecting road noise, 3 is a rectifying / smoothing circuit for changing the output of the low-pass filter 3 to DC, and 4 is control of the gain of input voice. A VCA (voltage control amplifier) type voice amplifier circuit 5 that amplifies according to the noise level is a high-pass filter that removes the main noise component of the voice amplifier circuit 4.

Reference numeral 6 is an A / D converter that converts an analog signal output from the high-pass filter 5 into a digital signal, 7 is a voice registration unit that registers and stores a plurality of voices in advance, and 8 is a microcomputer configuration. A registration / recognition processing unit 9 is an operation unit for selecting voice registration and voice recognition.

The low-pass filter 2 is for particularly detecting a road noise component from the input voice. As a result of measuring the road noise in the passenger compartment, the distribution shown in Fig. 2 was obtained. Road noise has a voice distribution of fc = 200Hz and 12dB / oct, and energy is concentrated below 200Hz. Therefore, the road noise can be detected by using the low pass filter 2 with fc = 200 Hz.

Next, the operation of this embodiment will be described.

First, at the time of voice registration, a quiet, noise-free place is selected. Also, voice registration is selected by the operation unit 9. In this state, the voice input from the microphone 1 is amplified by the voice amplification circuit 4 so that a predetermined gain control is performed regardless of the operations of the low-pass filter 2 and the rectifying and smoothing circuit 3 because there is no noise. After passing through the high-pass filter 5, it is converted into a digital signal by the A / D converter 6.

The digital signal is input to the registration / recognition processing unit 8 and recorded in the memory in the voice registration unit 7 under this control. After that, the same signal processing is performed every time a voice is input from the microphone 1, and a plurality of voices are registered in the memory in the voice registration unit 7.

Next, voice recognition is often performed in a place where there is a lot of noise in the surroundings, such as in a moving vehicle, and is therefore affected by noise. In this case, the voice recognition is selected by the operation unit 9. In this state, the voice input from the microphone 1 is output to the voice amplifier circuit 4 and the low pass filter 2.

Since the low-pass filter 2 is set to fc = 200 Hz, the main component of road noise of approximately 200 Hz or less passes through the low-pass filter 2 and is then converted into direct current by the rectifying / smoothing circuit 3. This direct current is output to the voice amplifier circuit 4.

As a result, the voice amplifier circuit 4 operates so as to control the gain of the input voice according to the noise level. That is, the gain is controlled to be low when the noise level is high, and the gain is controlled to be high when the noise level is low. As a result, the gain of the phoneme in the word of the input speech is controlled to be constant. When the noise level is low as in the voice registration as described above, the gain of the voice amplifier circuit 4 is controlled to be high.

The output of the audio amplifier circuit 4 whose gain is controlled according to the noise level is added to the high-pass filter 5 to remove the main component of noise, and then converted into a digital signal by the A / D converter 6. It This digital signal is a voice with the influence of road noise removed.

The digital signal is input to the registration / recognition processing unit 8, and based on this, the registration / recognition processing unit 8 records a voice pattern, which is a feature of the input voice, in a memory in the voice registration unit 7. Pattern matching is performed by comparing the voice patterns of multiple registered voices. When the registered voice for which pattern matching is established is found, it means that the specific speaker is recognized, and the registration / recognition processing unit 8 outputs a control signal to the device to be processed.

As described above, according to the present embodiment, the noise level of the voice input during voice recognition is detected by the low-pass filter 2 and the gain of the voice amplifier circuit 3 is controlled according to this noise level. The circuit 3 always outputs a constant level of sound. Therefore, when performing pattern matching with the registered voice, the input voice without the influence of road noise is compared, so that voice recognition with less mismatching can be performed. Further, it is possible to prevent saturation in which the road noise is excessively amplified by the audio amplifying circuit 4.

[0023]

[Effect of the device]

 As described above, according to the present invention, the voice amplification is performed so as to control the gain according to the noise level, and the pattern output is performed by comparing the voice output with the registered voice. It is possible to perform less voice recognition.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a voice recognition device of the present invention.

FIG. 2 is an energy distribution diagram of road noise for explaining the operating principle of the present invention.

[Explanation of symbols]

 2 Low-pass filter 4 Voice amplification circuit 5 High-pass filter 7 Voice registration unit 8 Registration / recognition processing unit

Claims (1)

[Scope of utility model registration request] 1. A voice recognition device for recognizing a specific speaker by registering a plurality of voices to be recognized in advance and performing pattern matching between the input voices and the registered voices. Noise detecting section for detecting the noise level of the sound, a voice amplifying section for controlling the gain of the input voice according to the detected noise level, a noise component removing section for removing the noise component of the output of the voice amplifying section, and a noise component A voice recognition device comprising: a voice recognition unit for recognizing a specific speaker by comparing the output of the removal unit and a registered voice.