JPH06337697A - Voice recognition device - Google Patents

Abstract

PURPOSE:To improve the precision in recognition by using a voice recognition result and adjusting an input part. CONSTITUTION:After a sound inputted from a microphone 1 is amplified by a preamplifier 2A and a variable gain amplifier 2B, is converted to a digital signal by an A/D converter 3 to be supplied to a digital signal processor(DSP) 4. After the digital signal is voice recognition processed in the DSP 4, is outputted from an interface circuit 5 through an optimum format or port. At this time, when no recognition succeeds, a signal adjusting an amplifier gain is outputted from the DSP 4, and a gain switching circuit 2C of the variable gain amplifier 2B is switched by the signal. That is, when no recognition succeeds, the circuit 2C is switched so as to increase the gain of the variable gain amplifier 2B. Further, since a success/a failure in recognition is a probable problem, a method feeding back the result by a human and the method, etc., probably learning in the DSP side 4 may be used also.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device which can be used in general information equipment and consumer equipment which use voice as input means.

[0002]

2. Description of the Related Art Human utterance sounds are different for each speaker and also vary for each speaker even if they are the same speaker. In the early speech recognition technology, how to match these fluctuations in the time domain and pattern matching was examined. further,
By introducing mathematical statistical methods and neural nets, modeling in both frequency and spectrum regions has been performed, and the recognition rate in word units of unspecified speakers has improved dramatically.

With the development of such technology, a large number of personal computer built-in boards capable of easily recognizing voice have been commercialized.

On the other hand, in the field of consumer products, an attempt has been made to input an operation by voice input as one approach for simplifying the operation of a complicated AV device. Pages 38-44
Page) has been partially commercialized as a remote control device using voice recognition technology is introduced.

FIG. 2 shows a circuit configuration example used for general voice recognition. Generally, since digital signal processing is performed, the input microphone 1, the amplifier system 2, the AD (analog / digital) conversion processing unit 3, the computer, the DSP (digital signal processor), or the recognition composed of a dedicated voice recognition IC It is divided into a processing unit 4 and an external interface unit 5.

[0006]

The technology currently commercialized is in the category of the above-mentioned word unit (unit) or its sequential processing. In this case, it is necessary to recognize only a predetermined word. The recognition rate is determined by how well the voice and the prepared database are matched.

[0007] This is an important factor as well as the problem of the recognition algorithm and how well the peripheral circuit can capture the necessary voice.

For simplicity, assume that the input voice signal has a single frequency as shown in FIG. When the input is at an excessive level, it becomes overrange in the amplifier system before being taken into the recognition calculation unit, and the peak is trapped in the waveform as shown in FIG. At this excessive input level, the waveform itself has changed. That is,
The frequency pattern, which is an important parameter for recognition, changes and it becomes impossible to recognize.

As a countermeasure against the excessive input as described above, it is conceivable to drop the amplifier gain in advance in order to prevent the range from being exceeded, but on the contrary, the recognition rate of the under level is greatly lowered. There is also a method of varying the gain according to the input level, but unlike a video signal that has a reference such as a synchronization signal whose input level has been determined in advance to some extent, the audio signal originally has no time axis / level reference and is quite dynamic. This method is dangerous because it fluctuates.

In view of this, the present invention intends to improve the recognition accuracy by adjusting the input section using the voice recognition result.

[0011]

[Means for Solving the Problems] When the same person occurs toward a microphone or the like, consider the time when a predetermined word is generated within a certain fixed time, and bring the microphone to a relatively similar position. It often occurs with similar sound pressure. Therefore, if the voice input is also distorted due to level over or the like, it is considered to occur continuously.

Therefore, when the voice recognition is unsuccessful, the gain is adjusted. At this time, a method having a learning effect, for example, data of the past several times is averaged and stored.

[0013]

By doing so, the recognition rate is improved due to the above-mentioned utterance characteristics.

[0014]

FIG. 1 shows an embodiment of the present invention.

The voice input from the microphone 1 is amplified by the preamplifier 2A and the variable gain amplifier 2B, converted into a digital signal by the AD converter 3, and supplied to the DSP 4. DSP
Then, after the voice recognition processing is performed, it is output from the interface circuit 5 to an external device through an arbitrary format or port.

When the recognition is unsuccessful, the DSP 4 outputs a signal for adjusting the amplifier gain, and this signal switches the gain switching circuit 2C of the variable gain amplifier 2B. That is, when the voice recognition is unsuccessful, the variable gain amplifier 2
Switch to increase B gain.

Since recognition success / failure is a probabilistic problem, a method in which a human feeds back the result, a method in which the DSP side learns stochastically, or the like may be used for gain adjustment.

In the voice recognition, the frequency component is an important parameter as described above, and if there is a characteristic distortion such as a range over, it may be possible to make a prediction of the frequency analysis. If the ROM has a margin, the accuracy can be further improved by adding this information to the discrimination element at the time of adjustment.

[0019]

According to the present invention, since the waveform distortion in the microphone amplifier section, which is an important factor in the success rate of voice recognition, is improved by feeding back the recognition result, the recognition rate can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram of a voice recognition device embodying the present invention.

FIG. 2 is a block diagram of a conventional voice recognition system.

FIG. 3 is a waveform diagram for explaining the operation.

FIG. 4 is a waveform diagram for explaining the operation.

[Explanation of symbols]

 1 Microphone 2A Preamplifier 2B Variable Gain Amplifier 2C Gain Switching Circuit 3 AD Converter 4 Digital Signal Processor 5 Interface Circuit

Claims (1)

[Claims] 1. A microphone, a variable gain amplifier for amplifying a voice signal obtained from the microphone, an analog-digital converter having an output obtained from the amplifier as an input, and a voice processing an output from the converter. A speech recognition apparatus, comprising: a recognition processing circuit, wherein the gain of the variable gain amplifier is adjusted based on the speech recognition rate in the speech recognition processing circuit.