KR20000032269A - Voice recognizing apparatus of sound equipment - Google Patents

Abstract

PURPOSE: A voice recognizing apparatus of a sound equipment is to control the sound equipment using a voice command in a noise environment, thereby enhancing the convenience and the value of the apparatus. CONSTITUTION: A voice recognizing apparatus comprises an adaptive filter(26) for removing a noise generated from a sound equipment. As a voice inputs, a voice section is detected by a voice sector detecting algorithm. From the detected voice section, the voice is analyzed to extract a characteristic of the voice. The extracted characteristic is compared with a reference pattern or an occurrence probability of the reference patter is computed. With the comparing result and the computed occurrence probability, the input voice is recognized or is refused. The extracted characteristic and recognition result is adapted for the recognizing apparatus. The apparatus controls the operation of various sound equipments with the extracted characteristic and recognition result.

Description

Voice recognition device of sound equipment

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to speech recognition, and more particularly, to a speech recognition apparatus of an acoustic device capable of controlling the acoustic device even in ambient noise.

At present, there are many attempts to operate electronic products that are used in daily life by voice commands.

In particular, TV, audio, and so on, play an important part in modern life such as leisure, information gathering and education.

Therefore, the remote control has appeared to improve the convenience of operation, thereby allowing the user to perform the desired action by sitting in the seat.

However, since this also operates other devices, there is a limit to the convenience, so that a device capable of recognizing the voice most familiar to humans has been devised.

The general operation principle of the speech recognition is as follows.

First, the speech recognition algorithm may be classified into a speech section detection process, a feature extraction process, and a matching process.

That is, as shown in FIG. 1, when a voice signal is input through the microphone 11, the A / D converter 12 converts the voice signal into a digital signal and then outputs the digital signal to the voice section detector 13.

The speech section detector 13 divides the digital speech signal into short section signals (ie, frames), and then uses the energy, zero crossing rate, and time length information of each frame to actually input the signals. After detecting only the spoken speech section, the speech signal is output to the feature extractor 14.

The feature extractor 14 extracts a feature of a frame corresponding to the speech section, creates a test pattern of the input speech, and outputs the test pattern to the matcher 15.

The matching unit 15 compares the test pattern with each of the reference patterns stored in the reference data memory 16 and outputs a reference pattern having a feature most similar to the test pattern as a recognized voice.

In this case, when the reference pattern of the voice signal is stored in the reference data memory 16, the feature extractor 14 extracts a feature of a frame corresponding to the voice section to create a reference pattern, and then uses the reference data. In the memory, the voice signal to recognize the operation is repeatedly performed to database the reference pattern in the reference data memory 16.

However, such a conventional speech recognition device is degraded when there is ambient noise.

In particular, when a microphone is mounted on a device generating a sound signal, such as a TV or an audio device, the voice may not be recognized because the sound generated by the sound device is very loud.

Therefore, there is a need to develop a device capable of speech recognition despite the presence of such ambient sounds.

An object of the present invention is to provide a voice recognition device for an audio device capable of recognizing voice despite the ambient sound generated by the sound device.

Another object of the present invention is to apply a voice recognition algorithm to a sound device with a built-in microphone to operate the sound device by voice.

1 is a block diagram of a general speech recognition system

2 is a block diagram of a voice recognition device of an acoustic device according to the present invention;

3 is a block diagram illustrating an adaptive filter of FIG. 2.

Explanation of symbols for main parts of the drawings

21: adaptive filter unit 22: speech recognition unit

25: first mixer 26: adaptive filter

27: second mixer 28: voice section detection unit

29: feature extraction unit 30: matching unit

31: recognition result checking unit 32: recognition result performing unit

Features of the speech recognition device of the acoustic device according to the present invention is an adaptive filter unit for adapting a noise signal generated in the audio device to the audio signal input from at least one microphone and outputting the speech signal from which the noise signal is removed, and an adaptive filter A speech section is detected by applying a negative speech signal to detect a speech section, extracts a feature by analyzing the detected speech signal, and compares the extracted feature with a reference pattern or calculates a probability of occurrence of the reference model. The speech recognition unit is configured to recognize or reject a voice, and to adapt the extracted feature value and the recognition result to a recognizer to perform the recognized result to a peripheral device.

According to another aspect of the present invention, there is provided a first mixer which adds a noise signal generated from an acoustic device and a voice signal input from a microphone, and outputs a voice signal mixed with noise, and a voice mixed with noise applied from the first mixer. A second mixer for subtracting the signal and the filtered signal applied from the following adaptive filter, and outputting a noise-free voice signal; and a filter for outputting the applied noise signal and filtering out the noise-free voice signal output from the second mixer. It consists of an adaptive filter that adjusts the coefficients of the filter by feedback.

Another feature of the present invention is to use the characteristics of the noise signal generated from the coefficient change of the adaptive filter and the noise signal from the audio device and the voice signal input from the microphone when detecting the speech section in the speech recognition unit.

Referring to the accompanying drawings, a voice recognition device for an acoustic device according to the present invention having the above characteristics is as follows.

First, the concept of the present invention forms an adaptive filter using a sound signal (noise) generated by the sound device as a reference signal to remove the noise generated by the sound device, thereby enabling speech recognition to the sound device. To make it happen.

FIG. 2 is a block diagram illustrating a speech recognition apparatus of an acoustic device according to the present invention. As shown in FIG. 2, the adaptive filter unit 21 and the speech recognition unit 22 may be largely divided.

As shown in FIG. 3, the adaptive filter unit 21 includes a first mixer 25 for adding a noise signal generated from an audio device and a voice signal input from a microphone, and outputting a voice signal mixed with noise, and a first mixer ( A second mixer 27 which subtracts the voice signal mixed with the noise applied from 25 and the filtered signal applied from the adaptive filter 26 to output the voice signal from which the noise is removed, and filters and outputs the applied noise. And an adaptive filter 26 for feeding back the speech signal from which the noise output from the second mixer 27 is removed to adjust the coefficient of the filter.

Here, the noise refers to an acoustic signal generated from the sound device to be controlled, and the audio signal input from the microphone refers to an audio signal including an external noise signal and a noise signal generated from the audio device.

In addition, one or more microphones may be used, and may be attached to an acoustic device or mounted elsewhere outside the acoustic device to receive an acoustic or audio signal.

The voice recognition unit 22 extracts a voice signal section from a signal applied through the microphone by using the coefficient change of the adaptive filter 26 and the voice signal and noise input through the microphone ( 28), a feature extraction unit 29 for extracting a parameter representing a characteristic from the detected speech signal, and a matching unit for recognizing the input speech by comparing the extracted parameter with a reference pattern or calculating the probability of occurrence of the reference model ( 30), a recognition result unit 31 which rejects recognition when the similarity or probability of occurrence of the recognized result is below a predetermined level, a recognition result performing unit 32 that performs the recognized result, and extracted feature values and recognition It is composed of a recognizer adaptor 33 adapted to adapt the speech recognition apparatus to the characteristics of the speaker or the use environment using the results.

The operation of the present invention configured as described above will be described with reference to FIGS. 2 and 3.

First, in order to recognize a voice in an acoustic device, an acoustic signal (noise) generated from the audio device must be removed from the voice signal applied from the microphone.

To do this, the adaptive filter coefficients should be adjusted to minimize the difference between the signal generated by the acoustic device and the voice signal applied from the microphone.

That is, the noise generated in the acoustic device and the voice signal input from the microphone are added by the first mixer 25 to output the voice signal in which the noise is mixed, and the adaptive filter 26 filters the noise signal.

The output signal of the adaptive filter 26 and the output signal of the first mixer 25 are applied to the second mixer 27 to output an audio signal from which noise is removed.

The speech signal from which the noise is removed is fed back to the adaptive filter 26 to adjust the coefficients of the adaptive filter.

In this way, the speech signal from which the noise output from the adaptive filter unit 21 is removed is applied to the speech section detecting section 28 to detect only the section actually spoken, i.e., the speech section.

Here, the voice section is detected using the coefficient change of the adaptive filter and the characteristics of the voice signal input from the microphone and the noise signal generated by the audio device.

The reason for using them is that in the case of the adaptive filter, the filter coefficient is very severely changed when the main signal input from the reference signal and the microphone are significantly different.

That is, while the voice is input, the filter coefficient change is remarkably different from other intervals.

Therefore, if the coefficient variation is continuously monitored, a large portion of the variation can be extracted as the voice interval.

Here, each coefficient change amount measures the average coefficient change amount of a short segment.

In addition to the filter coefficient variation, the speech section is extracted by using the spectrum information and the energy information of the reference signal and the main signal together.

Next, the feature extractor 29 extracts a feature of the detected speech section signal, and usually uses a parameter representing spectrum information.

The matching unit 30 recognizes the input voice by comparing the extracted parameter with a reference pattern or by calculating a probability of occurrence of the reference model.

That is, the voice most similar to the input voice is searched among the registered voices. After all the similarities of the registered words are all calculated, the most similar registered words are extracted.

As an example of measuring similarity, a dynamic time warping (DTW) algorithm or a hidden markov model (HMM) probability value may be used.

The matching unit 30 of the present invention may use a two-stage recognition method to improve the performance of the voice recognition device.

Two-stage recognition is divided into a key word recognition process and a command recognition process. The key word recognition process recognizes whether a single word is uttered.

That is, there is only one vocabulary to be recognized, and the "recognition" or "deny" is determined by the recognition result checking process, and the recognition rate is improved by minimizing the recognition words.

The device to be controlled operates in the keyword recognition mode as an initial state, and when it is determined as "recognition", sets the device to be controlled to the command recognition mode and limits the noise generated by the device to a predetermined level or less.

In addition, the command recognition process recognizes a command of the device, performs a recognition result, and sets the device to a keyword recognition mode in its original state.

Then, if the similarity between the most similar registered words and the input voice is rejected by the predetermined rule, the recognition result confirming unit 31 sees that the voice not registered is input and accepts the voice again.

In this case, a simple similarity may be measured and used, or various types of filter models may be used.

This applies to both keyword recognition and command recognition.

If the recognition process is successful, the recognition result execution unit 32 executes a command according to the result.

On the other hand, the recognizer adaptor 33 adapts the speech recognition apparatus to the characteristics of the speaker or the use environment using the extracted feature values and recognition results.

This can be done as a training course before the actual use of the device or during the actual use.

The speech recognition apparatus of the acoustic device according to the present invention has the following effects.

First, by controlling sound devices such as TV, audio, and radio by voice recognition, convenience and added value thereof can be increased.

Second, by solving the noise problem, which is the biggest problem in applying the voice recognition function to the audio device, it is possible to enable the voice recognition in spite of the noise generated in the sound device.

Claims (2)

An adaptive filter unit for adapting a noise signal generated by an audio device to at least one microphone and inputting a voice signal, and outputting a voice signal from which the noise signal is removed; The speech signal is applied to the adaptive filter unit to perform a speech section detection algorithm to detect a speech section, analyze the detected speech signal to extract a feature, and compare the extracted feature with a reference pattern or calculate a probability of occurrence of a reference model. And a voice recognition unit configured to recognize or reject the input voice and to adapt the extracted feature value and the recognition result to a recognizer to perform the recognized result to a peripheral device. The apparatus of claim 1, wherein the voice signal input from the at least one microphone is a voice signal including an external noise signal and a noise signal generated by the audio device.