JPS6142280B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to preventing malfunction of a speech recognition device due to external noise.

Extract the feature parameters of the audio signal, use these feature parameters to recognize the voice of a certain speaker,
2. Description of the Related Art Various voice recognition systems have been proposed in the past that output signals and the like to control external devices in accordance with recognized voice. An example of this is shown in the block diagram of FIG. An input unit 1 including an acoustic-electrical signal converter (for example, a microphone) that converts input audio into an electrical signal, an input band filter 2 that extracts only a frequency band to be processed from the input audio signal, and an audio signal. a feature extraction unit 3 that extracts the features of
A standard pattern storage unit 4 that stores standard patterns of pre-registered audio features, a recognition processing unit 5 that compares the feature pattern extracted from the input audio with the standard pattern and identifies the input audio, and an output that outputs the recognition results. The main component is a control unit 6, which includes an input signal amplitude normalization circuit 7 for improving the recognition rate, a time axis adjustment unit 8, and is responsible for processing when registering standard patterns of voice features in advance. A registration control section 9 is added.

There are many methods that can be considered as parameters for extracting voice features, such as frequency spectrum distribution, correlation function, number of zero crossings, formant frequency, or linear prediction coefficient. The so-called filter bank method, which separates and extracts patterns and checks the correlation with a standard pattern, is often used as a method that can obtain a high recognition rate with a relatively simple configuration.

Now, in such a speech processing device, the ability to separate and extract control commands from an input signal is extremely important, and it is no exaggeration to say that the recognition rate of the device is influenced by this ability. Particularly when the environment in which the recognition device is used is a place where there is a lot of external noise, the noise will be input through the microphone 1, which will pose a big problem in use. In this case, the frequency spectrum distribution of the noise input from microphone 1 after passing through two input band filters has a pattern that is very similar to the frequency spectrum distribution of one of the voices registered in advance in the device. If this is the case, the device may mistakenly recognize noise as sound having a certain meaning.

External noises that often occur and cause problems include people's applause and impact noises caused by objects colliding with each other.In modern times, these relatively high-level noises that contain many high-frequency components are produced by various types of noise. It is becoming more likely to occur in the environment. This type of sound has a short duration, making it difficult for the device to accept it due to the length of a typical word, but it may be accepted if it occurs several times in succession with short intervals.

A so-called close-talk type microphone is used in the input section for the purpose of removing external noise, but this tends to be less effective against high-level sounds that contain many high-frequency components as mentioned above. Prevention of malfunctions is still insufficient.

The present invention has been made for the purpose of making it difficult for such a speech recognition device to malfunction due to noise.

That is, it is intended to clarify the distinction between noise and voice by focusing on the fact that the above-mentioned types of noise have a wide frequency band and include many high-frequency components that are not included in normal voice.

An embodiment of the present invention will be described in detail below with reference to FIG. Input section 1 includes microphone 10 and amplifier 1
1. The signal from the input section 1 passes through the input band filter 2 and then enters the feature extraction section 3.The feature extraction section 3 has center frequencies of ₁ , _{2, and 2} , respectively.
..., n bandpass filters (BPF) 1
2, 13 to 14, a multiplexer 15 that switches the output of each filter, and an analog-digital (A/D) converter 16 that converts the output level of each filter that has passed through the multiplexer into a digital signal. . Note that the band covered by all of the bandpass filters 12 to 14 is the same as or narrower than the band of the input band filter 2. As a result, each filter component of the audio signal input from the input section 1 is sampled at appropriate time intervals (in most cases, around 10 milliseconds) and converted into digital code, and then sent via the microcomputer 17 including the I/O port. It is stored in the storage memory 18. Another standard pattern memory 4 is connected to the microcomputer 17, and control command voices are stored in advance together with codes specifying the control contents. In the voice recognition mode, the control voice is input as described above, and the signal string extracted by each filter 12, 13...14 of the feature extraction circuit 3 and converted into a digital signal is stored in the storage memory 18, and then the microcomputer 17 The input voice is specified by calculating the difference between this stored pattern and the standard pattern for all standard patterns and determining the standard pattern with the smallest difference. In general, human speech is not always the same over time even when the same language is uttered, so it is well known that some kind of time axis adjustment circuit 8 as shown in Figure 1 must be added. In Although,
In FIG. 2 showing the configuration of the present invention, the time axis adjustment circuit is omitted for convenience of explanation.

Speech is always captured in the recognition mode, and when the input speech is interrupted, that is, during a pause period, the above-mentioned recognition calculation is executed and the previous input speech is identified by the pattern matching method.
At this time, when it became possible to identify the input audio,
That is, when the input voice matches some standard pattern within an allowable error range, the microcomputer 17 controls the output control section 6 to output an external device control signal corresponding to the control voice. If the input voice cannot be specified, the control signal is not output, but the display device 19 is driven, for example, to prompt re-voice.

Next, a description will be given of how noise and audio signals are distinguished in the embodiment of the present invention.
The signal from the input section 1 enters the input band filter 2 and at the same time passes the signal above the band of the input band filter. The signal is input to a high pulse filter (HPF) 20. Since no higher-frequency sound than necessary is detected here, the high-frequency range is attenuated according to the characteristics of the microphone. The output of the HPF 20 is integrated by a high frequency signal integrator 21. The output of the input band filter 2 is input to the BPFs 12, 13 to 14 and at the same time is input to the voice band signal integrator 22. The outputs of the high frequency signal integrator 21 and the voice band signal integrator 22 are input to a divider 23 constituting a signal comparator. This divider 23 divides the audio band signal level as the denominator and the high frequency signal level as the numerator, and the output of the divider 23 is input to the voltage comparator 24. Since the above-mentioned noise contains many components above the voice band, when this type of noise is input, the high frequency signal level increases, and the ratio with the voice band signal level, that is, the output of the divider 23, is equal to or higher than the voice band. It is clearly larger than when only the signal is used. Therefore, if the reference voltage (V ₀ ) of the voltage comparator 24 is selected to an appropriate value (this is determined experimentally) and the divider output is higher than the reference voltage (V ₀ ), the input signal is noisy. By configuring the microcomputer 17 to output a signal to the microcomputer 17, the microcomputer 1
7 can interrupt the recognition operation and prevent erroneous recognition. In this case as well, the display device 19 is driven to prompt the user to speak again.

As described above, according to the present invention, it is possible to distinguish between noise and normal speech based on the characteristics of their spectral distribution, so that high discrimination is achieved.
It is highly practical as it can prevent misrecognition by the recognition device and improve the recognition rate.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of an existing speech recognition device, and FIG. 2 is a block diagram showing the configuration of the device of the present invention, in which 1 is an input section, 3 is a feature extraction section,
Reference numeral 4 indicates a standard pattern storage unit, 17 a microcomputer, 21 an integrator for high frequency signals, 22 an integrator for voice band signals, and 23 a divider.

Claims (1)

[Claims] 1. A voice range filter that divides the acoustic signal from the acoustic-electrical signal converter into a voice band signal and a high frequency signal above the frequency band; a high frequency filter; and an audio filter that detects the level of the signal passing through the voice frequency filter. a high-frequency range level detector, a high-frequency range level detector that detects the level of the signal passed through the high-frequency range filter, and a comparator that takes the ratio of the detection outputs from both of these level detectors. A speech recognition device characterized in that, as a result of the comparison, the speech recognition operation is interrupted when the treble range level is higher than the voice range level.