JPH0352639B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a speech recognition device, and in particular, an object of the present invention is to provide a speech recognition device that can reduce erroneous recognition caused by level fluctuations of an input electrical audio signal. It is something to do.

In general, in a speech recognition device, it is necessary to somehow reduce the fluctuations in the magnitude of the audio electrical signal that occur due to the speaking voice becoming louder or softer or due to changes in the distance between the speaker and the acoustic transducer. I can't get good results. For this reason, conventional speech recognition devices have been provided with an automatic gain control amplification device within the device body. FIG. 1 shows a conventional speech recognition device. In Fig. 1, 1 is an acoustic transducer that converts speech into an electrical signal, 2 is a speech recognition device main body, and inside this device main body 2 is an automatic gain control amplification device for making the input signal constant. I am wearing 3. FIG. 2 shows an example of the waveform of an audio signal. This is an example of a male voice occurring as a "channel". As can be seen in Figure 2, the speech signal has a very complex shape, and it is extremely difficult to maintain a constant volume without damaging this characteristic.The initial part of the word is particularly deformed, and as a result, it is difficult to recognize it. Errors often occurred in the results.

The present invention solves these conventional drawbacks, and includes a means for detecting the energy of an electrical audio signal that has reached an acoustic transducer, decomposing the detected output into a frequency spectrum, and digitizing the electrical signal. By setting the reference voltage to , the configuration is such that even if the magnitude of the audio signal fluctuates, misrecognition is reduced.

The present invention will be described below with reference to drawings of embodiments. FIG. 3 shows an embodiment of the present invention,
In FIG. 3, 17 is an acoustic transducer that converts audio into an electrical signal, and its output is amplified to an appropriate size by an amplifier 18 and provided to a filter device 19 for frequency analysis. . The filter device 19 is composed of a bandpass filter, a detector for detecting the energy of the bandpass filter, and a low-pass filter, and outputs an envelope of the frequency component. And these are frequency band 200Hz
15 filters 19-1 with a passband width of 1/3 octave over 5 octaves of ~6400Hz,
It is composed of 19-2...19-5. 24 is a total energy detection device for detecting the total energy of the output from the amplifier 18, and is composed of a detection and a resistance filter. 25 is the output of the total energy detection device 24 and the filter device 19;
The outputs of the total energy detection device 24 and the filter device 19 are sequentially switched under the control of a microcomputer 28. The output of the multiplexer 25 is connected to an analog-to-digital converter 27 via a sample hold 26. A switch circuit 22 is connected to a reference voltage terminal of this analog-to-digital converter 27.
A reference voltage from a reference voltage generator 23 fixed at
The output of a sample hold 21 that holds the output of the total energy detection device 24 is switched and connected. The output of the analog-to-digital converter 27 is taken into the microcomputer 28, converted into a pattern according to the instructions of an internal program, and registered in the memory device 29 in the registration mode.
In the recognition mode, recognition processing is performed by comparing with data stored in the memory device 29.
It also has a control device 20 for operating the recognition device and outputting its results.

As shown in FIG. 4, the analog-to-digital converter 27 includes a comparator 31 to which the signal from the multiplexer 25 is input, a successive approximation register 32 to which the signal from the comparator 31 is input, and a successive approximation register 32 to which the signal from the multiplexer 25 is input. An A/D converter 33 to which the signal of
A bus control circuit 34 that inputs data from the converter 33 into the microcomputer 28 under the control of the microcomputer 28, a comparison data processing circuit 35 to which data from this bus control circuit 34 is input, and this comparison data. a D/A converter 36 that converts the comparison data from the processing circuit 35 into analog data, compares it with the signal from the successive approximation register 32, and adds the comparison signal to the other input of the comparator 31; and a D/A converter 36 that processes the comparison data. It has a control circuit 37 that processes the circuit 35 and the D/A converter 36.

In the speech recognition device having such a configuration, the speech signal captured by the acoustic transducer 17 is transmitted to the amplifier 18.
After being amplified by , it is supplied to a filter device 19 and a total energy detection device 24 . Then, each output is input to the multiplexer 25. The multiplexer 25 has channel 1 as its initial value.
Under the control of the microcomputer 28, the sample hold 26 and analog/digital converter 27 are activated to measure the total energy. At this time, the switch circuit 22 is connected to a fixed reference voltage generator 23. If the measurement result does not reach the specified level, the switch circuit 22 switches the multiplexer 25 to the next channel 2 and performs the measurement while remaining connected to the fixed reference voltage generator 23. This operation is repeated up to channel 16 to complete one measurement.

If the total energy measurement result exceeds a specified level, the sample hold 21 is triggered under the control of the microcomputer 28 to hold the total energy level, and the switch circuit 22
is switched, and a reference voltage corresponding to the total energy detection device 24 is supplied to the analog-to-digital converter 27. After that, each output of the filter device 19 is sequentially measured in the same manner as in the above case. The above operations are repeatedly performed over the entire audio section. FIG. 5 shows an example of data taken into the microcomputer using only a fixed reference voltage. FIG. 6 shows an example of data captured using this method. As is clear from comparing the two, the data taken in with the total energy data as the reference voltage is the value obtained by dividing the output level of each filter by the energy. This shows that this method guarantees the same data even if the input signal level changes. This method of generating the same pattern even when the input audio volume fluctuates is as follows:
It also shows a good match when compared with the pattern stored in the memory device 29, significantly reducing erroneous determinations.

As described in detail above, according to the present invention, the energy of the audio electrical signal is detected and the output of the detection means is used as the reference voltage of the means for decomposing the audio electrical signal into a frequency spectrum and digitizing it. Therefore, it has the advantage of being able to generate the same pattern even when the input audio volume fluctuates, improving consistency in matching with the pattern stored in the memory device, and reducing misrecognition. be.

[Brief explanation of drawings]

Fig. 1 is a block diagram of a conventional speech recognition device, Fig. 2 is a waveform diagram of a speech signal, Fig. 3 is a block diagram showing an embodiment of the speech recognition device of the present invention, and Fig. 4 is an analog diagram of the same device. The specific circuit diagram of the digital converter, FIGS. 5 and 6, are diagrams showing the data taken in the main parts of the device. 17...Acoustic transducer, 18...Amplifier, 19...
...Filter device, 20...Control device, 21, 26
...Sample hold, 22...Switch circuit,
23... fixed reference voltage generator, 24... total energy detection device, 25... multiplexer, 27
...Analog-digital converter, 28...Microcomputer, 29...Memory device.

Claims (1)

[Claims] 1. Amplifying means for amplifying the audio electrical signal obtained through the acoustic transducer; Frequency analysis means for decomposing the signal from the amplification means into frequency spectra; and Frequency analysis means for decomposing the signal from the frequency analysis means into frequency spectra. A digitizing means for digitizing; a voice recognition means for taking in data from the digitizing means and storing it in a memory, or recognizing speech by comparing it with the data stored in the memory; and standards for the digitizing means. a fixed reference voltage generator for outputting a signal as a voltage; a means for detecting the energy of the audio electrical signal; 1. A speech recognition device comprising: means for detecting the energy of the audio electrical signal; and switch means for automatically switching between the means for detecting the energy of the audio electrical signal and supplying the reference voltage of the digitizing means.