JPS62178299A - Audio input system for voice recognition circuit - Google Patents

Abstract

(57) [Abstract] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] (Industrial Application Field) The present invention is directed to an audio input method in a speech recognition circuit.
It is 30T.

(Conventional technology) Currently, voice recognition circuits are being researched and developed in various fields such as robots, automobiles, and home appliances, but many conditions are required to make them usable, and the applications are extremely narrow. There's a problem. The reason for this is that the microphone used may receive a lot of ambient noise in addition to the desired sound, resulting in a different sound (different ambient noise conditions) than what was originally memorized. Things may fail or malfunction. For this reason, all current voice recognition systems use close-talk type microphones (for example, those in telephones) and cannot be used unless the noise is minimized. However, if a close-talking microphone is used as described above and it is used in a place with little noise, the use of voice recognition equipment will be limited to a very narrow range of specific areas, such as inside a factory, a subcommittee meeting, or a car. It must be said that it cannot be used in places where there are many equal m sounds, and its practicality is extremely low.

(Technical Problem) The present invention solves the above-mentioned problems, and its purpose is to provide a speech recognition circuit that does not require the use of a close-talking microphone, and in particular, a speech recognition circuit that does not require the use of a close-talking microphone. The purpose is to provide a recognition circuit. Furthermore, according to the present invention
:Speech recognition based on the low-cost zero-cross count method, which until now has been impractical due to low recognizers, will become fully practical.

(Technical Means) The present invention that achieves the above object uses microphones for target voice and noise, and amplifies the sound entering the target sound microphone and the sound entering the noise microphone, respectively,
An analog subtracter is used to remove each sound that has entered in the same way, and the output of the analog subtracter is made louder by an expander circuit, louder sounds are made smaller, and then signals below a certain level are cut. This removes unintended noise from small signals that could not be removed by the analog subtracter, while preventing malfunctions by activating the electronic volume circuit and temporarily interrupting the circuit in case of sudden large noises.
This is an audio input method in which the target signal is further compressed using a compressor circuit to equalize the voice and send it as an audio signal to the voice recognition circuit.

Note that in order to improve the quality of the circuit, it is better to use unidirectional microphones, and the number of microphones that can be used is of course not limited to one for the target voice and one for the noise. , more than one can be used. Furthermore, if a higher-grade IC is used in the speech recognition circuit, the recognition rate can be further increased.

(Operation of the Invention) Since the present invention has the above-described configuration, it is sufficiently resistant to noise, and can produce an output close to the ideal, similar to an analog signal as if it were obtained using a close-talking microphone in a place with little noise. Therefore, malfunctions are less likely to occur. Therefore, sufficient practicality can be expected even when applied to the zero cross count method.

(Example) Examples will be shown below, but the present invention is not limited to the contents thereof.

Microphone 1 is a unidirectional microphone for inputting the target voice and is set to face the speaker.

Microphone 2 is a unidirectional microphone for inputting ambient noise, and is preferably set in the opposite direction to microphone 1 with respect to the speaker. The audio signal from microphone 1 is
The signal is amplified by an amplifier 3 equipped with a bandpass filter, and is input to a terminal A of an analog subtracter 5. A noise signal from the microphone 2 is amplified by an amplifier 4 equipped with a bandpass filter, and its output is divided by a resistor, and one is inputted to a terminal B of an analog subtracter 5. The other divided signal C is rectified by the rectifier circuit 7r to become a direct current, passes through the time constant circuit 9, and becomes a control voltage for controlling the electronic volume 10.

The signals A and B input to the analog subtracter 5 are subtracted and do not appear in the output because the noise signals input to the microphones 1 and 2 have the same phase. For example, if the microphones 1 and 2 are only about 30 cm apart, the noise caused by a car driving, etc. will be input at approximately the same level. However, the speaker's sound is not generated from the entire surroundings, but is emitted from a single point close to microphone 1, so although a large amount of sound is input to microphone 1, microphone 2 receives sound that is reflected from an object. , Mawarikomi/
Interfered with the sound coming in at υ, only one-tenth to several-tenths of the sound entering microphone 1 is input. As a result, the noise input to microphones 1 and 2 with similar magnitude is
The -B to the speaker's voice signal which is eliminated by the analog attenuator 5 and entered the microphone 1 is attenuated by only about 10% at most, so it appears as a signal output of about 0.9A. This signal is further input to a low-pass filter 6,
High-pitched sounds unrelated to human speech that could not be removed by the circuit described above are removed, and the output of the low-pass filter is input to an expander circuit 8. This expander circuit has a logarithmic expansion rate of twice, so that the unnecessary signal that could not be removed by the previous circuit is small, the necessary audio signal is increased, and the difference between the necessary signal and the unnecessary signal is increased.

The output from the expander circuit 8 is the electronic volume circuit 1.
At 0, control so that the signal is not distorted.

When a sudden very loud noise occurs, the electronic volume circuit 10 is controlled so as not to output any output. In the event of such noise input, a time constant circuit is included in the control signal path to hold the circuit for an appropriate period of time, for example about 30 seconds, to prevent malfunction.

The control signal time constant circuit 9 is a circuit configured to selectively give a long delay time when the input to it is cut off, and various known circuits may be used as such a circuit. However, for example, if the series circuits of diode D1, resistor B1, diode D2, and resistor B2 are connected in parallel, and the diode D2 and resistor B2 are grounded via capacitor C, the structure can be simplified. Become. In addition, the initial operation can be made faster regardless of the hold time of the circuit. An attenuator 11 is installed at the output or input of the electronic volume 10 to cut off signals below a certain level. This signal enters a compressor circuit 12 with a logarithmic compression ratio of 1/2, and is output as a signal with little difference in level between the output signals. Weak signals below a certain level are not amplified by the compressor circuit 12 and therefore do not appear in the output.

The circuit described above removes noise other than the target speech signal, equalizes the speaker's signal tone, and produces only signals that are effective for the speech recognition circuit.

(Effects of the Invention) Since the present invention is configured and operates as described above, it is possible to obtain a voice signal with an extremely excellent SN ratio, thereby significantly increasing the voice recognition rate, reducing the risk of malfunction, and connecting Even without using a talking type microbond, a high-quality signal can be easily obtained without using any special elements, similar to when a close-talking type is used under ideal conditions.

[Brief explanation of drawings]

The drawing is a block diagram showing one embodiment of an audio input method in a speech recognition circuit according to the present invention. DESCRIPTION OF SYMBOLS 1... Microphone for target audio, 2... Microphone for noise, 5... Analog subtracter, 8... Expander circuit, 10... Electronic volume circuit. Patent applicant K.M.G. Co., Ltd.

Claims (1)

[Claims] Using a microphone for the target voice and a microphone for the noise, amplify the sound that entered the target voice microphone and the sound that entered the noise microphone, respectively, and remove the same sound that entered each with an analog subtracter. The output of the analog subtracter is output by an expander circuit that makes loud sounds louder and soft sounds softer, and then cuts signals below a certain level to eliminate unintended noise from small signals that could not be removed by the analog subtracter. On the other hand, sudden loud noises are prevented from malfunctioning by activating the electronic volume circuit and temporarily shutting down the circuit. Furthermore, the target signal is compressed by a compressor circuit to equalize the sound and recognize it as an audio signal. Audio input method to feed into the circuit.