JPS61109100A - Voice recognition system - Google Patents

Abstract

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

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a speech recognition method, and particularly to a method for adjusting the sensitivity of input speech in a speech recognition device.

(Prior Art) Sensitivity adjustment methods for input speech in conventional speech recognition devices can be roughly divided into two types: a fixed method and a UF transformation method. The fixed method fixes the gain of the amplifier that amplifies human voice. Variable methods include those equipped with a variable gain mechanism so that the sensitivity can be changed freely, and those using an AGC (automatic gain control) mechanism.

(Problems to be Solved by the Invention) However, the prior art system has the following problems.

In the case of a fixed method, there is a drawback that it is difficult to fix the gain of the amplifier to the best sensitivity because the sound volume varies depending on the operator.

In the case of the variable method, there is a drawback that the gain at the time when the sensitivity is adjusted to the best sensitivity cannot be stored. Therefore, there may be a difference in gain between standard pattern creation and speech recognition, resulting in a reduction in recognition rate. Another disadvantage is that the sensitivity must be adjusted again every time the operator changes.

The present invention solves the problems of the prior art and provides a speech recognition system having a function of automatically setting and automatically storing the gain of an audio signal amplifier.

Means for Solving Problem C) In order to solve the above problem, the present invention amplifies an input audio signal with an audio amplifier, and analyzes the amplified audio signal with an audio analysis means to create a audio pattern. , in a speech recognition method in which speech recognition is performed by comparing the speech pattern with a standard pattern created in advance in the same manner, a storage means for storing the standard pattern and a gain value of the audio amplifier at the time of creating the standard pattern; , a gain control means for setting a gain value of the audio amplifier and holding the gain value, and a recognition control means for performing speech recognition by pattern matching and controlling the storage means and the gain control means, When creating a pattern, the recognition control means outputs the gain value and the standard pattern held by the gain control means to the storage means, and when matching the pattern, the recognition control means outputs the standard pattern and gain value stored in the storage means. This is a voice recognition method in which the gain value is read out and set in the voice amplifier via the gain control section.

(Operation) According to the present invention, since the voice recognition system is configured as described above, the technical means operates as follows. When creating a standard pattern, the gain control means sets an appropriate input gain value to the audio amplifier and works to hold that gain value, and the recognition control means sets the input appropriate gain value to the audio amplifier and works to hold the gain value, and the recognition control means sets the input appropriate gain value to the audio amplifier and works to maintain the gain value. The recognition control means operates to output the parameterized standard pattern and the gain value held in the gain control means to the storage means, and during pattern matching, the recognition control means reads out the standard pattern and the fir value stored in the storage means. At the same time, the gain value is set in the audio amplifier through the gain control means, and the recognition control means collates the characteristic parameterized audio pattern with the standard pattern through the audio amplifier and audio analysis means to determine the audio output. Therefore, the above-mentioned problems can be solved.

(Embodiment) FIG. 1 is a diagram showing a first embodiment of the speech recognition method according to the present invention, and shows the configuration of a specific speaker speech recognition device. Microphone 1 converts input audio into an electrical signal. The audio signal amplifier 2 amplifies the audio signal output from the microphone l. The filter 3 emphasizes the high frequencies of the audio signal output from the audio signal amplifier 2. filter·
Bank 4 performs spectrum analysis on the high frequency emphasized signal to create a spectrum pattern. When creating a standard pattern, the recognition control unit 5 creates a standard pattern by converting the spectral pattern obtained by the filter bank 4 into feature parameters, and uses the gain value of the audio signal amplifier 2 held in the gain control unit 9, which will be described later, as well as the gain value of the audio signal amplifier 2. The data is stored in the storage medium 6 of. During pattern matching, the stored standard pattern and gain value are read out, the gain value is set in the audio signal amplifier 2 via the gain control section 9, and then the spectral pattern obtained by the filter bank 4 is converted into a characteristic parameter. The recognition control unit 5 has the function of performing voice recognition by pattern matching the voice pattern for comparison with the read standard pattern, and controlling the entire apparatus. The storage medium 6 stores the standard pattern transferred from the recognition control unit 5 and the gain value of the audio signal amplifier 2 used when creating the standard pattern. The recognition result display 7 displays the recognition result transferred from the recognition inter-gi5. The gain setting key 8 outputs the gain value to the gain control section 9 when the operator inputs the gain value of the audio signal amplifier 2 . As will be described later, the gain control section 9 outputs the gain value set by the gain setting key 8 to the audio signal amplifier 2 and holds it therein when creating the standard pattern. Further, the gain control unit 9 outputs to the audio signal amplifier 2 the gain value of the audio signal amplifier 2 which is transferred from the recognition control unit 5 and used when creating the standard pattern. The gain display 10 displays the gain value input from the gain control section 9 and set in the audio signal amplifier 2. The audio output level display 11 displays the output level of the audio signal amplifier 2.

The internal configuration of the audio signal amplifier 2 is shown in FIG.

The audio signal amplifier 2 includes an operational amplifier 12. Analog multiplexer 13. This is a programmable gain amplifier (known) consisting of a resistor 14 (Ro) and a resistor array 15 (R1 to R8). The non-inverting terminal of the operational amplifier 12 is grounded, and the inverting terminal is connected to the resistor 14.
It is connected to the input terminal 2a via the resistor array 15 and the analog multiplexer 13 to the output terminal of the operational amplifier 12. The audio signal amplifier 2 is an inverting amplifier due to the configuration of zero or more output terminals connected to the output terminal 2b.

The gain value is changed to the switching input terminals 13a, b, c of the analog self-multiplexer 13 via the gain control unit 9 by operation of the gain setting key 8 or by a signal from the recognition control unit 5 (CPU) in the 2.3 embodiment described later. is input, and a resistor Ri (1=1 to 8) corresponding to the gain value is selected. The audio signal input to the input terminal 2a is amplified by the amplification factor determined by the resistors RQ and R; and is passed from the output terminal 2b to the filter 3.
Output to. Further, its size is displayed on the audio output level display 11.

The internal configuration of the gain control unit 9 is shown in FIG.
is composed of a key controller 21, registers 22 and 23, a data selector 24, a display controller 25, and a driver 28. The key controller 21 input-controls the gain signal A indicating the gain value input from the gain setting key 8 and sends it to the register 22 . The register 22 outputs the gain signal A from the key controller 21 to the data selector 24 and holds it. When the register 23 receives a write signal from the recognition control section 5, it outputs a gain signal B indicating the gain value stored in the recognition control section 5 to the data selector 24 and holds it. The data selector 24 selects one of the gain signals A and B according to the key input enable signal and outputs it to the audio signal amplifier 2, display controller 25, and driver 26. The display controller 25 causes the gain display 12 to display the gain value indicated by the gain signal input from the data selector 24. The driver 26 outputs the gain signal input from the data selector 24 to the recognition control unit 5 when the read signal is input from the recognition control unit 5 . Therefore, the recognition control section 5 sends the key manual enable signal to the data selector 24 of the gain control section 9, thereby inputting the gain value from the gain setting key 8 (gain signal A) and the gain value from the recognition control section 5 (gain signal B). ) can be switched.

Further, the recognition control section 5 can read out the gain value (gain signal A) set by the gain setting key 8 by sending a readout signal to the driver 26 of the gain control section 9.

Next, the operation of the first embodiment will be explained.

When creating a new standard pattern, it operates as follows. While inputting sound into the microphone 1, the gain setting key 8 is operated to set the gain value of the audio signal amplifier 2 so that the display on the audio output level display 11 becomes a value suitable for recognition. The gain value at this time is displayed on the gain display 10 and held in the gain control section 9. The audio signal amplified by the audio signal amplifier 2 is passed through a filter 3 and a filter bank 4 for spectrum analysis. The recognition control unit 5 converts the spectrum pattern subjected to the spectrum analysis into characteristic parameters and creates them as a standard pattern on the internal memory 1B. Next, the recognition control section 5 transfers the gain value held by the gain control section 9 to the gain area 17 in the memory 16. Figure 5 shows how the memo January 6 was used at this time. Next, the contents of the memory 16 are transferred to the storage medium 6 (floppy disk, etc.) to complete recording of the display pattern.

When performing speech recognition, it operates as follows. Prior to speech recognition, the recognition control unit 5 transfers the opaque standard pattern and gain value stored in the storage medium 6 onto the memory 16. When the transfer is completed, the gain control unit 9 reads the gain value from the gain area on the ref memory 16 under the control of the recognition control unit 5, sets the gain value in the audio signal amplifier 2, and displays it on the gain display 10. do. Once this gain value is set in the audio signal amplifier 2, a speech recognition operation is performed. That is, when a voice is input to the microphone 1, the voice signal is amplified by the voice signal amplifier 2 set at the gain value at the time of creating the standard pattern, and then passed through the filter 3 and filter bank 4 to become a spectral pattern. The recognition control unit 5 converts this spectrum pattern into feature parameters and uses it as a speech pattern to be matched. The speech pattern to be matched and the standard pattern read out on the memory 16 are pattern matched to perform speech recognition, and this result is sent to the recognition control section 5.
is displayed on the recognition result display 7.

FIG. 4 is a diagram showing a second embodiment of the speech recognition method according to the present invention, and shows the configuration of a speech input Japanese language processing device. The same reference numerals as in FIG. 1 indicate identical components. The voice input device 18 is the gain setting key 8 in the first embodiment. This figure excludes the gain display 10 and the recognition result display 7. The Japanese language processing device 19 has a keyboard 19a for inputting gain values manually, and a display device 19b for displaying gain values, recognition results, etc. Voice input device 1
8 is connected to a Japanese language processing device 19.

When creating a new standard pattern, inputting voice into the microphone l, operate the keyboard 19a of the Japanese language processing device 18 so that the display on the voice output level display 11 becomes a value suitable for recognition.

At this time, the gain value entered manually from the keyboard 19a is displayed on the display screen of the display device 19b of the Japanese language processing device 13, and is also transferred to the gain area 17 of the memory 16 in the recognition control unit 5 of the voice input device 18. be done. Recognition control unit 5
Then sets a gain value in the gain control section 9. Once an appropriate gain value has been set, a standard pattern is created on the memory 16 in the recognition control section 5 in the same manner as in the first embodiment.

After creating the standard pattern, the contents of the memory 16 are transferred to the storage medium 6 (floppy disk, etc.) to complete recording of the standard pattern.

The recognition operation is the same as in the first embodiment.

Next, the third embodiment of the speech recognition method according to the present invention will be explained in the fourth embodiment.
This will be explained using figures. The third embodiment is an improved version of the second embodiment. The Japanese language processing device 18, which is a higher-level device than the voice input device 18, first enters the standard pattern creation mode. The Japanese language processing device 18 is the recognition control section 5 in the voice input device 1B.
, the recognition control section 5 instructs the gain control section 9 to be in the standard pattern creation mode, and sets the initial value of the gain to the gain control section 9 . Next, the operator utters several designated syllables according to the guidance on the display screen of the display device 19b of the Japanese language processing device 18. The recognition control unit 5 analyzes the voice power of the generated voice and calculates the most suitable gain value. The recognition control unit 5 transfers the optimum gain value to the gain control unit 9 and the gain value area 17 in the memory 16.

Once the setting of the optimal gain value is completed, the Japanese language processing device 1
Following the guidance on the display screen of the display device 19b of 8, a standard pattern is created. Once the creation of the standard pattern is completed, the standard pattern and gain value in the memory 16 are stored in the recording medium 6.

The recognition operation is the same as in the first embodiment.

No.? , 3, the recording medium 6 is connected to the recognition control section 5, but the Japanese language processing device 18
This can also be done with a storage medium connected to.

This embodiment has a function that automatically stores the gain value when creating the standard pattern in the same storage medium as the standard pattern, and also transfers the gain value at the same time when the standard pattern is transferred from the storage medium to the memory prior to recognition. Since this is a voice recognition method that has a function that automatically completes the sensitivity adjustment of the input voice, if there are multiple operators, even if the operator changes, you only need to select the standard pattern to use, and there is no need to adjust the sensitivity again. It has the advantage of not having In addition, during the detection operation, the gain value at the time of standard pattern creation is automatically set to the audio signal amplifier, improving the reproducibility of the audio signal.
Misrecognition can be reduced.

(Effects of the Invention) As explained above, according to the present invention, the gain value of the audio amplifier during standard pattern creation can be used during pattern matching, thereby improving the reproducibility of audio signals and achieving highly accurate speech recognition. can be done easily.

[Brief explanation of drawings]

FIG. 1 is a diagram showing a first embodiment of the speech recognition system according to the present invention, FIG. 2 is a diagram showing the audio signal amplifier in detail, FIG. 3 is a diagram showing the gain control section in detail, and FIG. FIG. 5 is a diagram showing the second and third embodiments of the present invention, and FIG. 5 is a diagram showing how the memory is used. 2... Audio signal amplifier, 3... Filter, 4...
... Filter puncture, 5. Recognition control section, 6.
...Storage medium, 9...Gain control section.

Claims (1)

[Claims] The input voice signal is amplified by a voice amplifier, the amplified voice signal is analyzed by a voice analysis means to create a voice pattern, and the voice pattern is compared with a standard pattern created in advance in the same manner to perform voice recognition. In the voice recognition method used,
A storage means for storing a standard pattern and a gain value of the audio amplifier at the time of creating the standard pattern, a gain control means for setting the gain value of the audio amplifier and holding the gain value, and pattern matching to perform speech recognition. and a recognition control means for controlling the storage means and the gain control means, and when creating a standard pattern, the recognition control means outputs the gain value held by the gain control means and the standard pattern to the storage means. , a voice recognition system characterized in that during pattern matching, the recognition control means reads out a standard pattern and a gain value stored in the storage means, and sets the gain value in the voice amplifier via the gain control section. .