JPH05224694A - Speech recognition device - Google Patents

Abstract

PURPOSE:To obtain the largest recognition rate regardless of whether the speech level of a speaker is high or low by judging the speech level of an inputted hot key and setting the gain of the whole system corresponding to the judged level. CONSTITUTION:An speech signal which is inputted is amplified by a microphone amplifier, divided into respective frequency bands by a filter bank 4, and digitized, and then, a process part 6 extracts features. The feature-extracted speech pattern is divided into speech sections, which are stored in an input pattern storage part 8. When the stored speech signal is inputted to a recognition and collation part 9, it is judged that a hot key process is completed; when the hot key is inputted, a selector 15 selects one of hot key dictionaries 14 provided by gains and the recognition and collation part 9 performs collation by using this dictionary to detect the dictionary which gives maximum similarity. The speech level of the speaker is estimated from the gain of this dictionary and the speech level of the whole system is so set as to obtain the maximum recognition rate.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device for recognizing voice input through a microphone.

[0002]

2. Description of the Related Art One of the major obstacles to improving the recognition rate of this type of voice recognition device is the individual difference in voice level. From the viewpoint of the power of voice, a person with a loud sound and a person with a low sound make several tens of dB. There are differences. As shown in FIG. 1, between the voice level and the recognition rate, a high recognition rate cannot be obtained in a low level region, and an effective recognition rate cannot be obtained unless the level is above a certain level. In other words, it is difficult at present to guarantee a high recognition rate over a wide range from low level to high level, and as a result, the device must be set so that the maximum recognition rate can be obtained at an average voice level. I don't get it. Therefore, in that case, the necessary recognition rate cannot be obtained for low-level voice input.

Therefore, conventionally, there has been proposed a microphone amplifier to which an AGC (automatic gain control circuit) is added so that a substantially constant voice input is always obtained (Japanese Utility Model Laid-Open No. 59-60700). reference). However, the above AGC can handle only about 40 dB, and as described above, it is not possible to sufficiently cope with the individual difference of several tens of dB.

There is also proposed a method in which a speaker is informed of the voice level of the speaker by an appropriate means so as to urge the speaker to speak at an optimum level (Japanese Utility Model Publication No. 01-137497). JP-A-63-014200
(See Japanese Patent Publication). However, the utterance level is individual and individual, and if it is forcibly changed, the utterance itself becomes unnatural, which may rather lower the recognition rate, and even if the same person utters, the consonant and vowel sounds are 10 dB.
Due to the above power difference, there is a problem that it is actually difficult to set the optimum level.

[0005]

SUMMARY OF THE INVENTION Therefore, a technical problem of the present invention is to automatically obtain the maximum recognition rate in accordance with the utterance level of the utterer.

[0006]

Therefore, according to the present invention, a microphone, a microphone amplifier for amplifying a sound signal output of the microphone, an A / D converter for converting an output signal of the microphone amplifier into a digital signal, and an A / D converter A feature extraction unit for extracting features of the audio signal output from the D converter,
A voice section detection unit that detects a voice section from the voice signal, an input pattern storage section that stores the feature-extracted pattern between the detected voice sections as an input pattern, and an input pattern that compares the input pattern with a dictionary pattern. In a voice recognition device having a voice recognition unit that specifies the maximum similarity, a plurality of hot key dictionaries for each gain are provided, and when a hot key is input, the similarity is calculated using each of these hot key dictionaries to obtain the maximum similarity. Provided is a voice recognition device comprising means for setting a gain of the voice recognition device for a voice to be subsequently input according to the gain set in the hot key dictionary obtained.

That is, in the present invention, the hot key input is used, and the gain of the entire apparatus is set so as to obtain the maximum recognition rate according to the voice level of the input hot key. Here, the gain of the entire device is
Not only the gain of the microphone amplifier that directly amplifies the voice input, but also means for indirectly controlling the gain for the voice input, for example, the reference voltage applied to the A / D converter and the maximum recognition rate in the low level region. Means a dictionary or the like such that

More specifically, a plurality of gains can be selectively set for the microphone amplifier, and the optimum gain is set for the microphone amplifier according to the gain set in the hot key dictionary for which the maximum similarity is obtained. Alternatively, the reference voltage of the A / D converter may be optimally set.

Further, a plurality of gain-specific dictionaries are prepared, and a dictionary is selected according to the gain of the gain-specific hot key dictionaries for which the maximum similarity is obtained, for subsequent recognition. May be.

[0010]

According to the present invention, the gain of the entire voice recognition device can be optimally set according to the voice level of the speaker by using the hot key input, and an unnatural voice is given to the speaker. It becomes possible to always perform voice recognition at the maximum recognition rate without being forced.

According to the invention of claim 2, the maximum recognition rate can be guaranteed by preparing a plurality of microphone amplifiers for each gain and selecting the most suitable microphone amplifier for the voice level of the speaker. .. According to the invention of claim 3,
The maximum recognition rate can be guaranteed by optimally setting the reference voltage for the A / D converter according to the voice level of the speaker.

Further, according to the invention of claim 4, the maximum recognition rate can be secured by preparing a plurality of dictionaries for each gain and selecting a dictionary according to the voice level of the speaker.

[0013]

EXAMPLES Examples of the present invention will be specifically described below. (Basic System) FIG. 2 shows a basic system of the voice recognition device according to the present invention. The audio signal input from the microphone 1 is pre-processed by the pre-processing unit 2 and then amplified by the automatic gain control circuit (AGC) that constitutes the microphone amplifier 3, and separated by the filter bank 4 for each frequency band. A
The signal is converted into a digital signal by the / D converter 5, is input to the processing unit 6, and feature extraction is performed. The voice signal for each frequency is also input to the voice section detection circuit 7,
The voice section is detected. The feature-extracted voice pattern is stored in the input pattern storage unit 8 as an input pattern in a state of being divided by the detected voice section.

The stored input pattern is recognized by the recognition and collation unit 9
In, the template is collated with each template of the dictionary 10 and the similarity is calculated in the similarity calculator 11. The similarity calculation unit 11 detects a template that gives the maximum similarity among the calculated similarities, and controls the entire system.
Output this to. The control unit 12 outputs the recognition result corresponding to the input template to the result output unit 13, and the result output unit 13 displays the recognition result.

The above speech recognition apparatus is designed to disclose the operation by so-called hot key input. When a hot key input is made, a plurality of hot key dictionaries 14 provided for each gain are selected by a selector 15. Then, the recognition matching unit 9 performs matching using the hot key dictionary 14 and detects a hot key dictionary that gives the maximum similarity.

FIG. 3 shows a hot key input processing routine executed by the control unit 12 for optimally setting the volume level of the system.

When the voice signal input stored in the input pattern storage unit 8 is input to the recognition and collation unit 9, first, step S
In 1, the flag is used to determine whether or not the hot key processing is completed. If the hot key processing is completed, normal recognition processing is executed. If the flag is not set, that is, if the hot key processing is not completed, step S2
In, it is determined whether the voice input is a hot key,
If it is determined that the key is a hot key, step S3
In, the similarity is compared using the hot key dictionary 14 provided in advance for each gain. The hot key dictionary for each gain is, for example, a standard voice level (0 dB), ± 6d.
A hot key dictionary that produces the maximum similarity to the input hot key is detected in advance with the voice level of B, ± 12 dB. In step S4, the voice level of the speaker is estimated from the gain of the hot key dictionary that gives the maximum similarity. For example, if the hot key dictionary of -6 dB gives the maximum similarity, the standard level is calculated in step S5. The audio level of the system as a whole is set so that the recognition rate of the system is maximized at the audio level lower than the audio level of 6 dB. When the setting of the voice level is completed, a flag indicating the end of the hot key process is set in step S6, and the process returns.

(System Audio Level Setting Method) (Part 1) FIG. 4 shows an example of the audio level setting method. As shown in the figure, in this example, the gain-specific microphone amplifiers 3-1, ..., 3-5 corresponding to the five gain-specific hot key dictionaries 14-1 ,. I try to select a microphone amplifier with a gain corresponding to the level.

That is, with respect to the standard voice level KdB,
By the microphone amplifier selector 20, a total of five microphone amplifiers 3-1, 3-2, ..., 3-5, which are scheduled to have an audio level of ± 12 dB and ± 6 dB and have been given an optimum gain for each audio level, are selected. select. In the hot key processing shown in FIG. 2, for example, when it is determined that the hot key dictionary 14-2 of +6 dB gives the maximum similarity, this selection is temporarily stored in the most similar dictionary storage unit 21, The microphone amplifier selector 20 reads the type of the hot key dictionary stored in the dictionary storage unit 21, and if it is the +6 dB hot key dictionary 14-2, it corresponds to this (K
Select +6) dB microphone amplifier 3-2.

Although FIG. 4 does not correspond exactly to FIG. 2, the same parts as those shown in FIG. 2 are designated by the same reference numerals and further description will be omitted.

(Part 2) FIG. 5 shows another example of the audio level setting method. In this case, the A / D converter 5
The reference voltage to be applied to is selected according to the audio level. That is, with respect to the reference voltage (displayed as 0 dB) corresponding to the standard audio level, the reference voltages of ± 12 dB and ± 6 dB are set to be selectable in advance and stored in the most similar dictionary storage unit 21. The selector 22 selects the reference voltage corresponding to the gain of the hot key dictionary. In FIG. 5, as in FIG. 4, the same parts as those in FIG.

(Part 3) FIG. 6 shows another example of the audio level setting system. As shown in the figure, in this example, a dictionary used for normal voice recognition is provided for each gain. Specifically, in addition to the dictionary created at the standard voice level (displayed as 0 dB dictionary),
Five dictionaries 10-1 and 10 created at voice levels that differ from the standard voice level by ± 6 dB and ± 12 dB.
-2, ..., 10-5 is a hot key dictionary for each gain 14-
1, 14-2, ... 14-5 are prepared in advance,
For example, if the hot key dictionary stored in the most similar dictionary storage unit 21 is +6 dB 14-2, the selector 23 selects the corresponding +6 dB dictionary 10-2. Note that, in FIG. 6, as in FIG. 4, the same parts as those in FIG.

[Brief description of drawings]

FIG. 1 is a graph showing a relationship between an input voice level and a recognition rate.

FIG. 2 is a schematic system diagram of a voice recognition device according to the present invention.

FIG. 3 is a flowchart of hot key input processing executed in the present invention.

FIG. 4 is a schematic explanatory view showing an embodiment of the present invention.

FIG. 5 is a schematic explanatory view showing another embodiment of the present invention.

FIG. 6 is a schematic explanatory view showing another embodiment of the present invention.

[Explanation of symbols]

1 ... Microphone 3 ... Microphone amplifier 3-1, 3-2, ..., 3-5 ... Gain-specific microphone amplifier 5 ... A / D converter 6 ... Processing unit 7 ... Voice section detection unit 9 ... Recognition / collation unit 10 ... Dictionary 11 ... Similarity calculation unit 12 ... Control unit 14 ... Hot key dictionary 10-1, 10-2, ..., 10-5 Gain-specific dictionary 14-1, 14-2, ..., 14-5 Gain-specific hot key dictionary

Claims (4)

[Claims] 1. A microphone, a microphone amplifier for amplifying an audio signal output of the microphone, an A / D converter for converting an output signal of the microphone amplifier into a digital signal, and
A feature extraction unit that extracts a feature of a voice signal output from the D / D converter, a voice segment detection unit that detects a voice segment from the voice signal, and a feature extracted pattern between the detected voice segments as an input pattern. In a voice recognition device having an input pattern storage unit for storing and a voice recognition unit for identifying an input pattern by comparing the input pattern with a dictionary pattern, a plurality of hot key dictionaries for each gain are provided, and at the time of hot key input, A means for calculating the similarity using each of the plurality of hot key dictionaries, and setting the gain of the voice recognition device for the subsequently input speech according to the gain set in the hot key dictionary for which the maximum similarity is obtained. A voice recognition device comprising: 2. The voice recognition device according to claim 1, wherein the microphone amplifier has a plurality of selectable gains, and the gain set in the hot key dictionary that obtains the maximum similarity is obtained. Correspondingly, the microphone amplifier gain is selected. 3. The voice recognition device according to claim 1, wherein the reference voltage of the A / D converter is set according to the gain set in the hot key dictionary for which the maximum similarity is obtained. What is characterized by doing. 4. The voice recognition apparatus according to claim 1, wherein a plurality of dictionaries for each gain are provided, and a dictionary corresponding to the gain set in the hot key dictionary for which the maximum similarity is obtained is selected. Characterized by the fact that