JPH01197796A - voice recognition device - 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] To the rightful person” The present invention relates to a speech recognition device.

Research on speech recognition is active, but there are two types of speech recognition: the speaker-specific method, in which the user registers his or her own voice, and the speaker-independent method, which can be used without any preparation. The former has a higher recognition rate than the latter. On the other hand, instead of using these as separate machines, it is conceivable to have both functions in one and use them properly as needed. In this case, it is convenient to recognize words that are commonly used by a large number of people using the speaker-independent method, and to recognize words that are not commonly used using the speaker-specific method. Alternatively, it may be convenient to leave commands for specific specifications unspecified, so that when you say your name, the person's voice dictionary is searched from the disk and loaded. . In such cases, one device is used by several people, and if one person gets up while using it, the voice dictionary of the previous person will be loaded when the next person uses it. There are up to
It has the disadvantage of not being recognized.

Purpose The present invention was made in view of the above-mentioned circumstances.
In particular, when a single recognition device is used by multiple people, the purpose of this invention is to provide a device that can recognize any person's voice at any time.

Configuration In order to achieve the above object, the present invention provides means for converting audio into an electrical signal, a feature converter for extracting features from this signal, a first storage for storing the features, and the like. and a second storage unit separate from the .

In a speech recognition device having a comparison unit for comparing feature amounts and a holding unit for holding the compared similarity degree.

It has a time measuring section, and the time measuring section measures the time during which no voice is input, and when the value exceeds a predetermined value, the first inputted voice is stored in the second storage section. The feature is that the comparison is made only with the feature quantity, or it is equipped with a place for placing a device that converts audio into an electrical signal, and is also equipped with a switch, and by placing the device, the true key switch can be opened and closed. Accordingly, the objects to be compared in the comparison section are limited to those stored in the second storage section. Hereinafter, the present invention will be explained based on examples.

FIG. 1 is a configuration diagram for explaining one embodiment of the present invention, in which 1 is a microphone, 2 is a group of band-pass filters, 3 is a changeover switch, 4.5 is a register, 6 is a changeover switch, 7 is a comparison section, 8 is a summation section, 9 is a comparator, 10 is a threshold generation section, 11 is a time counter, and 12 is a recognition result output section. A feature conversion unit that extracts features from a signal, a storage unit that stores the feature values, a comparison unit that compares the feature values with a separate second storage unit, and a storage unit that stores the compared similarity degrees. In a speech recognition device having a holding section, a time measuring section is provided to measure the time during which no voice is input, and when the value exceeds a predetermined value, the first input voice is recorded as a characteristic in the second storage section. It is designed to compare only the quantity. In FIG. 1, a microphone is used as the acoustic/electrical signal converter 1, and its output is applied to a group of band-pass filters, which is a feature converter. - It goes without saying that a microphone amplifier or a voice section detection section may be inserted between the bandpass filter group 2 and the microphone 1. Thereafter, the switch 3 separates voice registration and recognition. When switch 3 is turned to the C side, it is registered, and when it is turned to the D side, it is recognized. The register 4 stores a dictionary for unspecified speakers, and the register 5 stores voices registered by each user. for example,
Register 4 is “Start” “End J rlJ
r2Jr3J r4J... It can be thought of as a ROM in which voice patterns such as "9" are registered. Also, register 5
It is desirable to be able to store or load the contents into a hard disk or the like (the hard disk is not shown).
Or, the pattern No. 5 is compared with the input pattern to determine the degree of similarity. In this case, the patterns may be compared using a matching method using dynamic programming known as DP matching or a method using binarized data.

As a result of the comparison by the comparison unit 7, a signal corresponding to the voice determined to be most similar is output as a recognition result 12.

Also, when the switch 3 is turned to the D side and used, the output is always summed up by the number of bandpass filters in the summation section 8, that is, the energy for each time is calculated and the value is added to the threshold section 1o. Compare whether it is greater or less than a specified value. The threshold value can be set by averaging the output at point E over a certain period of time when no voice is input from the microphone and determining that value. If the input is larger than this threshold, the comparator 9
11, the tie 11 counter 11 is reset. The time counter 11 may be of any type as long as it can measure a certain amount of time. For example, it may be reset and set to a large value, and the time taken until it reaches O by repeating subtraction is measured. When the value becomes 0, the switch 6 is turned to the A side by a signal. The setting value of the time counter 11 is preferably about 1 minute to 5 minutes. First, a number is assigned by the user.

Assuming that user number 3 is using it, when he flips switch 3 to the D side and utters "3" from microphone 1, it is checked against the dictionary for unspecified speakers in register 4, and the result is "3". . This recognition result may be used in any way, but here it is assumed that the voice data of the third speaker from the disk is loaded into the register 5 (not shown). Turn the switch 6 to the B side, and the speaker will be recognized using the dictionary of the third speaker. If this user gets up from his or her seat midway, the time counter 11 will operate and the switch 6 will automatically return to the A side.Next, when the second speaker uses this device, he or she should face the microphone 1. You can just say "2" and there is no need to check whose dictionary is loaded.

FIG. 2 is a configuration diagram for explaining another embodiment of the present invention. In FIG. 1, numeral 13 indicates a microphone stand;
In the figure, parts having the same function as in the embodiment are given the same reference numerals as in FIG. 1.

Therefore, this embodiment includes a means for converting audio into an electrical signal, a feature amount converting section that extracts features from this signal, a storage section that stores the feature amounts, and a second separate device. A speech recognition device having a storage section, a comparison section for comparing feature amounts, and a holding section for holding the compared degree of similarity, the speech recognition device including a place for arranging a device for converting speech into an electrical signal and a switch, By arranging the device, the switch is opened and closed, thereby limiting the objects to be compared in the comparison section to those in the second storage section. In the case of the embodiment shown in FIG. 2, there is no time counter 11 or comparator 9 used in the embodiment shown in FIG. 1, and a microphone stand 13 is provided instead.
When a load is applied to the microphone stand 13, the switch 6 is turned to the A side. In the case of the embodiment shown in FIG. 1, the switch 6 is moved to the A side, that is, the register 4 side for unspecified speakers, by measuring the time when the user does not use the device. In the embodiment shown in FIG. 2, when the user leaves his/her seat and places the microphone 1 on the microphone stand 13, the switch 6 is tilted toward the A side.゛Although in the above example, register 4 for unspecified speakers was explained as a ROM, this is not necessary, and it can also be made registrable so that each user can register his or her name using his/her own voice. You can leave it there.

Kazuki HiroAs is clear from the above explanation, according to the present invention, even when the user changes, the voice recognition device can be used from the beginning.
There is no need for complicated procedures.

[Brief explanation of the drawing]

FIG. 1 and FIG. 2 are configuration diagrams for explaining the present invention in detail, respectively. DESCRIPTION OF SYMBOLS 1... Microphone, 2... Bandpass filter group, 3... Changeover switch, 4.5... Register, 6
... Selector switch, 7... Comparison section, 8... Adder, 9... Comparator, 10... Threshold value generation section, 11...
Time counter, 12... Recognition result output unit, 13...
·Mike stand.

Claims (1)

[Claims] 1. A means for converting audio into an electrical signal, a feature amount converting section for extracting features from this signal, a first storage section for storing the feature amounts, and a separate second storage section. 2, a comparison unit for comparing feature amounts, and a storage unit for holding the compared similarity degree, the speech recognition device having a time measurement unit,
The time measurement unit measures the time during which no audio is input, and when the value exceeds a predetermined value, the next input audio is compared only with the feature amount stored in the second storage unit. A voice recognition device characterized by: 2. means for converting audio into an electrical signal, a feature converter for extracting features from this signal, a first storage for storing the features, and a second storage separate from this; A speech recognition device having a comparison section for comparing feature amounts and a holding section for holding the compared similarity degree, the speech recognition device having a place for arranging a device for converting speech into an electrical signal, and a switch for arranging the device. The voice recognition device is characterized in that the switch is opened and closed by arranging the switch, thereby limiting the objects to be compared in the comparison section to those stored in the second storage section.