JPS6324297A - Voice dictionary generation system for specified speaker's voice recognition equipment - 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] [Summary] The present invention provides a voice recognition dictionary for a specific speaker voice pressure ggli device, which uses a common dictionary created in advance from the utterances of words used by a plurality of people, and a common dictionary of input person's utterances. This system is configured with a specific dictionary created exclusively for the input user for words that are misrecognized due to recognition of words, thereby reducing the number of utterances and dictionary area required for registration.

[Industrial application field]

The present invention relates to a speech dictionary creation method in a specific speaker speech recognition device.

When using a specific speaker speech recognition device, the speech of a skilled person is registered in advance, and during recognition, the uttered speech is compared against the registered speech and used as a recognition result. The registration and
Furthermore, it is desirable for the user to reduce the number of times of utterance, that is, learning, for correcting registered voice parameters in order to increase the recognition rate.

Also, regarding the dictionary area, the smaller the dictionary area for a specific person, the greater the number of user's dictionaries that can be built into the same device. Alternatively, if the number of users is small, it is desirable because the device can be made smaller.

[Prior Art] A speech recognition dictionary creation method of a conventional speaker-specific speech recognition device is shown in the device configuration diagram of FIG.

show.

Reference numeral 5 denotes a host device that connects to the speech recognition device 1, controls it, and uses the speech recognition results.

10 is a speech recognition dictionary provided in the speech recognition section.

30 is an input line, 31 is an internal interface line, and 32 is an external interface line.

Conventionally, as shown in FIG. 4, the speech recognition dictionary 10 was comprised only of a specific dictionary 20 dedicated to the input user.

That is, all the people who perform voice input speak out all the words to be input, and perform the registration process of storing the voice parameters as a voice recognition dictionary dedicated to each inputter.

After that, in order to improve the recognition rate, the robot performs a learning process in which it utters the words multiple times and corrects the speech recognition dictionary.

[Problems to be Solved by the Invention] As mentioned above, the voice input dictionary of the conventional speaker-specific voice input device requires each user to speak a large number of times in order to register and learn the voice input dictionary. In addition, since it was configured with only a specific dictionary, if the person who inputs the data changes, the entire specific dictionary must be recreated, or if it has already been created, it can be completely rewritten using an external storage device, etc. Must be replaced.

Furthermore, when a plurality of voice input dictionaries are built in, there is a problem in that a large voice input dictionary containing all the voices for the number of users is required.

The present invention solves these conventional problems.

The present invention aims to provide a new voice input dictionary creation method for a specific speaker voice input device.

Means for Solving Problem E] FIG. 1 shows a block diagram of the principle of a speech recognition dictionary creation method for a specific speaker speech recognition apparatus of the present invention.

In FIG. 1, reference numeral 21 denotes a common dictionary, which is created in advance based on voice parameters uttered by several or dozens of people regarding the words to be used.

41 is a speech parameter extraction function provided in the speech recognition section.

42 is a comparison and verification function provided in the speech recognition section.

The utterances of the inputting person are processed by the sound parameter extraction section 41 and the sound comparison/verification function 42. Words that are not correctly inputted are registered in the specific dictionary 22 for that user. Correctly recognized words are not registered in the specific dictionary 21.

In this way, the speech recognition dictionary 10 is composed of the common dictionary 21 and the specific dictionary 22, and during recognition, the speech parameters of the user are compared and recognized with the common dictionary 21 and the specific dictionary 22 for the user.

[Operation] The speech recognition dictionary 10 is configured to be divided into a common dictionary 21 and a specific dictionary 22, and since the common dictionary 21 has already been created, there is no need for a special task of voice registration by the input person.

Furthermore, by storing voice patterns unique to the input person in the specific dictionary 22, a high recognition rate can be obtained.

[Example] The present invention will be explained in more detail with reference to an example shown in FIGS. 2 and 3.

FIG. 2 is a diagram showing the configuration of an apparatus according to an embodiment of the present invention.

In Fig. 2, the differences from the conventional example shown in Fig. 4 are as follows.
This is because the speech recognition dictionary 10 is divided into a common dictionary 21 and a specific dictionary 22.

FIG. 3 is a diagram showing the flow of processing in an embodiment of the present invention.

In FIG. 3, reference numeral 21 denotes a common dictionary, in which voice parameters of words to be used uttered by several or several dozen people are collected in advance into several patterns and stored in the voice recognition dictionary 10.

22 is a specific dictionary dedicated to each inputter.

23 indicates audio parameters extracted from the input person's utterance.

The operation of the apparatus of this embodiment will be described below with reference to FIGS. 2 and 3.

(1) When the input person utters a word to be used from the microphone 4, the voice input from the microphone 4 is input to the voice recognition unit 2 of the specific speaker voice recognition device 1 via the input line 30.

(2) The speech recognition unit 2 creates speech parameters 23 from the input speech.

(3) The created audio parameters 23 and the common dictionary 2] are compared, and if there is a difference of more than a certain value, the audio parameters are created separately as the specific dictionary 22.

For example, the audio parameter 231 of Mr. A's first word is checked against all the words in the common dictionary 21, and the most similar word is the planned floor 1 word barameku 211, and the difference is within a certain numerical value. Therefore, no parameters are created in the specific dictionary 22.

Next, the second word parameter 2 of the phonetic parameter of the second word.
Similarly, 32 was compared with all words and the most similar word was the scheduled two-word parameter 212, but there was a difference of more than a certain value, or the most similar word was the scheduled two-word parameter 212. 212, the voice parameter is stored in the specific dictionary 22 as the Kan 2 language Mr. A specific parameter 221.

(4) Similarly, in the case of Mr. 8, if there is a difference of more than a certain value from the common dictionary 21 among the words uttered, 8 is added to the specific dictionary 22 as in the ltn word B limit parameter 222. A dictionary exclusively for him will be created.

(5) At the time of speech recognition, the host device 5 and the voice input control unit 3 specify the input person for the specific dictionary 22 created as described above, and limit the individual dictionaries in the specific dictionary 22. Decide which dictionary to check. Therefore, speech recognition is performed by comparing with designated personal dictionaries in the common dictionary 21 and the specific dictionary 22.

[Effects of the Invention] As explained above, according to the present invention, by using only the utterances that do not match the common dictionary created in advance as a specific dictionary, it is possible to create a dictionary with a high recognition rate with a small number of utterances. , Also, dictionaries for multiple people can be configured with fewer dictionary @ areas,
The effect of contributing to improving the performance and operability of the speech recognition device is significant.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of the principle of the present invention, Fig. 2 is a block diagram of a device according to an embodiment of the present invention, Fig. 3 is a diagram showing the processing flow of an embodiment of the present invention, and Fig. 4 is a diagram of a conventional device. FIG. In the drawings, ■ is a specific speaker voice recognition device, 2 is a voice recognition unit, 3 is a voice input control unit, 4 is a microphone, 5 is a host device,
10 is a speech recognition dictionary, 20 is a specific dictionary (
2] is a common dictionary, 22 is a specific dictionary (invention), 23 is a voice parameter, 30 is a human power line, 31 is an internal interface line, 32 is an external interface line, 41 is a voice parameter extraction function, 42 is a comparison Verification function, 211 to 214 are parameters for each word in the common dictionary, 22L 2
22 is a parameter in the specific dictionary 22, and 231 to 233 are audio parameters. 1N1. ,' Fig. 4 Block diagram of the principle of the present invention

Claims (2)

[Claims] (1) Speech recognition dictionary provided by the specific speaker speech recognition device (1
0), it consists of a common dictionary (21) created from the audio parameters uttered by multiple speakers regarding the words used, and a specific dictionary (22) created from the audio parameters uttered by the inputter about the words used. A speech dictionary of a specific speech speech recognition device, characterized in that the speech dictionary is configured to recognize the speech uttered by comparing and collating the speech uttered with the dictionary for the inputting person in the common dictionary (21) and the specific dictionary (22). Creation method. (2) The specific dictionary (22) is created from only the audio parameters of words spoken by the user that were unrecognizable through comparison with the common dictionary (21). A speech dictionary creation method for a specific speaker speech recognition device according to claim 1.