JPH10116093A - Voice recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide the voice recognition device which always recognizes voices even through the device is used by the person whose mother language is other than the language that is recognized by the device. SOLUTION: When voices are inputted from the external through a voice inputting section 1, a Japanese voice recognition processing section 2 obtains the recognition likelihood in terms of Japanese against the inputted data and moreover, an English voice recognition processing section 3 obtains the recognition likelihood in terms of English against the inputted voice data. Then, a recognition likelihood comparison processing section 5 compares the respective recognition likelihoods obtained by the sections 2 and 3 and the language, which has a largest recognition likelihood, is outputted as the recognition result.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition device, and more particularly to a speech recognition device capable of recognizing even a person who does not speak a recognizable language as its native language.

[0002]

2. Description of the Related Art Conventionally, various techniques have been proposed for speech recognition, and various devices have been developed which can recognize input speech at a certain recognition rate.

In a conventional speech recognition apparatus, a speech model of a recognizable language is stored in advance, and when a speech to be recognized is input, the input speech is compared with the speech model. Voice recognition is being performed.

As an example of a conventional speech recognition apparatus, Japanese Patent Laid-Open No. 6-27988 discloses an HMM (Hidden MMM) of a keyword to be recognized when a speech to be recognized is input.
A speech recognition apparatus that obtains likelihood with respect to an arkov model and thereby performs speech recognition is disclosed.

[0005]

In the above-described conventional speech recognition apparatus, a speech model is generated using a basic pattern of a person whose native language is a recognizable language. Other than the native language,
In addition, when a person whose recognizable language pronunciation is not close to the basic pattern uses the device, there is a possibility that his / her voice may not be recognized.

For example, in a speech recognition apparatus for recognizing English, since a speech model is generated based on the utterance of a person whose native language is English, a Japanese who does not speak English uses the apparatus. If you do, some of the Japanese have pronunciations that are similar to those whose native English is English, or have pronunciations that are similar to Japanese (Katakana foreign words). There is a possibility that the voice of a person whose pronunciation is close to "" may not be recognized.

[0007] The above-mentioned problem occurs, for example, in a cockpit. In the cockpit, there are people who have various pronunciations, and some of the pronunciations of English are similar to those whose native language is English, and some of them are similar to Japanese (Katakana foreign language). It is.

Here, in the problem caused by individual differences described above, an English voice recognition method is used for a person whose pronunciation in English is close to a person whose native language is English, and a pronunciation similar to Japanese (Katakana foreign language) is used. ) Can be solved by using the Japanese language recognition method, but the switching is not performed automatically, so check in advance to which voice the person belongs, and A switch must be made.

[0009] Also, in English utterances by Japanese,
Even for the same person, the pronunciation is not always the same, but the pronunciation may be English or Japanese, so the setting was switched for each person who uses the device However, there is a possibility that a part that is not recognized may occur.

[0010] The present invention has been made in view of the above-described problems of the conventional technology, and is intended to always provide voice even when a person whose native language is a language other than a recognizable language is used. It is an object of the present invention to provide a voice recognition device capable of recognizing a speech.

[0011]

In order to achieve the above object, the present invention recognizes an input speech as a first language and a second language, and recognizes a language having the highest likelihood of recognition. A speech recognition device for outputting a result, wherein a speech is inputted, and a speech input unit for encoding the inputted speech, and recognition of the data encoded in the speech input unit as the first language. A first speech recognition processing unit for finding a likelihood, and a second speech recognition unit for finding a recognition likelihood as the second language for the data encoded in the speech input unit.
A speech recognition processing unit, a recognition likelihood normalization processing unit that normalizes the recognition likelihood obtained in the first and second speech recognition processing units, and a recognition likelihood normalization processing unit A recognition likelihood comparison processing unit that performs a comparison operation of the recognized likelihoods,
A recognition result output unit that outputs a language having the highest recognition likelihood as a recognition result by the comparison operation in the recognition likelihood comparison processing unit.

Further, the first and second speech recognition processing units are characterized in that a speech model for comparison with data inputted for obtaining the recognition likelihood is stored in advance.

[0013] The first language is Japanese and the second language is English.

(Operation) In the present invention configured as described above, when a voice is input from the outside via the voice input unit, the input is performed in each of the first and second voice recognition processing units. Recognition likelihoods as first and second languages are obtained for the speech data, and thereafter, in a recognition likelihood comparison processing unit, the recognition likelihoods obtained in the first and second speech recognition processing units, respectively. Are compared, and as a result, the language with the highest recognition likelihood is recognized as the recognition result by the recognition result output unit 6.
Is output via.

In this way, the recognition likelihood as a plurality of languages is obtained for the input speech, and the language having the largest recognition likelihood is output as a recognition result. Even when the second language is input by a person who is a national language, there is no possibility that an unrecognized part may occur due to a difference in pronunciation.

If the first language is Japanese and the second language is English, the input voice is similar to that of a person whose native language is English. However, even if the pronunciation is close to Japanese (Katakana foreign words), there is no possibility that the speech will not be recognized.

[0017]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of a speech recognition apparatus according to the present invention, which is a speech recognition apparatus for Japanese whose recognition target is English.

In this embodiment, as shown in FIG. 1, a voice is inputted, a voice input unit 1 for encoding the input voice, and data encoded by the voice input unit 1 are recognized as Japanese. Japanese speech recognition processing unit 2 for obtaining likelihood, English speech recognition processing unit 3 for obtaining the recognition likelihood of data encoded in speech input unit 1 as English, Japanese speech recognition processing unit 2 and Comparison between the recognition likelihood normalization processing unit 4 that normalizes the recognition likelihood obtained by the English speech recognition processing unit 3 and the recognition likelihood of Japanese and English normalized by the recognition likelihood normalization processing unit 4 It comprises a recognition likelihood comparison processing unit 5 for calculating, and a recognition result output unit 6 for outputting, as a recognition result, a language having the highest recognition likelihood by the comparison operation in the recognition likelihood comparison processing unit 5.

Hereinafter, the operation of the speech recognition apparatus configured as described above will be described.

First, an external microphone (not shown)
When a voice signal (input voice) is input to the voice input unit 1 via the voice input unit 1, the input voice signal is converted into a digital signal in the voice input unit 1, and the voice recognition unit 2 converts the input voice signal into voice code data. Output to the English speech recognition processing unit 3.

Here, the Japanese speech recognition processing unit 2 stores in advance a speech model for each syllable.
In the English speech recognition processing unit 3, the Dip
A phone model is stored for each phone.

When the speech code data outputted from the speech input unit 1 is inputted to the Japanese speech recognition processing unit 2, the Japanese speech recognition processing unit 2 outputs the speech code data from the speech input unit 1 using a half-syllable HMM. A matching process is performed between the input speech code data and the stored speech model in units of syllables, and the word having the maximum similarity and the Japanese recognition likelihood at that time are calculated.

When the voice code data output from the voice input unit 1 is input to the English voice recognition processing unit 3, the English voice recognition processing unit 3 outputs the voice code data from the voice input unit 1 using the Diphone HMM model. A matching process is performed between the obtained speech code data and the stored speech model in units of Diphones, and the word having the maximum similarity and the English recognition likelihood at that time are calculated.

Here, the recognition likelihood of the Japanese recognition likelihood calculated by the Japanese speech recognition processing unit 2 and the English recognition likelihood recognized by the English speech recognition processing 3 are different from each other. In this state, the recognition likelihood cannot be compared.

When the Japanese speech recognition likelihood and the English speech recognition likelihood are calculated in each of the Japanese speech recognition processing section 2 and the English speech recognition processing section 3, the Japanese recognition likelihood and the English recognition likelihood are recognized. It is input to the likelihood normalization processing unit 4 and normalized in the recognition likelihood normalization processing unit 4 so that the Japanese recognition likelihood and the English recognition likelihood can be compared with each other.

When the Japanese recognition likelihood and the English recognition likelihood are normalized in the recognition likelihood normalization processing unit 4, the recognition likelihood comparison processing unit 5 compares the two normalized recognition likelihoods. An operation is performed.

Thereafter, in the recognition result output unit 6, based on the result of the comparison operation in the recognition likelihood comparison processing unit 5,
The word with the higher recognition likelihood is output as the recognition result.

In the above-described embodiment, as the speech models stored in advance, a half-syllable HMM in the Japanese speech recognition processing unit 2 and a D
Although the case where the iPhone HMM is used has been described,
The present invention is not limited to this, and it is also possible to perform each recognition using another model.

Also, in the present embodiment, the case where both the Japanese speech recognition processing and the English speech recognition processing are used has been described. However, the present invention is not limited to this, and other plural speech recognition processing may be performed depending on the language to be recognized. It is also possible to use the speech recognition processing of the language of FIG.

[0031]

As described above, in the present invention,
When voice is input from outside via the voice input unit, the first
In each of the first and second speech recognition processing units, the recognition likelihood as the first and second languages is obtained from the input speech data. 2 is compared with the recognition likelihood obtained by the speech recognition processing unit 2, and as a result, a language having the largest recognition likelihood is output as a recognition result via the recognition result output unit. Thus, a high recognition rate can be obtained without switching the recognition method.

For example, when a voice of English generation by a Japanese is input, if the first language is Japanese and the second language is English, the input voice is to a person whose native language is English. A high recognition rate can be obtained whether the pronunciation is close or the pronunciation is close to Japanese (Katakana foreign words).

[Brief description of the drawings]

FIG. 1 is a block diagram showing one embodiment of a speech recognition device of the present invention.

[Explanation of symbols]

 Reference Signs List 1 speech input unit 2 Japanese speech recognition processing unit 3 English speech recognition processing unit 4 recognition likelihood normalization processing unit 5 recognition likelihood comparison processing unit 6 recognition result output unit

Claims (3)

[Claims] 1. A speech recognition apparatus for recognizing an input speech as a first language and a second language, and outputting a language having the highest likelihood of recognition as a recognition result. A speech input unit that encodes input speech, a first speech recognition processing unit that determines a recognition likelihood as the first language for data encoded by the speech input unit, A second speech recognition processing unit that obtains a recognition likelihood as the second language for the data encoded in the input unit; and a recognition likelihood obtained by the first and second speech recognition processing units. A recognition likelihood normalization processing unit for normalizing; a recognition likelihood comparison processing unit for comparing and calculating the recognition likelihood normalized in the recognition likelihood normalization processing unit; and a comparison calculation in the recognition likelihood comparison processing unit Words with the highest recognition likelihood A recognition result output unit that outputs a word as a recognition result. 2. The speech recognition apparatus according to claim 1, wherein the first and second speech recognition processing units are configured to compare a speech model with data input in advance to obtain the recognition likelihood. Is stored. 3. The speech recognition device according to claim 1, wherein the first language is Japanese and the second language is English.