JPH02124599A - Continuously voiced speech recognition system - Google Patents

Abstract

PURPOSE:To improve recognition processing efficiency by outputting a correction signal from a voice power calculating means when a misrecognized word is included at the time of voice recognition, correcting the word by a voice recognition means which receives the correction signal according to a word dictionary, and outputting a correct recognition result to a recognition result output means. CONSTITUTION:When the voice recognition means 3 misrecognizes one word and a recognition result containing the misrecognized word is outputted from a recognition result output means 4, an operator inputs a voice from a voice input means 2 again so that the voice power of the word to be corrected is larger than the voice power of other uncorrected words and exceeds a constant value. Then the voice power calculating means 5 outputs the correction signal and the voice recognition means 3 extracts the word after correction out of the word dictionary 1 and recognizes it. Therefore, the correct recognition result containing the corrected word is outputted from the recognition result output means 4. Consequently, the misrecognition rate in correction input is reduced and the efficiency of the recognition processing is improved.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a continuous utterance recognition method for performing partial correction when erroneously recognizing continuous utterance in a speech recognition device that recognizes continuous utterance. .

[Prior Art] FIG. 4 is a block diagram showing the configuration of a continuous speech recognition device that employs a conventional continuous speech recognition method. In the figure,
1 is a word dictionary in which words are registered; 2 is a voice input means realized by a microphone for inputting voice; 3 is a voice recognition means for recognizing the voice input from the voice input means 2 based on the word dictionary 1; 4 is a recognition result output means for outputting the result recognized by the voice recognition means 3. The recognition result output means 4 is realized by an earphone, a speaker, a CRT, etc., but here, it is assumed that the earphone or the like is used to output audio.

Next, the conventional method will be explained with reference to the flowchart shown in FIG. Continuously uttered voices from the operator are input to the voice input means 2 and converted into electrical signals (step SL). The speech recognition means 3 converts the electric signal into a digital signal, extracts the corresponding word from the word dictionary 1, and recognizes the input speech (step S2). After this recognition process, in order to have the operator confirm the recognition result, the voice recognition means 3 provides the recognition result to the recognition result output means 4, and continuously informs the operator of the recognition result by voice (step S3). That is, the recognition result output means 4 outputs the voice as a result of recognizing the input continuous voice, but if the recognition result is correct (step S4), the process ends and the input continuous voice and the recognition result voice are output. If there is even a partial difference, it means that the voice recognition means 3 has misrecognized the voice. Therefore, when the above-mentioned erroneous recognition occurs, the operator performs a process of inputting the continuous voice again.

[Problem to be solved by the invention] Since the conventional continuous speech recognition method processes as described above, if even one of N consecutive words is misrecognized, the entire recognition process is redone and the misrecognition is eliminated. It is necessary to repeat the input process of continuous utterances until all the utterances are exhausted, which poses a problem of reducing the efficiency of the recognition process.

This invention was made to solve the above-mentioned problems, and when there is a misrecognition, by redoing the recognition process only for the misrecognized word, the misrecognition rate at the time of corrected input is reduced, and the recognition process is improved. The purpose of the present invention is to provide a continuous utterance speech recognition method that can improve processing efficiency.

[Means for Solving 8 Problems] The continuous utterance speech recognition method according to the present invention calculates the power of the speech input from the speech input means 2, and calculates the speech power difference between the word to be corrected and the word not to be corrected to be constant. It is equipped with a voice power calculation means 5 that outputs a correction signal when the value exceeds the value, and when a word that is incorrectly recognized during speech recognition is included, in order to correct that word, the voice power of the word to be corrected is calculated by other words that are not corrected. The voice input means 2 inputs the voice again so that the voice power is higher than the voice power of the word and exceeds the above-mentioned fixed value, the voice power calculation means 5 outputs a correction signal, and the voice recognition means 3 which receives this correction signal recognizes the word. It is characterized in that it is corrected according to the word dictionary 1 and outputs the correct recognition result to the recognition result output means 4.

[Operation] In this continuous utterance speech recognition method, when the speech recognition means 3 misrecognizes a certain word and the recognition result output means 4 outputs a recognition result including the misrecognized word, the operator corrects the word. To do this, perform the following processing. That is,
Speech is input again from the speech input means 2 so that the speech power of the word to be corrected is greater than the speech power of other words not to be corrected and exceeds a certain value. As a result, the speech power calculation means 5 outputs a correction signal, and the speech recognition means 3 extracts the corrected word from the word dictionary 1 and recognizes it. Therefore, the recognition result output means 4 outputs a correct recognition result including the corrected word.

[Embodiment of the Invention] FIG. 1 is a block diagram showing the configuration of a continuous utterance speech recognition apparatus employing a continuous utterance speech recognition method according to an embodiment of the present invention. In FIG. 1, components corresponding to those shown in FIG. 4 are designated by the same reference numerals, and their explanations will be omitted. In FIG. 1, 5 calculates the power of the voice input from the voice input means 2, and corrects it if the voice power difference (the difference between the voice power of the word to be corrected and the voice power of the word not to be corrected) exceeds a certain value. This is an audio power calculation means that outputs a signal.

Next, the operation of this embodiment will be explained with reference to flowchart 1 shown in FIG. The voice input means 2 inputs continuous voice from the operator and converts it into an electrical signal (
Step Nl>. The speech recognition means 3 converts the electric signal into a digital signal, extracts the corresponding word from the word dictionary 1, and recognizes the input speech (step N2). on the other hand,
The voice power calculation means 5 calculates the voice power corresponding to each word of the continuous utterance inputted in step N1 (step N3). In order to let the operator confirm the recognition result of the voice recognition means 3 by voice response etc., the recognition result output means 4 is used.
continuously outputs the recognition results (step N4). However, the recognition result is incorrect (step N5), and when the recognition result includes the incorrectly recognized word, the operator re-inputs the continuous word utterance audio (step N5).
6). In other words, when re-entering, the operator increases only the incorrectly recognized word (word to be corrected) (so as to exceed the above-mentioned certain value).
Other words (words that are not corrected) are uttered at the same loudness as when they were first input. This uttered voice is input to the voice recognition means 3 in the same manner as above, and is subjected to recognition processing (
Step N7). Further, this uttered voice is input to the voice power calculation means 5, and the voice power is calculated (step N8). That is, the voice power calculation means 5 extracts words whose input value exceeds a certain input value from the word dictionary 1, recognizes them, and replaces the initially erroneously recognized word with a correct word (step N9). Thereafter, the process returns to step N4, where the recognition results are continuously output, and the operator checks again, and if the recognition results are correct (step N5), this process ends.

FIGS. 3(1) to 3(6) are signal waveform diagrams for explaining the operation of this embodiment. FIG. 3(1) shows the first voice utterance signals 2a, 5a, la, 3a, and FIG. 3(2) shows the first recognition results 2a, 5a, 8a, 3a 2a, 5a,
la, 3a, and the audio power of signal 1a is expressed as the other signal 2.
It is larger than that of a, 5a, and 3a. Figure 3 (
4) shows provisional recognition results 2a, 5a, la, and 4a at the time of correction. In this recognition result, 1a has been corrected, but 4a has been misrecognized. However, the first recognition result correctly recognized it as 3a and it was set to voice recognition means 3, so
There is no change from 3a to 4a. FIG. 3(5) shows the audio power at the time of initial input (broken line) and the audio power at the time of corrected input (solid line). FIG. 3(6) shows recognition results 2a, 5a, la, and 3a at the time of correction.

Even if the recognition result includes a word that was initially misrecognized in this way, the correct recognition result can be obtained by increasing the voice power of the word to be corrected and inputting it.

[Effects of the Invention] As described above, according to the present invention, when a word that is incorrectly recognized during speech recognition is included, in order to correct the word,
Input the voice again from the voice input means so that the voice power of the word to be corrected is larger than the voice power of other words to be corrected and exceeds a fixed value, output a correction signal from the voice power calculation means, and receive the correction signal. Since the words are corrected by the voice recognition means according to the word dictionary and the correct recognition result is output to the recognition result output means, if there is an erroneous recognition, only the erroneously recognized word is recognized and corrected. The effect is that the rate of recognition errors during input is reduced, partial correction can be performed with high accuracy, flexibility and recognition processing efficiency can be improved.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of the configuration of a continuous speech recognition device that employs a continuous speech recognition system according to an embodiment of the present invention, and FIG. 2 is a flowchart 1 for explaining the operation of this embodiment. , Fig. 3 (1) to (6) are signal waveform diagrams for explaining the operation of this embodiment, and Fig. 4 is a structural block diagram of a continuous utterance speech recognition device employing a conventional continuous utterance speech recognition method. Figure 5I2I is a flowchart 1 for explaining the operation of this conventional example. 1... Word dictionary, 2... Voice input means, 3... Voice recognition means, 4... Recognition result output means, 5...... Voice power calculation means.

Claims (1)

[Claims] A word dictionary in which words are registered, a voice input means for inputting voice, a voice recognition means for recognizing the voice input from the voice input means based on the word dictionary, and a result recognized by the voice recognition means. A continuous speech recognition device that recognizes continuously uttered speech includes a recognition result output means that outputs a recognition result output means, which calculates the power of the speech input from the speech input means and distinguishes between words to be corrected and words not to be corrected. A voice power calculation means that outputs a correction signal when the voice power difference exceeds a certain value is provided, and if a word that is incorrectly recognized during speech recognition is included, the voice power of the word to be corrected is adjusted to correct the word. The speech input means again inputs speech so that the speech power is greater than the speech power of other words to be corrected and exceeds the certain value, the speech power calculation means outputs a correction signal, and the speech recognition receives this correction signal. A continuous utterance speech recognition method, characterized in that the means corrects words according to the word dictionary and outputs correct recognition results to the recognition result output means.