JPH0421899A - Voice recognizing device - Google Patents

Abstract

PURPOSE:To quickly access a word purposed to express by a speaker by sorting plural word canditates or sentence candidates in each sorting item and outputting and displaying the sorted result. CONSTITUTION:A voice analyzing circuit 1 inputs an input voice and outputs a feature parameter time sequence expressing voice features to a pattern collating circuit 2. The circuit 2 collates the feature parameter time sequence with a parameter time sequence supplied from a reference pattern storing circuit 6 and utilized as a reference pattern in each prescribed word group and outputs the collated result to a candidate selecting circuit 3. The circuit 3 selects plural word candidates having higher collated results and supplies the selected candidates to a candidate sorting circuit 4, which collates sorting items stored in a sorting item information storing circuit 7, sorts the inputted word candidates to the sorting items. A picture display device 5 displays the word candidates in respective sorting items.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a speech recognition device that recognizes word speech or sentence speech produced by humans.

BACKGROUND OF THE INVENTION Conventionally, there have been various techniques for realizing a device that recognizes human-generated speech.

An example is a method based on matching acoustic parameter patterns. Hereinafter, as a conventional technique, a method based on acoustic parameter pattern matching will be described in the case of word recognition.

FIG. 3 shows a typical configuration example of a conventional speech recognition system.

In FIG. 3, the input voice is input to the voice analysis port fI8Ll. The speech analysis circuit 11 analyzes the input speech and outputs a time series of feature parameters representing the characteristics of the speech to the pattern matching circuit 12.

The pattern matching circuit 12 includes a standard pattern storage circuit 16.
Also provides parameter time series.

The standard pattern storage circuit 16 stores a parameter time series prepared for each word used as a standard pattern. The pattern matching circuit 12 is the voice analysis circuit 11
The characteristic parameter time series obtained in step 1 and the standard pattern parameter time series supplied from the standard pattern storage circuit 16 are compared for a predetermined group of words, and the results of the comparison are outputted to the recognition result determination circuit 13 for each word. . The recognition result determination circuit 13 inputs the results of the matching performed by the pattern matching circuit 12, determines the word that gives the best matching result, and outputs the word to the outside as the recognition result of the speech recognition device.

Problems to be Solved by the Invention In conventional speech recognition devices, the time-series parameters of each input pattern and standard pattern are compared, and the word that gives the best matching result is obtained as a recognition result. However, in the case of speech recognition, so-called recognition errors, in which a recognition result different from the content intended by the speaker is obtained, occur with a certain frequency. If this recognition error occurs, it may be necessary to take measures such as re-entering the same word, but with the method of obtaining only one recognition result as shown in the conventional example in Figure 3, recognition errors occur. In some cases, there is a problem in terms of operability in that it takes time and effort to access the word that the speaker intends.

In view of the problems described above, the present invention provides the following:
It is an object of the present invention to provide a speech recognition device that can access words intended by a speaker in a short time.

Means for Solving the Problem In order to achieve this object, the present invention provides a speech recognition means for outputting a plurality of word candidates or sentence candidates as recognition results, and a speech recognition means for outputting a plurality of word candidates or sentence candidates as recognition results, and A candidate classification means for classifying word candidates or sentence candidates according to classification items, and a screen display device for outputting and displaying a plurality of word candidates or sentence candidates according to classification items based on the classification results by the candidate classification means. It is configured.

111 With the above configuration, the present invention displays a plurality of recognition results as candidates for each classification item, and if a correct word is included in the plurality of candidates, the correct word is selected from among the displayed candidates. By selecting a word, it is possible to access one of the words intended by the speaker with a single utterance.

Example The present invention will be explained below with reference to Examples.

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention.

In FIG. 1, human voice is input to a voice analysis circuit 1. In FIG. A speech analysis circuit 1 analyzes human speech and outputs a time series of feature parameters representing speech characteristics to a pattern matching circuit 2.

The pattern matching circuit 2 is also supplied with parameter time series from the standard pattern storage circuit 6. The standard pattern storage circuit 6 stores a parameter time series prepared for each word used as a standard pattern. The pattern matching circuit 2 matches the feature parameter time series obtained by the speech analysis circuit 1 and the parameter time series used as a standard pattern supplied from the standard pattern storage circuit 6 for a predetermined word group, and performs matching. The results are output to the candidate selection circuit 3 for each word.

The candidate selection circuit 3 inputs the matching results of the pattern matching circuit 2, selects a plurality of word candidates based on the best matching results, and supplies them to the candidate classification circuit 4.

The candidate classification circuit 4 receives the word candidates supplied from the candidate selection circuit 3, and classifies the word candidates by classification item by comparing the word candidates with the classification items stored in the classification item information storage circuit 7. The contents classified by the candidate classification circuit 4 are input to the screen display device 5 at the final stage, and are displayed for each classification item.

FIG. 2 shows an example of a screen display on the screen display device 5.

In addition, in this example, if the word ``-invention'', 1- is ``dan''.
, for the word ``speech'', the word is displayed as classified by setting the first syllable of the 50 sounds to which the first syllable of the word belongs as a standard for word classification, such as ``a-dan''.

FIG. 2 shows an example in which recognition result candidates are displayed on the screen display device 5 when the input voice "invention" is recognized. In this example, the word "-shitsurei" is recognized as the first candidate, and the correct answer "1 is Tsumei" is recognized as the second candidate.

That is, the first-place candidate is displayed with a double underline, and the second-place candidate is displayed with an underline.

In the conventional speech recognition device shown in Fig. 3, only the first result is obtained, so if a recognition error occurs, it is necessary to re-speak the same word. With the voice recognition device of the present invention shown in FIG. Alternatively, by selecting the desired word using an input means such as a mouse, the user can access the desired word in a short time, thereby improving operability.

Furthermore, when speech recognition is used as an input for information retrieval, etc., a word constant that exceeds the limit of actual speech recognition performance may be required. Even in such a case, according to the speech recognition device of the present invention, the probability that the obtained candidates include the word intended by the speaker is lower than the conventional method of displaying singular recognition results. Since the cost is higher than that of conventional methods, it is possible to set the number of words to be handled larger than in the past.

DETAILED DESCRIPTION OF THE INVENTION According to the present invention, there is provided a speech recognition means for outputting a plurality of word candidates or sentence candidates as recognition results, and a method for classifying the plurality of word candidates or sentence candidates outputted by the speech recognition means. The present invention is configured to include a candidate classification means for classifying by item, and a screen display device for outputting and displaying a plurality of word candidates or sentence candidates for each classification item based on the results of classification by the candidate classification means.

With this configuration, by displaying multiple recognition results as candidates for each classification item, if the correct word is included in the multiple candidates, the correct word can be selected from the displayed candidates. It becomes possible to access the words intended by the speaker in a short time.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a speech recognition device in an embodiment of the present invention, FIG. 2 is a front view of a screen display device for explaining the operation of the speech recognition device in an embodiment of the invention, and FIG. 3 1 is a block diagram showing a conventional speech recognition device. 1...Speech analysis circuit, 2...Pattern matching circuit,
3. Candidate selection circuit, 4. Candidate classification circuit, 5. Screen display device, 6. Standard pattern storage circuit, 7. Classification item information storage circuit.

Claims (1)

[Claims] a speech recognition means for outputting a plurality of word candidates or sentence candidates as recognition results; a candidate classification means for classifying the plurality of word candidates or sentence candidates outputted by the speech recognition means according to classification items; A speech recognition device comprising: a screen display device that outputs and displays the plurality of word candidates or sentence candidates for each of the classification items based on the classification results.