JPS60130798A - Voice indentifier - 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 Field of the Invention The present invention relates to speech identification, and more particularly to a speech identification device that changes an output code depending on the difference in meaning of recognized TDT speech.

Problems with the Structure of the Conventional Example In recent years, voice input methods have been attracting attention from the viewpoint of efficiency as a text input method for word processors and the like.

As shown in FIG. 1, a conventional speech recognition device consists of a speech recognition section and a speech standard pattern dictionary, and outputs a code, such as a flat name character string, corresponding to the recognized speech.

By the way, there are many characters in a text that have the same pronunciation, that is, homophones, but these are treated as exactly the same by the speech recognition unit. Therefore, it has been impossible for conventional speech recognition devices to distinguish between linguistic meanings.

OBJECTS OF THE INVENTION The present invention solves the above-mentioned conventional problems, and it is an object of the present invention to provide a speech identification device that changes the output code depending on the difference in linguistic meaning.

Structure of the Invention The present invention is a speech identification device equipped with a linguistic meaning information dictionary for distinguishing recognized speech from the viewpoint of linguistic meaning, and a meaning flag for retaining the linguistic meaning of input speech. By referring to the meaning flag that holds the linguistic meaning of the voice input immediately before, the burden of processing homophones can be reduced.

DESCRIPTION OF THE EMBODIMENTS FIG. 2 is a block diagram of a voice recognition device according to an embodiment of the present invention. In FIG. 2, 11' is a voice feature extraction unit that extracts voice feature parameters from the input voice and outputs the extracted feature parameters to the voice standard pattern determination unit;
Speech standard pattern determination that inputs feature parameters output from the 12-digit speech feature extraction section 11, judges speech by searching a speech standard pattern dictionary, and outputs a judgment result consisting of a syllable code string to a linguistic semantic information judgment section. Part 1
The syllable code string that is the judgment result of the speech outputted from the 3-key speech standard pattern judgment unit 12 is manually inputted, and reference is made to the meaning flag and linguistic meaning information dictionary that hold the linguistic meaning of the speech input immediately before. 1 is a speech feature extraction section which adds semantic information to the input syllable code string and simultaneously stores the determined linguistic meaning in a meaning flag and outputs a recognition code corresponding to the input speech. A speech recognition section 11, a speech standard pattern judgment section 12, and a linguistic semantic information judgment section 13, which recognizes input speech and outputs a recognition code; A speech standard pattern dictionary incorporating standard patterns; 3 a linguistic semantic information dictionary used to refine semantic information into the input syllable code string in the linguistic semantic information determining section 13; 4 a linguistic semantic information dictionary used in the linguistic semantic information determining section 13; This is a meaning flag that holds the linguistic meaning of the syllable code string of the input speech in order to select the optimal linguistic meaning from a plurality of pieces of linguistic meaning information for the syllable code string.

The operation of the voice recognition device of this embodiment configured as described above will be described below.

Sentence 1'-Press the REQUEST button 19 times' in voice / R / E / Cue / Knee / E / Varnish / T /
Let's consider the case of inputting ``bo/ri/n/o/ichi/cue/force/i/o/su/. /R/ is voice recognition 2 sum 1
is input, it is called an alphabet by referring to the linguistic meaning information dictionary 3! If you have a word meaning, you can understand good fortune.

As a result, the speech recognition unit 1 outputs the character rRJ, and causes the semantic flag 4 to hold the semantic information "alpha vent". Similarly, when /E/ is input, the character "E"
At the same time, the semantic flag 4 is made to hold the semantic information of the alphabet.

Now, when /cue/ is input next, by referring to the linguistic meaning information dictionary 3, it is found that it has two linguistic meanings of an alpha bent number, and the output code is not determined. At this time, by referring to the meaning flag 4, since the voice input just before was an alphabet, it is determined that the human input /cue/ is also an alphabet, and the character rQJ is output.

Similarly, for the /cue/ of "...19 times...", semantic flag 4 indicates that the previous input /ichi/ had the linguistic semantic information of a number, so the character "9" is output.

It is known from the statistical conclusions of natural language information that there is a fact that ``alphabets and numbers appear consecutively more often than they appear randomly alone'' in sentences. , the principle of Honkomei is based on this fact.

As described above, according to this embodiment, by being able to retain the linguistic semantic information of the voice input immediately before,
The homophones "Q" and "9" can be easily distinguished.

Note that in the above embodiment, the distinction between alphabento and numbers was taken as an example, but it goes without saying that various other semantic groupings can be considered from a linguistic perspective. In particular, by categorizing the text into groups appropriate to the content of the text to be input by voice, the burden of processing homophones can be further reduced.

In addition, in the above example, we mainly focused on monosyllable input, and took up the case where numbers and alpha vents were input as consecutive syllables.
Since the essence lies in the linguistic semantic grouping of the dictionary, the present invention is effective even with continuous voice input. For example, by dividing "government" into a commercial group and "positive/negative" into a technical group, whether the text input to the present invention is commercial text or technical data, the homonyms of /safe/ It can be expected that the post-processing burden on the speaker will be reduced.

In this example, the speech standard Bakun dictionary 2 and the linguistic semantic information dictionary 3 are explained separately, but it is natural to create a linguistic semantic information dictionary that separates the speech standard patterns from the viewpoint of linguistic meaning. However, it is not necessary to have a separate speech standard pattern dictionary.

Effects of the Invention The present invention makes it possible to easily process homophones according to the context by referring to the linguistic meaning flags of human speech, improving the accuracy rate in that context and improving the processing speed. The number of steps is reduced, speeding up, and ease of operation can be achieved.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a conventional voice recognition device, and FIG. 2 is a block diagram of an embodiment of the present invention. 1...Speech recognition unit, 2...Speech standard pattern dictionary, 3...Linguistic meaning information dictionary, 4.
... Meaning flag, 11 ... Voice feature extraction section, 12 ... Voice standard pattern determination section, 13.
...Linguistic semantic information determination unit.

Claims (2)

[Claims] (1) A speech recognition device characterized by comprising speech recognition means, a linguistic meaning information dictionary for distinguishing recognized speech from the viewpoint of linguistic meaning, and a meaning flag for retaining linguistic meaning. . (2) The speech identification device according to claim 1, wherein the speech recognition means includes a speech feature extraction section, a speech standard pattern judgment section, and a linguistic meaning information judgment section.