JPS58159594A - Polysyllabic word recognition system - 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 (a) Technical Field of the Invention The present invention relates to a speech recognition system, in particular, to recognize multi-syllabic words and mono-syllabic words without the need for mode switching, etc. This paper relates to a multi-syllable word recognition method with short recognition time.

(b) Technical Background A speech recognition system that primarily recognizes monosyllabic words (in this case, if it is possible to recognize multi-syllabic words as well, the new line 1
It is now possible to simultaneously input special symbols such as periods, commas, (,), etc., and control terms such as save, print, etc., eliminating the need for mode switching keys such as word mode, kana mode, etc., and simplifying the configuration of the speech recognition system. It is simpler, has improved operability, and has the advantage of a single L7). Therefore, it is desired to develop a speech recognition system that can recognize multi-syllable words at a high speed.

(C) Purpose of the Invention Based on the above-mentioned requirements, the purpose of the present invention is to recognize all monosyllabic words and multi-syllabic words without the need for mode switching.
There are a number of speech recognition systems that can quickly recognize multi-syllable words.

(d) Structure of the Invention The structure of the present invention corresponds to different syllables in a speech signal. Method for extracting syllable representative points using stationarity The gold film extracts sample points that indicate the syllables that correspond to the vowels of a word, and if there are multiple sample points, it is determined that it is a multi-syllable word and it is a multi-syllable word. If the sample point is singular, it is recognized by a recognition circuit dedicated to monosyllabic words.

The present invention is based on the fact that there are sections in which the characteristic parameters of an audio signal change rapidly over time and sections in which they change gradually over time.
Note that one feature parameter is smooth in time.
Multi-syllable words are selected by utilizing the fact that the points at which these changes, that is, the points at which the characteristic sound of the voice has constancy, are syllables corresponding to the vowels of the words. Therefore, it is necessary to extract a sample point representative of the syllable, and the means for extracting the sample point is described in detail in Japanese Patent Laid-Open No. 145408/1983. To briefly explain, unconstrained sampling is performed using all the accumulated variation of feature parameters, and the results omitted in the non-uniform sampling are associated with representative weights, and the higher the weight, the more stationary the interval This corresponds to the position of the vowel.

(e) Examples of the invention The figure is a block diagram of a circuit showing one embodiment of the present invention.

First, the speaker connects the switching unit 3 from the control unit 11 to the monosyllabic word recognition unit 9 and the multi-syllable recognition unit 10ζ in order to register voice sounds in advance, and inputs monosyllabic words and specific multi-syllable words. Add. The preprocessing unit 1 performs all audio level adjustment, analog-to-digital conversion, etc., and sends it to parameter extraction 'm2. The variable extraction unit 2 extracts all feature parameters of the word, and then passes through the switching unit 3 to the monosyllabic word recognition unit 91, which then extracts all the feature parameters of the monosyllabic word. Store parameters. The next speaker selects the switching unit 3 under the control of the control unit 11 in order to perform speech recognition.
It is connected to the total sampling time determining section 4 and the storage section 7, and utters words. The audio signal of the input word passes through the preprocessing section 1, parameter extraction section 2, and switching section 3 in the same manner as described above.
The sampling time determination section 4 determines nonuniform sampling times, and the syllable representative point extraction section 5 enters the sampling time determination section 4 . The syllable representative point extraction unit 5 performs the cumulative number of times 1! while the non-uniform sampling points are being determined. (The count value of the non-uniform sampling points is kept as a weight, and a predetermined number of sample points are determined from the non-uniform sampling points with the largest weight. In other words, the sample points corresponding to the syllable t) The vowel number detection section 6 counts the number of sounds at the sample point, and if the number of vowels is more than one, it is determined that it is a multi-syllable word, controls the selection section 8, and selects the single Wr+ stored in the storage section 7ζ. ' is sent to the multi-syllable word recognition unit 10, and the recognition result is sent to the control unit 11.
The signal is then sent to output 4-. If the number of parent vowels is 1, it is determined to be a monosyllabic word, and the selection unit 8 is controlled to send the word in the storage unit 7 to the monosyllabic word recognition unit 9, and the recognition result is outputted via the control unit 11. .

(f) As described in detail, the present invention provides an audio +4! Utilizing the small amount of constancy, we can detect vowel parts that correspond to different syllables in the audio signal, then count the number of vowels to determine whether it is a monosyllabic word or a multi-syllabic word, and calculate the unknown input word. It is possible to provide an entire speech recognition system that can recognize multi-syllabic words without the need for mode switching for selection.

[Brief explanation of drawings]

FIG. 3 is a block diagram of a circuit showing an embodiment of the present invention. 2 is a preprocessing section, 2 is a parameter extraction section, 4 is a sampling time determining section, 5 is a syllable representative point extraction section, 6 is a vowel number determining section, 8 is a selection section, and 1] is a control section.

Claims (1)

[Claims] 1) In a speech recognition system that recognizes multi-syllable words and monosyllabic words, it takes full advantage of the constancy of the characteristic sounds of speech to recognize different syllables in a speech color code. A multi-syllable word recognition method, characterized in that a means for detecting all corresponding vowel parts is provided, and an unknown input word is recognized as a multi-syllable word when a plurality of vowel parts are detected by the detection means. 2) The means for detecting the vowel part sequentially calculates and accumulates the cumulative variation amount of the characteristic parameter of the audio signal, and determines a non-uniform sampling point where the cumulative variation amount reaches a predetermined threshold l, Using sample points extracted from non-uniform sampling points extracted in descending order of weight corresponding to the cumulative number l of determining the non-uniform sampling point for each non-uniform sampling point, the sampling point is Corresponding to the extracted non-uniform sampling points in descending order of magnitude, other non-uniform sampling points within a predetermined neighborhood range of the extracted non-uniform sampling points are excluded from the above extraction process, and then A multi-syllable word recognition method according to claim 1F, characterized in that the non-uniform sampling points with large weights are selected.