System operable as an automaton for recognizing continuously spoken words with reference to demi-word pair reference patterns

• US 4,513,435 A
• Filed: 04/21/1982
• Issued: 04/23/1985
• Est. Priority Date: 04/27/1981
• Status: Expired due to Term

• Current Assignee: Nippon Electric Company Limited (NEC Corporation)
• Original Assignee: Nippon Electric Company Limited (NEC Corporation)
• Inventors: Sakoe, Hiroaki, Chiba, Seibi
• Primary Examiner(s): Kemeny, E. S. Matt
• Application Number:
  US06/370,637
• Time in Patent Office: 1,098 Days
• Field of Search: 381/41-45, 382/33, 382/34, 382/37
• US Class Current: 704/255
• CPC Class Codes: G10L 15/00 Speech recognition G10L17/00 takes precedence

Abstract

A system for recognizing a continuously spoken word sequence with reference to preselected reference words with the problem of coarticulation removed, comprises a pattern memory for memorizing demi-word pair reference patterns consisting of a former and a latter reference pattern segment for each reference word and a word pair reference pattern segment for each permutation with repetition of two words selected from the preselected reference words. A recognition unit is operable as a finite-state automaton on concatenating the demi-word pair reference patterns so that no contradiction occurs at each interface of the reference patterns in every concatenation. It is possible to use the automaton in restricting the number of reference patterns in each concatenation either to an odd or an even positive integer.

Claims

1. A continuous speech recognition system for recognizing a succession of continuously spoken words with reference to a plurality of preselected reference words, each reference word having an intra-word characteristic point approximately at a point of bisection of said reference word, said system comprising:

means for memorizing a plurality of discrete demi-word pair reference patterns, each reference pattern being given by a sequence of parameters characteristic of a demi-word pair which lies between the intra-word characteristic point of a preceding reference word and the intra-word characteristic point of a succeeding reference word, said preceding and said succeeding reference words being continuously pronounced; and

means connected to said reference pattern memorizing means to be operable as a finite-state automaton to recognize said succession by using a pattern concatenation of optimum demi-word pair reference patterns which are selected from said discrete demi-word pair reference patterns and concatenated at least at an interface, said automaton being for controlling said concatenation so that no contradiction occurs at every interface.

2. A continuous speech recognition system as claimed in claim 1, wherein said reference pattern memorizing means comprises:

means for holding, at one time, parameters characteristic of at least those two of said preselected reference words which are continuously pronounced;

a reference pattern memory; and

means for operatively connecting said parameter holding means and said reference pattern memory to segmenting means operable at first to segment, with each of said preselected reference wors selected, a former and a latter reference pattern segment from a parameter sequence representative of the selected reference word at the intra-word characteristic point thereof and to store said former and said latter reference pattern segments in said reference pattern memory as those two of said discrete demi-word pair reference patterns in which the preceding and the succeeding reference words are null reference words, respectively, and thereafter operable to form, for each pair of those first and second selected ones of said preselected reference words which are successively continuously pronounced, a segment concatenation of the former and the latter reference pattern segments for the first selected reference word and the former and the latter reference pattern segments for the second selected reference word, to find a pair of intra-word characteristic points in a parameter sequence representative of the first and the second selected and successively continuously pronounced reference words with reference to said segment concatenation, to segment a word pair reference pattern segment from the last-mentioned parameter sequence at said intra-word characteristic point pair, and to store said word pair reference pattern segment in said reference pattern memory as one of said discrete demi-word pair reference patterns.

3. A continuous speech recognition system as claimed in claim 2, wherein the intra-word characteristic point is an instant which substantially bisects a duration of a parameter sequence representative of each of said preselected reference words.

4. A continuous speech recognition system as claimed in claim 2, wherein the intra-word characteristic point is an instant at which spectra of utterance of each of said preselected reference words have a maximum variation adjacently of an instant which bisects a duration of a parameter sequence representative of said each reference word.

5. A continuous speech recognition system as claimed in claim 2, wherein the intra-word characteristic point is an instant at which spectra of utterance of each of said preselected reference words have a minimum variation adjacently of an instant which bisects a duration of a parameter sequence representative of said each reference word.

6. A continuous speech recognition system as claimed in any one of claims 2 through 5, wherein said automaton has an initial state, a final state, and a plurality of intermediate states for the respective ones of said preselected reference words, the former reference pattern segment for a first of said preselected reference words giving rise to state transition from said initial state to one of said intermediate states that is for said first reference word, the latter reference pattern segment for a second of said preselected reference words giving rise to state transition to said final state from one of said intermediate states that is for said second reference word, the word pair reference pattern segment for said first and said second reference words giving rise to state transition from a former one of said intermediate states to a latter one of said intermediate states, said former and said latter intermediate states being for said first and said second reference words, respectively.

7. A continuous speech recognition system as claimed in any one of claim 2 through 5, wherein said automaton has an initial state, a final state, a first group of intermediate states, and a second group of intermediate states, the intermediate states of each group being for the respective ones of said preselected reference words, the former reference pattern segment for a first of said preselected reference words giving rise to state transition from said initial state to one of the intermediate states of said first group that is for said first reference word, the latter reference pattern segment for a second of said preselected reference words giving rise to state transition to said final state from that one of the intermediate states of a predetermined one of said first and said second groups which is for said second reference word, the word pair reference pattern segment for said first and said second reference words giving rise to state transition from a former one of the intermediate states of one of said first and said second groups to a latter one of the intermediate states of the other of said first and said second groups, said former and said latter intermediate states being for said first and said second reference words, respectively.