Speech recognition method and system

• US 6,631,349 B1
• Filed: 05/09/2000
• Issued: 10/07/2003
• Est. Priority Date: 11/27/1997
• Status: Expired due to Fees

• Current Assignee: Renesas Electronics Corporation
• Original Assignee: Hitachi, Ltd.
• Inventors: Ishiwatari, Kazuyoshi, Kondo, Kazuo, Wakisaka, Shinji
• Primary Examiner(s): ABEBE, DANIEL DEMELASH
• Application Number:
  US09/554,003
• Time in Patent Office: 1,246 Days
• Field of Search: 704/255, 704/239, 704/238, 704/240, 704/232, 704/245, 704/252, 704/256, 704/251
• US Class Current: 704/239
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units

Abstract

Frames making up an input speech are each collated with a string of phonemes representing speech candidates to be recognized, whereby evaluation values regarding the phonemes are computed. The frames are each compared with part of the phoneme string so as to reduce computations and memory capacity required in recognizing the input speech based on the evaluation values. That is, each frame is compared with a portion of the phoneme string to acquire an evaluation value for each phoneme. If the acquired evaluation value meets a predetermined condition, part of the phonemes to be collated with the next frame are changed. Illustratively, if the evaluation value for the phoneme heading a given portion of collated phonemes is smaller than the evaluation value of the phoneme which terminates that phoneme portion, then the head phoneme is replaced by the next phoneme. The new portion of phonemes obtained by the replacement is used for collation with the next frame.

Claims

1. A speech recognition method for collating a target speech with each of a plurality of speech candidates in order to recognize said target speech, said speech recognition method comprising the steps of:

(a) collating one of said plurality of speech candidates successively with an ordered plurality of speech frames obtained by dividing said target speech; and

(b) performing said step (a) on another plurality of speech candidates;

wherein said step (a) includes the steps of;

(a1) determining a plurality of likelihoods representing similarities between one of said ordered plurality of speech frames, and consecutive ones of ordered phonemes in a plurality of phoneme strings representing one of said plurality of speech candidates;

(a2) determining a plurality of evaluation values representing similarities between said consecutive ones of said ordered phonemes and said target speech, based on said plurality of likelihoods determined in said step (a1) and on a plurality of transition probabilities corresponding to different combinations of said consecutive ones of the ordered phonemes; and

(a3) if said evaluation value for the head phoneme in said consecutive ones of said ordered phonemes is smaller than that of the last phoneme in said consecutive ones of said ordered phonemes, replacing phonemes to be collated for the next one of said speech frames with new consecutive ones of said ordered phonemes, wherein said new consecutive ones are said consecutive ones with the head phoneme removed from therefrom and with the next phoneme in the corresponding phoneme string added to said consecutive ones;

wherein said new consecutive ones of the ordered phonemes are used when said step (a) is carried out on said next speech frame in said ordered plurality of speech frames.

2. A speech recognition method according to claim 1, wherein said step (a3) includes the steps of:

if said evaluation value for the head phoneme in said consecutive ones of said ordered phonemes is not smaller than that of the last phoneme in said consecutive ones of said ordered phonemes, then determining said consecutive ones of the ordered phonemes with no modification as said new consecutive ones of the ordered phonemes.