Method of speaker adaptive speech recognition

• US 5,170,432 A
• Filed: 09/21/1990
• Issued: 12/08/1992
• Est. Priority Date: 09/22/1989
• Status: Expired due to Term

• Current Assignee: Alcatel NV (Nokia Corporation)
• Original Assignee: Alcatel NV (Nokia Corporation)
• Inventors: Hackbarth, Heidi, Immendorfer, Manfred
• Primary Examiner(s): Shaw, Dale M.
• Assistant Examiner(s): Tung, Kee M.
• Application Number:
  US07/586,086
• Time in Patent Office: 809 Days
• Field of Search: 381/41-43
• US Class Current: 704/254
• CPC Class Codes: G10L 15/07 to the speaker

Abstract

A method for recognizing spoken words of a speech includes extracting feature vectors from a speech signal which corresponds to a spoken phrase, and segmenting and classifying the successive extracted feature vectors into syllable oriented word subunits by means of a stored supply of word subunits to form a set of hypotheses. The set of hypotheses is used to generate, by three dimensional time dynamic comparision, a set of word hypotheses by comparing the segmented and classified word subunits with standard pronunciations and pronunciation variants of a plurality of words stored in a reference pattern vocabulary. The generated set of word hypotheses are then subjected to syntactic analysis to determine the spoken phrase.

Claims

1. A method for recognizing spoken words of a speech, comprising the steps of:

extracting feature vectors from a speech signal corresponding to a spoken phrase to be recognized;

segmenting and classifying the successive extracted feature vectors into syllable oriented word subunits by means of a stored supply of word subunits to form a set of hypotheses;

comparing the set of hypotheses formed from the segmented and classified word subunits with standard pronunciations and pronunciation variants of a plurality of words stored in a reference pattern vocabulary over a three-dimensional time dynamic period to generate a set of word hypotheses; and

subjecting the generated set of word hypotheses to syntactic analysis in order to determine the spoken phrase.

2. A method according to claim 1, wherein the speech is continuous and the comparing step generates overlapping word hypotheses, the step of subjecting the generated set of word hypotheses to syntactic analysis enabling the spoken phrase to be determined.

3. A method according to claim 1, further comprising the step of adapting the stored reference pattern vocabulary of speech data to a new speaker by means of a hybrid statement based on spoken utterances made during a brief training phase for this new speaker.

4. A method according to claim 3, wherein the adapting step includes adapting the feature vectors as well as the stored word subunits to the new speaker.

5. A method according to claim 1, further comprising the steps of compiling and expanding the stored reference pattern vocabulary by inputting written text and converting this text on the basis of syntactic rules into symbols for word subunits.

6. A method according to claim 1, further comprising the step of preselecting a sub-vocabulary of the stored reference pattern vocabulary with the aid of the stored syllable oriented word subunits in order to accelerate recognition of speech with large stored vocabularies.

7. A method according to claim 1, wherein the feature vectors extracted from a speech signal are based on intensities of various frequency ranges present in the signal.

8. A method according to claim 1, wherein the pronunciation variants include a group consisting of:

linear variations of individual word subunits, variations which omit a syllable of a word, and variations which insert an additional syllable to a word.