Speech recognition method having noise immunity

• US 4,713,777 A
• Filed: 05/27/1984
• Issued: 12/15/1987
• Est. Priority Date: 05/27/1984
• Status: Expired due to Term

• Current Assignee: Voice Industries Corporation
• Original Assignee: Exxon Research and Engineering Company (Exxon Mobil Corporation)
• Inventors: Lee, Chin-Hui, Ganesan, Kalyan, Klovstad, John W.
• Primary Examiner(s): Kemeny, E. S. Matt
• Application Number:
  US06/593,892
• Time in Patent Office: 1,297 Days
• Field of Search: 381/41-43, 364/513, 364/513.5
• US Class Current: 704/233
• CPC Class Codes: G10L 15/00 Speech recognition G10L17/00 takes precedence

Abstract

In a speech recognition system, the beginning of speech versus non-speech (a cough or noise) is distinguished by reverting to a non-speech decision process whenever the liklihood cost of template (vocabulary) patterns, including silence, is worse than a predetermined threshold, established by a Joker Word which represents a non-vocabulary word score and path in the grammar graph.

Claims

1. In a speech recognition apparatus wherein speech units are each characterized by a sequence of template patterns, and having

means for processing a speech input signal for repetitively deriving therefrom, at a frame repetition rate, a plurality of speech recognition acoustic parameters, and

means responsive to said acoustic parameters

for generating likelihood costs between said acoustic parameters and said speech template patterns, and

for processing said likelihood costs for determining the speech units in said speech input signal,

a method for inhibiting a response to nonvocabulary utterances in a speech input for which template patterns have not been created, comprising the steps of

repeatedly, at a frame repetition rate, generating acoustic parameters representing said speech input,

generating likelihood costs at each frame time for said acoustic parameters and said template patterns, said template patterns including a pattern representing silence,

beginning a normal speech recognition process whenever said cost for an active template pattern is better than a predetermined threshold value, and

reverting to a non-speech recognition process whenever said cost of said template patterns, including silence, is worse than said predetermined threshold value.

2. The method of claim 1 further comprising the steps of

setting a second threshold value and

remaining in said non-speech recognition process until a likelihood cost of said silence template in better than said second threshold.

3. The method of claim 2 further wherein there are two silence template patterns, a first short silence pattern employed during said reverting step, and a second long silence pattern employed during said remaining step.

4. In a speech recognition apparatus wherein speech units are each characterized by a sequence of template patterns, and having

means for processing a speech input signal for repetitively deriving therefrom, at a frame repetition rate, a plurality of speech recognition acoustic parameters, and

means responsive to said acoustic parameters

for generating likelihood costs between said acoustic parameters and said speech template patterns, and

for processing said likelihood costs for determining the speech units in said speech input signal,

a method for inhibiting a response to nonvocabulary utterances in a speech input for which template patterns have not been created, comprising the steps of

repeatedly, at a frame repetition rate, generating acoustic parameters representng said speech input,

generating likelihood costs at each frame time for said acoutic parameters and said template patterns, said template patterns including a pattern representing silence,

employing dynamic programming and a grammar graph for determining in response to said likelihood costs whether there has been a nonvocabulary utterance, said grammar graph having a normal speech recognition branch and a non-speech recognition branch, said non-speech recognition branch corresponding to nonvocabulary utterances for which said template patterns have not been created, and

said employing step determining and selecting, using said dynamic programming, the better scoring of said speech recognition and said non-speech recognition branches.

5. The method of claim 4 further comprising the steps of

assigning a fixed predetermined threshold score to an entrance arc of said non-speech recognition branch of said grammar,

assigning a second fixed predetermined threshold score to a self loop are of said non-recognition branch for providing a path to remain in said non-recognition branch, and

providing a silence arc leading from said self loop arc to the beginning of said entrance arc.

6. The method of claim 5 further comprising the step of

assigning a self loop silence arc at a grammar node from which said entrance arc originates.

7. The method of claim 6 further comprising the steps of

providing said silence arc with a good score for a long silence, and

providing said self loop silence arc with a good score for a short silence.