Speech endpoint detection method and apparatus and continuous speech recognition method and apparatus

• US 5,740,318 A
• Filed: 09/27/1995
• Issued: 04/14/1998
• Est. Priority Date: 10/18/1994
• Status: Expired due to Term

• Current Assignee: KDDI Corporation
• Original Assignee: Kokusai Denshi Denwa Company Limited (KDDI Corporation)
• Inventors: Naito, Masaki, Takeda, Kazuya, Kuroiwa, Shingo, Yamamoto, Seiichi
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Dorvil, Richemond
• Application Number:
  US08/534,805
• Time in Patent Office: 930 Days
• Field of Search: 395/2.62, 395/2.65, 395/2.54, 395/2.48, 395/2.47, 395/2.5, 395/2.42, 395/2.6, 395/2.63, 395/2.64, 395/2.41, 395/2.19, 395/2.57, 395/2.59, 395/2.24
• US Class Current: 704/253
• CPC Class Codes: G10L 15/193 Formal grammars, e.g. finite state automata, context free grammars or word networks
• CPC Class Codes: G10L 25/87 Detection of discrete points within a voice signal

Abstract

A speech endpoint detection unit determines that a first condition is met when a matching score of a partial sentence accepted by a grammatical rule unit is the highest of all partial sentences, and that a second condition is met when a duration time of input speech determined to coincide with a silent standard pattern is longer than a predetermined time. A speech endpoint is determined when the first and second conditions are both met and a speech endpoint detection signal is sent to a word prediction unit and a recognition result output unit. By requiring that both first and second conditions be met, a speech endpoint can correctly be detected even when a long silent period is present in the course of a sentence.

Claims

1. A speech endpoint detection method used in a continuous speech recognition process which recognizes continuous speech by continuously matching an input speech with word standard patterns according to a grammatical rule, said method comprising the steps of:

checking a matching score and structure of a partial sentence obtained by matching the input speech with the word standard patterns;

determining a duration time of an input speech determined to coincide with a silent standard pattern; and

judging an end of speech based on said checking and determining steps.

2. The method according to claim 1, wherein the continuous speech recognition process is performed by a speech recognition apparatus.

3. A speech endpoint detection method comprising the steps of:

continuously matching an input speech with word standard patterns according to a grammatical rule;

determining partial sentences and corresponding matching scores by matching the input speech with the word standard patterns;

determining whether or not a first condition is met depending on whether the matching score of a partial sentence accepted by the grammatical rule is the best of all partial sentences;

determining whether or not a second condition is met depending on whether a duration time of an input speech determined to coincide with a silent standard pattern is more than a predetermined time; and

determining a time at which the first and second conditions are both met as an end of speech.

4. A continuous speech recognition apparatus for recognizing a continuous speech by continuously matching an input speech with word standard patterns according to a grammatical rule, comprising:

structure analysis means for searching for a partial sentence which is acceptable by the grammatical rule from partial sentences obtained by matching the input speech with the word standard patterns;

matching score comparison means for determining whether or not a matching score of the partial sentence accepted by the grammatical rule is the best matching score of all the partial sentences;

silent duration time determination means for determining whether or not a duration time of an input speech determined to coincide with a silent standard pattern is longer than a predetermined time, from the result of matching the input speech with the word standard patterns; and

speech endpoint determination means for determining an end of speech when the matching score of a partial sentence accepted by the grammatical rule is determined to be the best of matching scores of all the partial sentences, and the silent duration of time is determined to be longer than the predetermined time.

5. A speech endpoint detection apparatus used in a continuous speech recognition apparatus for recognizing a continuous speech by continuously matching an input speech with word standard patterns according to a grammatical rule, comprising:

matching score comparison means for searching for a partial sentence having the best matching score from partial sentences and corresponding matching scores obtained by matching the input speech with word standard patterns;

structure analysis means for analyzing whether or not the partial sentence having the best matching score is accepted by the grammatical rule;

silent duration time determination means for determining whether or not a duration time of an input speech determined to coincide with a silent standard pattern is longer than a predetermined time, from the result of matching the input speech with the word standard patterns; and

speech endpoint determination means for determining an end of speech when the partial sentence having the best matching score is determined to be a sentence accepted by the grammatical rule, and the silent duration time is determined to be longer than a predetermined time.

6. A continuous speech recognition method for recognizing continuous speech by continuously matching an input speech with word standard patterns according to a grammatical rule, said method comprising the steps of:

starting speech recognition when a power of input speech exceeds a threshold value for more than a predetermined time;

determining whether or not a first condition is met depending on whether a matching score of a partial sentence accepted by the grammatical rule is the best of matching scores of all partial sentences, from individual partial sentences and corresponding matching scores obtained by matching the input speech with the word standard patterns;

determining, based on the result of matching the input speech with the word standard patterns, whether or not a second condition is met depending on whether a duration time of an input speech determined to coincide with a silent standard pattern is more than a predetermined time; and

ending speech recognition when the first and second conditions are both met.

7. A continuous speech recognition method used in a continuous speech recognition process for recognizing continuous speech by continuously matching an input speech with word standard patterns according to a grammatical rule, said method comprising the steps of:

starting speech recognition when the continuous speech recognition apparatus begins operation;

determining whether or not a first condition is met depending on whether the matching score of a partial sentence accepted by the grammatical rule is the best of matching scores of all partial sentences, from individual partial sentences and corresponding matching scores obtained by matching the input speech with the word standard patterns;

determining, from the result of matching the input speech with the word standard patterns, whether or not a second condition is met depending on whether a duration time of the input speech determined to coincide with a silent standard pattern is more than a predetermined time; and

ending speech recognition when the first and second conditions are both met.

8. A continuous speech recognition apparatus for recognizing a continuous speech by continuously matching an input speech with word standard patterns according to a grammatical rule, comprising:

speech startpoint detection means for determining whether a power of the input speech exceeds a threshold value for more than a predetermined time, and outputting a speech startpoint detection signal when the condition continues for more than the predetermined time;

structure analysis means for searching for a partial sentence which is accepted by the grammatical rule from partial sentences obtained by matching the input speech with the word standard patterns;

matching score comparison means for determining whether or not a matching score of the partial sentence accepted by the grammatical rule is the best of all partial sentences, from matching scores of the partial sentences obtained by matching the input speech with the word standard patterns;

silent duration time determination means for determining whether or not a duration time of an input speech determined to coincide with a silent standard pattern is longer than a predetermined time, from the result of matching the input speech with the word standard patterns;

silent duration time determination means for determining whether or not a duration time of an input speech determined to coincide with a silent standard pattern is longer than a predetermined time, from the result of matching the input speech with the word standard patterns; and

speech endpoint determination means for determining an end of speech when the matching score of a partial sentence accepted by the grammatical rule is determined to be the best of the matching scores of all the partial sentences, and the silent duration time is determined to be longer than the predetermined time,

whereby speech recognition is started by the output of the speech startpoint detection signal, and speech recognition is ended by an output of a speech endpoint detection signal.

9. A continuous speech recognition apparatus for recognizing a continuous speech by continuously matching an input speech with word standard patterns according to a grammatical rule, comprising:

speech startpoint detection means for determining whether or not a power of the input speech exceeds a threshold value for more than a predetermined time, and outputting a speech startpoint detection signal when the power of the input speech exceeds the threshold value for more than the predetermined time;

matching score comparison means for searching for a partial sentence having the best matching score from partial sentences and corresponding matching scores obtained by matching the input speech with the word standard patterns;

structure analysis means for analyzing whether or not the partial sentence having the best matching score is accepted by the grammatical rule;

silent duration time determination means for determining whether or not a duration time of an input speech determined to coincide with a silent standard pattern is longer than a predetermined time, from the result of matching the input speech with the word standard patterns; and

speech endpoint determination means for outputting a speech endpoint detection signal when the partial sentence having the best matching score is determined to be a sentence acceptable to the grammatical rule, and the silent duration time is determined to be longer than the predetermined time,

whereby speech recognition is started by the output of the speech startpoint detection signal, and speech recognition is ended by the output of the speech endpoint detection signal.

10. A continuous speech recognition apparatus for recognizing a continuous speech by continuously matching an input speech with word standard patterns according to a grammatical rule, comprising:

structure analysis means for searching for a partial sentence which is accepted by the grammatical rule from partial sentences obtained by matching the input speech with the word standard patterns;

matching score comparison means for determining whether or not a matching score of the partial sentence accepted by the grammatical rule is the best of all partial sentences, from the matching scores of partial sentences obtained by matching the input speech with the word standard pattern;

silent duration time determination means for determining whether or nor a duration time of an input speech determined to coincide with a silent standard pattern is longer than a predetermined time, from the result of matching the input speech with the word standard patterns; and

speech endpoint determination means for outputting a speech endpoint detection signal when the matching score of the partial sentence accepted by the grammatical rule is determined to be the best of all the partial sentences, and the silent duration time is determined to be longer than the predetermined time;

whereby speech recognition is carried out from the starting operation of the apparatus until the speech endpoint detection signal is output.

11. A continuous speech recognition apparatus for recognizing a continuous speech by continuously matching an input speech with word standard patterns according to a grammatical rule, comprising:

matching score comparison means for searching for a partial sentence having the best matching score from partial sentences and corresponding matching scores obtained by matching the input speech with the word standard patterns;

structure analysis means for analyzing whether or not the partial sentence having the best matching score is acceptable by the grammatical rule;

silent duration time determination means for determining whether or not a duration time of an input speech determined to coincide with a silent standard pattern is longer than a predetermined time, from the result of matching the input speech with the word standard patterns; and

speech endpoint determination means for outputting a speech endpoint detection signal when the partial sentence having the best matching score is determined to be a sentence acceptable to the grammatical rule, and the silent duration time is determined to be longer than the predetermined time;

whereby speech recognition is carried out from the starting operation of the apparatus until the speech endpoint detection signal is output.