Speech recognition apparatus

• US 5,526,466 A
• Filed: 04/11/1994
• Issued: 06/11/1996
• Est. Priority Date: 04/14/1993
• Status: Expired due to Fees

• Current Assignee: Matsushita Electric Industrial Company Limited (Panasonic Holdings Corporation)
• Original Assignee: Matsushita Electric Industrial Company Limited (Panasonic Holdings Corporation)
• Inventors: Takizawa, Yumi
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): ONKA, THOMAS
• Application Number:
  US08/225,630
• Time in Patent Office: 792 Days
• Field of Search: 395/2.25, 395/2.4, 395/2.45, 395/2.46, 395/2.47, 395/2.54, 395/2.57, 395/2.62, 395/2.63, 381/41, 381/43
• US Class Current: 704/253
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/1807 using prosody or stress
• CPC Class Codes: G10L 2015/0638 Interactive procedures

Abstract

At the time of training reference speech, the relationship between durations of each recognition unit is obtained by a duration training circuit and, at the time of recognizing speech, a beginning and an end of input speech is detected by a speech period sensing circuit, and then by using the relationship and the input speech period length, the durations of the recognition units in the input speech are estimated. Next, the reference speech and the input speech are matched by the matching means by using the calculated estimation values in such a manner that the recognition units have a duration close to that of the estimated values.

Claims

1. A speech recognition apparatus which recognizes input speech input to said apparatus, the input speech including plural recognizable units which have a shorter duration than the input speech, said apparatus obtaining a recognition unit corresponding to each recognizable unit and outputting final recognition results by successively connecting each obtained recognition unit, said apparatus comprising:

speech period sensing means for detecting a speech period by sensing a beginning and an end of the input speech;

duration training means for calculating a relationship between durations of recognition units by using influencing factors which influence the duration of the recognition units;

duration estimation means for estimating a duration of each of the recognizable units within the input speech by using the speech period of the input speech detected by the sensing means and a duration relationship calculated by the training means; and

matching means for matching a recognition unit to a recognizable unit to produce a final recognition result based upon durations estimated by the estimation means.

2. A speech recognition apparatus according to claim 1, wherein

the duration training means calculates the relationship between the durations of the recognition units by obtaining an average duration of recognition units for each influencing factor.

3. A speech recognition unit according to claim 2, wherein

the duration training means calculates the relationship between the durations of the recognition units by obtaining, for each of the influencing factors, an extent α

, to which the influencing factor affects the duration of the recognition units and obtaining an average duration of the recognition units for each influencing factor.

4. A speech recognition apparatus according to claim 3, wherein

the duration training means calculates the duration, S(n), of each recognition unit for such case where a total length of the speech period is taken to be 1, the duration training means obtaining, for each factor, an average duration, f_j (t), of the recognition units and then obtaining a ratio between an average duration of a total speech period length for the input speech and the average duration of each recognition unit by using the obtained average duration, and further assigning weight to and adding together the ratio for each factor, and

the duration estimation means calculates estimated values, d(n), for the duration of each of the recognizable units of the input speech from S(n) and the total length, L, of the speech period, and wherein ##EQU8## where f_j (n) is the average duration of all recognition units as the nth recognition unit for factor j,

N is the total number of recognition units included in one speech period,

α

_j is the weighting coefficient for each factor, and

S(n) is the duration estimation value for each recognition unit when the total speech period length is 1, and wherein

space="preserve" listing-type="equation">d(n)=L×

S(n)

where d(n) is the duration estimation value for the nth recognition unit, and L is the total speech period length.

5. A speech recognition apparatus according to claim 3, wherein

the duration training means calculates the duration, S(n), of each recognition unit for such case where a total length of the speech period is taken to be 1, the training means obtaining an average duration, f_j (i), of the recognition units belonging to the same category for each factor and then estimating the duration of every recognition unit of the input speech by using the obtained average duration and further by obtaining a ratio between the estimated value and the total length of the speech period, and

the duration estimation means calculates estimated values for the duration of each of the recognizable units of the input speech from the duration S(n) and said total length of the speech period, L, wherein ##EQU9## where S(n) is the duration estimation value for each recognition unit when the total speech period length is 1,

N is the total number of recognition units included in one speech period, and ##EQU10## f_j (n) is the average duration of all recognition unite as the nth recognition unit for factor j, and

α

j is the weighting coefficient for each factor, and wherein

space="preserve" listing-type="equation">d(n)=L×

S(n)

where

d(n) is the duration estimation value for the nth recognizable unit and L is the total speech period length.

6. A speech recognition apparatus according to claim 1, wherein

the matching means carries out its matching and recognition result determinations using estimated duration values for the recognizable units such that a matching period for a recognition unit is close to an estimated duration value for a matched recognizable unit.

7. A speech recognition apparatus according to claim 1, further comprising

matching period deciding means for determining a matching period for each of the recognizable units in the input speech by using estimated duration values for the recognizable units, and wherein

the matching means carries out said matching and recognition results decisions such that each of the recognizable units is matched within a matching period determined by the period deciding means.

8. A speech recognition apparatus according to claim 1, further comprising

error calculation means for calculating a difference between an estimated duration and a duration of a recognition candidate and

results evaluation means for removing recognition candidates based upon a difference calculated by the error calculation means.