Title:
【発明の名称】音声認識方式
Document Type and Number:
Japanese Patent JP3135931
Kind Code:
B2
Abstract:
PURPOSE:To execute the precise continuation time control by giving prescribed similarity, when the continuation time of a partial pattern of an input voice is within a specific allowable range, and giving similarity corresponding to a shift from the average, when the continuation time is outside the specified allowable range. CONSTITUTION:The above system is provided with a feature sequence converting means for converting an inputted voice waveform to an input voice pattern being a time series of a feature vector, and similarity/non-similarity of a partial pattern X and a standard pattern Y of the input voice pattern is calculated by the sum of similarity/non-similarity Rv of the respective feature vectors x, y, and similarity/non-similarity Rt of the continuation time i-m of the partial pattern X and the continuation time l of the standard pattern Y. In this case, the similarity/non-similarity of the continuation time is calculated as Rt=min(a, b-c (i-m-l)<2>). In this regard, a to c denote parameters for evaluating the continuation time. Accordingly, when the continuation time i-m of the partial pattern of the input voice is within an allowable range of the expression, prescribed similarity (a) is given, and when the time is outside the allowable range, b-c (i-m-l)<2> is given.
Inventors:
Tetsuya Muroi
Application Number:
JP4248991A
Publication Date:
February 19, 2001
Filing Date:
February 14, 1991
Export Citation:
Assignee:
株式会社リコー
International Classes:
G10L17/08; G10L15/02; G10L15/10; (IPC1-7): G10L15/10
Domestic Patent References:
| JP2248999A | ||||
| JP277099A | ||||
| JP63104098A | ||||
| JP431118B2 |
Other References:
【文献】特許2804265(JP,B2)
【文献】特許3083855(JP,B2)
【文献】特許3083855(JP,B2)
Attorney, Agent or Firm:
Akichika Takano