Title:
METHOD AND DEVICE FOR VOICE RECOGNITION
Document Type and Number:
Japanese Patent JP3513284
Kind Code:
B2
Abstract:
PROBLEM TO BE SOLVED: To provide the method and the device for voice recognition that is small in size, fast in speed and high in performance by eliminating trellis and viterbi searches for rejection and reducing the trellis space for the rejection.
SOLUTION: Inputted voices are acoustically analyzed and voice parameters are obtained (202). Then, an output probability against each state of each hidden Markov model is computed (203). Then, a language search is conducted using the language information including a dictionay and a grammar, a hidden Markov model and its state output probability and recognition candidates of the inputted voices and their likelihood are computed (205). Simultaneously, the output probability against each state of each hidden Markov model is computed and using this result, a maximum state output probability and its state are obtained against the analyzed voice parameters (207). Then, the likelihood for rejection for the voice input is obtained through the use of the state transition probability determined by the way of state transition (208). Then, based on the ratio of the two likelihood, a decision is made to output the recognition result of the voice input by a threshold value or to reject the result (209).
More Like This:
| JP3549681 | Vocal identification proof for recognition of concatenated numbers |
| JP2006171185 | SPEECH RECOGNITION DEVICE AND METHOD |
| WO/2001/026092 | ATTRIBUTE-BASED WORD MODELING |
Inventors:
Komori, Yasuhiro
Yamada, Masaaki
Ohora, Yasunori
Yamada, Masaaki
Ohora, Yasunori
Application Number:
JP26152795A
Publication Date:
March 31, 2004
Filing Date:
October 09, 1995
Export Citation:
Assignee:
CANON INC
International Classes:
G10L15/14; G10L15/08; G10L15/28; (IPC1-7): G10L15/28; G10L15/14
Other References:
Ｏｚｅｋｉ，Ｔｈｅ ｍｕｔｕａｌ ｉｎｆｏｒｍａｔｉｏｎ ａｓ ａ ｓｃｏｒｉｎｇ ｆｕｎｃｔｉｏｎ ｆｏｒ ｓｐｅｅｃｈ ｒｅｃｏｇｎｉｔｉｏｎ，電子情報通信学会技術研究報告［音声］，日本，１９９５年１２月１５日，ＳＰ９５−１０１，ｐ．５３−６０
渡辺， 塚田，音声認識を用いたゆう度補正による未知発話のリジェクション，電子情報通信学会論文誌 Ｄ−ＩＩ，日本，１９９２年１２月，Ｖｏｌ．Ｊ７５−Ｄ−ＩＩ， Ｎｏ．１２，ｐ．２００２−２００９
野田， 嵯峨山，前向き尤度を用いたＡ＊ビーム探索によるＨＭＭ−ＬＲ音声認識，電子情報通信学会技術研究報告［音声］，日本，１９９４年 ６月１７日，ＳＰ９４−２３，ｐ．１−７
渡辺， 塚田，音声認識を用いたゆう度補正による未知発話のリジェクション，電子情報通信学会論文誌 Ｄ−ＩＩ，日本，１９９２年１２月，Ｖｏｌ．Ｊ７５−Ｄ−ＩＩ， Ｎｏ．１２，ｐ．２００２−２００９
野田， 嵯峨山，前向き尤度を用いたＡ＊ビーム探索によるＨＭＭ−ＬＲ音声認識，電子情報通信学会技術研究報告［音声］，日本，１９９４年 ６月１７日，ＳＰ９４−２３，ｐ．１−７
Attorney, Agent or Firm:
大塚 康徳 （外１名）