SPEAKER CERTIFYING PROBABILISTIC MATCHING METHOD

Japanese Patent JP3630216

B2

PROBLEM TO BE SOLVED: To improve speaker recognizing performance by efficiently performing probabilistic matching with a corresponding case of training voice data on aggregation of input test voice data.
SOLUTION: A first covariance matrix to express a probabilistic characteristic of information on a characteristic of an input test voice is generated on the basis of information on a characteristic of a concerned input test voice. Then, the conversion of information on a characteristic of its input test voice is performed. Its conversion is based on the first covariance matrix and a second covariance matrix to express a probabilistic characteristic of information on a characteristic of a training voice. Information on a characteristic of an already converted input test voice having a probabilistic characteristic exactly adapted by a probabilistic characteristic of information on a characteristic of the training voice, can be successfully obtained as a result by such conversion. This conversion is also desirably based on a probabilistic average value of information on a characteristic of the training voice.

Kipee.

JP6345198A

March 16, 2005

March 13, 1998

Click for automatic bibliography generation  

Lucent Technologies, Inc.

G10L15/10; G10L15/00; G10L15/06; G10L15/14; G10L17/00; (IPC1-7): G10L17/00; G10L15/00

JP3054600A
JP5241589A

WO1996041334A1

柳井春夫外著「多変量解析ハンドブック」(株式会社現代数学社)第18~20頁(20.04.1986)

Masao Okabe
Nobuaki Kato
Kazuo
Shinichi Usui
Ikuo Fujino
Takao Ochi
Teruhisa Motomiya
Norimichi Takanashi
Asahi Shinmitsu
Seiichiro Takahashi
Koji Yoshizawa
Hirofumi Mimata

Previous Patent: MEDULLARY CAVITY NAIL FOR TREATMENT TO HIP JOINT BREAKAGE, AND DEVICE TO HANDLING THE SAME

Next Patent: METHOD FOR DETECTING SHORT-CIRCUIT OF BATTERY