| 摘要 |
PROBLEM TO BE SOLVED: To provide a method and a device for voice recognition adaptive to even a voice of high speaking speed. SOLUTION: The voice recognition device provided with a microphone 10, a sound analyzer 20, a sound model 30, a langue model 40, a probability network generator 50, and a memory 60 and is further provided with a collator 70 capable of collating a voice with a plurality of transitional states of a probability network in a prescribed time section. A self-transitional state, a transitional state adjacent to it, and a transitional state adjacent to it are collated. Thus even the voice of high speaking speed is collated with all transitional states from the start point to the end point of the probability network, so that the device can be adaptive to the voice of high speaking speed. Multiple-time collation with the same transitional state is added, so that the device can be adaptive to a voice of normal speaking speed by the same algorithm of multiple-time collation. Larger one of both collation results is selected as a score for recognition. Thus accurate voice recognition of not only the voice of high speaking speed but also the voice of normal speaking speed is possible. COPYRIGHT: (C)2004,JPO
|
