摘要 |
PROBLEM TO BE SOLVED: To realize a method and a device in which automatic speech recognition is performed by using stochastic accuracy information as to decoded bits in a decentralized ASR(automatic speech recognition) system to be used through wireless channel. SOLUTION: The reliability posessed by the system as to the decoded values of the feature of the ASR(automatic speech recognition) is computed (, for example, by using a maximum post probability algorithm) by using bit error probability with respect to respective bits which are used for coding the given feature of the ASR (which is a ceptstrum coefficient and a signal energy component based on, for example, a linear predictive coding analysis and is coded by using an ununiform error protection method). A feature having high damaged probability is not used or is weighted low by an acoustic distance computation which is performed in a speech recognizing device in more commonly. The method and device improve greatly the performance of the ASR under a bad channel condition. |