摘要 |
In a speech recognition device, the threshold value which establishes the degree of similarity between a spoken command and a stored reference command and on the basis of which the commands are deemed to correspond is, according to the invention, automatically corrected in a type of learning process. The correction is performed taking into account the mean value or the spread of the Tschebyscheff interval between the spoken command and the reference command most similar to it. The correction is performed every time that the command concerned is entered. Even if the reference command has been recorded under abnormal conditions, the recognition frequency can be improved automatically in this way. To perform the aforementioned corrections, the device according to the invention comprises, in addition to the assemblies of a conventional speech recognition device, a similarity threshold memory (20) and a threshold calculation unit (19). <IMAGE> |