| 摘要 |
PURPOSE: To perform voice recognition without a DMA controller by divisionally and alternately performing the transfer of the voice data in terms of analytic frame to a memory and decision of voice/non-voice. CONSTITUTION: An input voice is amplified by a microphone 1, and converted into the digital data (A/D conversion data) by an A/D conversion means 2 and stored successively in an FIFO incidental to the means 2. An A/D conversion data transfer means 41 transfers the A/D conversion data of one analysis frame to the memory 5, and the data are stored in the memory 5. Then, the means 41 reads out them, and decides the voice/non-voice. In a process alternately performing such a processing, a voice section decision means 42 performs voice section decision. A frequency analysis means 43 frequency analyzes related to the voice section and specifies an input voice pattern. A collation means 44 collates the input voice pattern with the patterns awaited beforehand in a voice dictionary 6, and retrieves the nearest one, and specifies the input voice. Then, the input voice is outputted from a recognition result output means 45. |
