| 摘要 |
When subjected to frequency analysis and plotted in a frequency spectral distribution, a voice signal typically includes a monotonously and relatively slowly changing component and a relatively rapidly changing component. For the recognition of voice sound, the relatively rapidly changing component contains phonemic information and thus is more important. In order to extract such a relatively rapidly changing component containing phonemic information from a voice signal, the voice signal is first subjected to frequency analysis to obtain a frequency spectral distribution, which is then sampled from one end to the other and then in the reversed order in timed sequence repetitively to produce a periodic waveform. Then, the thus obtained periodic waveform is filtered to remove the relatively slowly changing component thereby extracting the relatively rapidly changing component. |
