| 摘要 |
A speech feature vector extracting apparatus and a method, and a speech recognition system and a method are provided to suppress pitch harmonic components for a speech spectrum to obtain a speech spectrum emphasizing formant, and to extract an exact feature vector based on the obtained speech spectrum, thereby improving speech recognition performance. A speech feature vector extracting apparatus includes a preprocessing unit(210), an FET processing unit(230), a formant emphasizing unit(250), a filter bank processing unit(270), and a DCT processing unit(290). The preprocessing unit constitutes a speech signal in units of frames. The FET processing unit converts the speech signal into a frequency signal. The formant emphasizing unit suppresses pitch harmonic components included in component components of the frequency signal to emphasize formant. The filter bank processing unit band-pass-filters the frequency signal using a mel-scale filter bank. The DCT processing unit performs a DCT process for each filter bank signal from the filter bank processing unit to extract a final MFCC feature vector. |
