| 摘要 |
PURPOSE: A multi-pulse speech coding system using a least square method is provided to correct a reproduced voice waveform using the least square method in case a voiced sound and an unvoiced sound exist in the same frame, thereby improving sound quality. CONSTITUTION: A silence, a voiced sound and an unvoiced sound (S/V/UV) are detected (206). A transition segment including unvoiced consonant (TSIUVC) is extracted from an individual pitch pulse and an error signal of a finite impulse response and simplified technique for recursive estimate autocorrelation K parameters (FIR-STREAK) filter (203) used for detecting the S/V/UV. The error signal is controlled by applying a least mean square (LMS) algorithm (209) to the TSIUVC. A maximum error signal is produced to control the error signal by applying the LMS algorithm to the TSIUVC. Distortion of a voice waveform is controlled by applying the maximum error signal. |
