Speech signal determination method for vocoder

• Main Authors: LIN, CHIN-TENG, TAINAN, TW, LIN, HSIN-AN, I-LAN, TW
• Format: Patent
• Language: English; German
• Published: 02.07.1998
• Edition: 6
• Subjects: ACOUSTICS; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

The method involves receiving speech frames of the input speech and dividing the frame into four subframes. It is evaluated if the normalised correlation (NC) of the speech frame is bigger or the same as a high threshold. The four subframes are set as frames with speech when the value is greater than the threshold otherwise the subframes are set to frames without speech. The coefficients of the energy and of the line spectrum (LSP) of the speech signal are calculated separately for determination if the speech frame is stationary or non-stationary. If the energy coefficient and the LSP coefficient are not smaller than the corresponding thresholds further determination of the states with speech and without speech is accomplished by calculating the energy ratio of the lower band to the upper band for each subframe. If the energy coefficient and the LSP coefficient are smaller than the corresponding thresholds each subframe is set to the same value as the previous speech frame. If the energy ratio is bigger than a threshold the subframes are set to frames with speech otherwise the subframes are set to frames without speech.