| 摘要 |
PURPOSE:To enhance the decision accuracy and to attain a high speed treatment by judging a sound voice, a silent voice or a silent section automatically by using plural judgement conditions. CONSTITUTION:First at a noise level measuring section 11, the noise level N is measured and subsequently the segmentations of a short section (frame) at about 10-30msec are executed and at every each short section, with a linear predictive analysis section 13, a high speed Fourier transform section 15 and a correlation device 17, a primary partial autocorrelation coefficient K 1, a power ratio gamma (=PL/PH) of the power PH of a frequency component higher than it at the boundary of 2kHz, the power PT of the whole frequency band region of the relevant frame and the maximum value rhoMAX of the autocorrelation coef. rho for the predictive error from the linear predictive analysis are obtained. The prescribed logic is conducted at the decision section 19 by using those values and the discrimination of a sound voice, a silent voice and a silence section of the voice signal is conducted. |
