摘要 |
<p>PROBLEM TO BE SOLVED: To sufficiently eliminate the breathing noise in gullet voicing. SOLUTION: Input signals 10 of gullet voicing are digitized (12). Then, the digital signals are fast Fourier transformed(FFT) (14). Then, the average value of the FFT transformation is computed (14). Then, the transformation is inputted to a form filter to produce a filter spectrum (16). Then, the differential coefficients of the filter spectrum, that is the output of the filter, are computed. Then, based on the average value and the differential coefficients, the positions and the values of a maximum peak among the filter spectrum and a second largest peak (called a second maximum peak) are determined. Then, based on the positions and the values of the maximum peak and the second maximum peak, the generation of the breathing noise of the uttered signals is determined. Moreover, during a prescribed period, the generation of silence is determined based on whether there exists energy or not while moving a 10 ms window (18). Then, after the detection of the generation of breathing noise, uttered signals are passed and the uttered signals after the generation of the silence is prevented.</p> |