Extraction
Qualitative parametrization of the time-frequency characteristics for speaker recognition

We use the voicing measurement correlated with SNR, the equivalent of the harmonicity index . It is extracted from the autocorrelogram of the demodulated signal (see figure 1). According to H. Glotin , in the case of Gaussian noise (figure 2), the correlogram of the noise window is less modulated than an eigencorrelogram. . The peaks in the demodulation window of the autocorrelogram isolate the many harmonics in the signal. This can then be used to separate the mixed harmonic noise from the dominant harmonic signal, using a time window of almost the same duration as the average phoneme duration.