Spectral Multi-scale Product Analysis for Pitch Estimation from Noisy Speech Signal

In this work, we present an algorithm for estimating the fundamental frequency in speech signals. Our approach is based on the spectral multi-scale product analysis. It consists of operating a short Fourier transform on the speech multi-scale product. The multi-scale product is based on making the product of the speech wavelet transform coefficients at three successive dyadic scales. The wavelet used is the quadratic spline function with a support of 0.8 ms. We estimate the pitch for each time frame based on its multi-scale product harmonic structure. We evaluate our approach on the Keele database. Experimental results show the effectiveness of our method presenting a good performance surpassing other algorithms. Besides, the proposed approach is robust for noisy speech.