7th International Workshop on Models and Analysis of Vocal Emissions for Biomedical Applications (MAVEBA 2011)

Florence, Italy
August 25-27, 2011

Multiscale Product Correlation For The Open Quotient Estimation From The Noisy Speech Signal

Wafa Saidi, Aicha Bouzid, Noureddine Ellouze

LSTS, ENIT-Tunis Le Belvedere B.P 37, Tunis, Tunisia

In this paper, an improved Multiscale Product-based Method is evaluated for the open quotient (OQ) estimation from the noisy speech signal. The method consists of making the multiscale wavelet transforms coefficients product at three scales. Our proposed approach is based upon correlation functions computed on the negative and the positive parts of the speech Multiscale Product (MP). It operates without determining the glottal opening and closing instants. Over each frame, the pitch period is given by the first maximum of the MP negative part autocorrelation and the open phase is given by the first maximum of the inter-correlation between the negative and positive parts. OQ is the ratio of the open phase and the pitch period. Tested on the Keele University database, our new approach proves to be robust to noise degradation. Key words: open quotient, speech, multi-scale product, correlation, noise.

Full Paper (reprinted with permission from Firenze University Press)

Bibliographic reference.  Saidi, Wafa / Bouzid, Aicha / Ellouze, Noureddine (2011): "Multiscale product correlation for the open quotient estimation from the noisy speech signal", In MAVEBA-2011, 119-122.