Estimation by means of wavelet analysisof the signal - to - noise ratio of dysphonicvoicesM

| The objective is to explore a procedure for isolating additive noise in vowels, and estimating the signal-to-noise ratio. The method is founded on the wavelet transform, which is a family of multi-resolution analysis methods. The breakdown of sonorants into a noise and a signal constituent relates to the observation that additive noise is mostly high-frequency or low-amplitude. The segregation of the noise is therefore based on the isolation of the wavelets of the highest-resolution scale or, alternatively, on those whose amplitude is below a critical threshold. The analysis method was tested on a corpus of a],,i],,u] vowels sustained by healthy and dysphonic speakers. Results suggest that isolating noise via threshold-ing is preferable to assigning high-resolution scales to noise and low-resolution scales to the signal. Keywords| Signal-to-noise ratio, Wavelet Transform .