15th International Congress of Phonetic Sciences (ICPhS-15)

Barcelona, Spain
August 3-9, 2003

Simulating the Effects of Additive White Noise on Speech - A Perceptual Study

Noam Amir, Liat Kishon-Rabin

Tel Aviv University, Israel

In this study we suggest a model for the masking process and attempt to verify it experimentally. The model is based on a sinusoidal coding algorithm, where sinusoidal components that would be masked at certain a noise level are removed. Two signals are resynthesized: one is the signal composed of all the sinusoids remaining above the masking threshold, and the second is the signal composed of all the sinusoids below the masking threshold.
   Extensive listening tests show that the identification percentages on words that underwent "noiseless masking" (NM) were somewhat lower than those achieved on lists with additive white noise. Interestingly, adding white noise to the NM lists improved the recognition rates. Though further refinements to the model are necessary to provide a better match to standard masking with white noise, the results here give promising insights into the masking process.

Full Paper

Bibliographic reference.  Amir, Noam / Kishon-Rabin, Liat (2003): "Simulating the effects of additive white noise on speech - a perceptual study", In ICPhS-15, 813-816.