Project Abstract

Conventional speech processing methods primarily rest on the time-frequency domain to design statistical tools to enhance noisy speech signals. Recent research shows that the modulation domain offers a greater demarcation between signals such as speech and environmental noise.

This is because the modulation domain offers additional information, such as the non-stationarity of the vocal tract, which better represents the evolution of spectral-temporal information of speech. Nevertheless, there is still much to be explored with the multi-dimensional space of the modulation domain and its integration framework into speech applications.

This project will delve deep into the modulation domain to define its representation and establish the associated properties with speech intelligibility. These findings will then be used to develop new tools and statistical models to enhance speech intelligibility or other acoustics signals. Applications for the output of this research are assistive listening devices, mobile applications for clearer speech communication and hearing aids.