Novel Spectro-Temporal Codes and Computations for Auditory Signal Representation and Separation
Abstract
In the past three years, we have developed algorithms that emulate the phenomenon of "synchrony capture" in the auditory nerve. Synchrony capture means that the dominant component in a given frequency band preferentially drives auditory nerve fibers innervating the entire corresponding frequency region of the cochlea. Our algorithm, called the synchrony capture filterbank (SCFB) consists of a bank of broadly tuned filters (not unlike the basilar membrane) in cascade with narrower filters (not unlike outer hair cells) that adaptively lock onto locally-dominant frequency components to produce synchrony capture. This local behavior enables a robust encoding of the running power spectrum based on relative numbers of channels recruited by different frequency components. The filterbank precisely tracks individual time-varying frequency components, such as low harmonics and formant frequencies in speech, in the midst of noise and auditory clutter. This precise tracking in turn can be used to enhance the separation of concurrent periodic sounds. We envision that the project will result in improved front-ends that can enhance voices in noise and better separate sounds.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 2013
- Accession Number
- ADA581226
Entities
People
- Peter Cariani
- R. Kumaresan
- Vijay K. Peddinti
Organizations
- University of Rhode Island