A perception-based processing strategy for cochlear implants and speech coding

We propose a perception-based signal processing strategy to encode slowly-varying amplitude and frequency modulations within a frequency analysis band. We have simulated a four-band version of this processor and showed that vowel recognition is improved by adding frequency modulation, particularly in noise. This strategy can be used to efficiently encode fine structure in cochlear implants and to improve speech recognition in noise. Because the center frequency is removed in the analysis, this technique also has the potential to provide high-quality, low-bit speech coding and synthesis.