We propose a novel algorithm for tracking multiple sinusoidal signals that is motivated by neural coding in the mammalian peripheral auditory system. A striking feature of auditory nerve activity is the phenomenon of “synchrony capture,” whereby the most intense frequency components in the stimulus dominate the temporal firing patterns of whole subpopulations of auditory nerve fibers (ANFs). A novel adaptive filterbank structure that emulates key aspects of synchrony capture is presented. The proposed filterbank has two components: a fixed bank of traditional gammatone (or equivalent) filters that are cascaded with a bank of adaptively-tunable bandpass filter triplets. The bandpass filters are tuned by using a voltage controlled oscillator (VCO) whose frequency is steered by a frequency discriminator loop (FDL). The resulting filterbank is used to process synthetic signals and speech. It is shown that the VCOs can track the low frequency harmonics in speech that evoke voice pitch at their fundamental (F0). For vowels, the VCOs faithfully track the strongest harmonic present in each formant region.
展开▼
