Biologically motivated pitch detector by using PLLs and bayesian estimation
AbstractThis work proposes a signal processing system based on biological evidence that reflects an aspect of the representation of the speech signals in the peripheral auditory system, namely the pattern of multiple pitch synchronous signals in the inner ear. There are two main characteristics observed in this representation: on one hand the recovery of the pitch envelope using the high frequency portion of the spectrum; on the other hand, synchronous behavior with the fundamental harmonic displayed by these signals. Phase Locked Loops (PLLs) are used to determinate the frequencies present into the signals, obtaining at the same time good noise robustness. The system generates multiple observations of the harmonic frequencies corresponding to high portion of the spectrum, then those observations are fed to a Bayesian statistical detector which finally obtains the pitch using a variation of a Gaussian mixtures model. As a results this system shows pitch estimation quality similar to state-of-the-art systems, resulting in a possible validation of the biological hypothesis included in the model.