Theory of Modulation Frequency Analysis and Modulation Filtering, with Applications to Hearing Devices

Theory of Modulation Frequency Analysis and Modulation Filtering, with Applications to Hearing Devices Steven Marco Schimmel Chair of the Supervisory Committee: Professor Les E. Atlas Electrical Engineering Natural signals such as speech and music can be faithfully represented by low frequency modulators which modulate higher frequency carriers. In this work, a principled signal processing framework for modulation frequency analysis and synthesis is developed. Its core components are a filterbank that separates a broadband signal into narrowband subbands, and a carrier estimator or envelope detector that decomposes each subband into a carrier and a modulator. A distinction is made between “incoherent” envelope detectors, such as the magnitude or the Hilbert envelope, and “coherent” carrier estimators, such as the instantaneous frequency carrier. Several limitations of incoherent detectors are demonstrated, and three coherent carrier estimators are proposed and evaluated. Practical aspects of modulation filtering such as modulation filtering effectiveness and signal reconstruction from a modified time-frequency representation are also addressed. The modulation analysis framework is applied to the problem of talker separation in the context of hearing devices such as hearing aids and cochlear implants. It is demonstrated that two co-channel talkers are largely non-overlapping in the modulation frequency domain. In preliminary experiments, talker separation is achieved with a supervised incoherent modulation masking approach, and an optimal coherent modulation filtering approach. However, both approaches require more information about the target talker than would be available in hearing device applications. Therefore, a target talker enhancement technique is proposed which requires only knowledge of the target talker’s fundamental frequency range. It uses incoherent modulation analysis to determine coefficients for a time-varying acoustic filter that enhances the target talker and suppresses an interfering talker, whose fundamental frequency range must also be known. Several weaknesses of this incoherent talker enhancement technique are pointed out, and a novel coherent modulation filtering approach to target talker enhancement is proposed. The coherent approach uses custom implementations of the carrier estimator and the modulation filter, in order to be more robust to noise and free of modulation filtering artifacts. The coherent approach to talker enhancement is evaluated in a subjective listening test on normal hearing subjects over a cochlear implant simulation and on hearing aid users. The test shows a modest improvement in intelligibility for coherently enhanced speech.