Separation of simultaneous voices
Richard J. Stubbs, Quentin A. Summerfield · The Journal of the Acoustical Society of America · 1987
People with sensorineural hearing impairments find the task of separating speech from interfering noises particularly difficult. For this reason, the ideal hearing aid would selectively amplify a target voice while attenuating competing voices and other background noises. In exploring possible noise-reduction strategies for future aids, two algorithms for separating the voices of talkers who are speaking simultaneously, with periodic excitation at similar overall intensities, are evaluated. Both approaches are pitch based and exploit the regularity in the harmonic structure of voiced speech. The first involves attenuating the harmonics of the competing voice via a cepstrumlike representation. The second method is derived from the procedure for harmonic selection [T. W. Parsons, J. Acoust. Soc. Am. 60, 911–918 (1976)]. Perceptual evaluation of the two processing methods, in a test involving the separation of concurrent vowel pairs synthesized on fixed fundamental frequencies, has demonstrated an increase in performance for both normal-hearing and hearing-impaired subjects. The results of experiments involving the separation of simultaneous sentences containing a high proportion of voiced excitation and varying fundamental frequencies will also be discussed.