Monaural echo suppression: Detectability of a delayed signal based on vowel formant structure

Abstract

In order for echo suppression to occur, the ear must be convinced that the echo makes a reasonable acoustic match with the direct sound. It is plausible that the potential acoustic cues used to perform this task are similar to those used for auditory grouping, and include spectral location of vowel formants. An experiment was performed to investigate the role of vowel formant structure in monaural echo suppression. In a 2IFC adaptive procedure, the detectability of a masked signal, consisting of the second formants of two synthetic vowels, presented successively, was measured. The masker consisted of the first, third, and fourth formants of the same two vowels, repeated, so that it had four alternating segments. The signal either preceded the second segment of the masker, or was delayed relative to that segment. When the vowel order of the signal matched that of the second and third segments of the masker, the delayed signal was less detectable than when the vowel order was reversed in the signal. Likewise, when the vowel order of the signal matched that of the masker, the delayed signal was less detectable than the leading signal. These data suggest that the auditory system is able to group vowel formants based on expected spectral location and use that for echo suppression. [Work supported by AFOSR.]