Contribution of temporal fine structure information and fundamental frequency separation to intelligibility in a competing-speaker paradigm

Abstract

The speech reception threshold (SRT) for identifying a target speaker in a background speaker was measured as a function of the difference (F0sep) in fundamental frequency (F0) between the two speakers. The amount of original temporal fine structure (TFS) information in the mixed signals was manipulated by tone vocoding channels above a certain cutoff channel (CO). When the natural variations in F0 of both speakers were preserved, the SRT did not decrease with increasing F0sep, indicating that short-term differences in F0 can allow perceptual segregation of two speakers even when their F0s cross. When F0 variations were removed from both speakers, increasing F0sep led to decreased (better) SRTs. The decrease was greater for unprocessed signals than for fully tone-vocoded signals. However, the decrease was similar for unprocessed signals and for signals with original TFS below 1600 Hz, suggesting that most of the benefit from increasing F0 difference depends on the use of TFS information at lower frequencies. Adding original TFS information to channels centered above 1600 Hz produced roughly the same decrease in SRT as adding original TFS information to channels centered below 1600 Hz, suggesting a benefit from original TFS information apart from that related to differences in F0.