Effects of Vocoder Processing on Speech Perception in Reverberant Classrooms

Channel vocoders have been widely used as acoustic models for current vocoder-centric cochlear implant (CI) signal processing strategies. Previous studies found that 4- to 8-channel vocoded speech in normal hearing (NH) subjects can derive comparable recognition scores to CI subjects who may have 8 to 22 frequency channels. The reasons for this overestimation might include 1) classic vocoders preserve natural intensity dynamic range within the envelopes rather than the much narrower range in CIs and 2) classic vocoders cannot simulate the effect of electric pulse rate. This study presents a novel vocoder based on a direct electric-pulse to acoustic-pulse mapping (i.e., electrodogram to spectrogram) to deal with the overestimation problem. The effects of the proposed vocoders with 22 and 16 channels on speech intelligibility in two real-measured classrooms were tested in NH listeners. Results showed that the proposed vocoders are more sensitive to changes of reverberant environment, like the previously reported actual CI results, than the classic ones, which implies that the new vocoding method could be a better alternative for acoustic modelling of current CI processing.