Intelligibility of Vocoded and Waveform Speech for Native and Non-Native Speakers of English

Abstract

This study applied Signal Detection Theory investigating the inconsistent findings from previous studies of intelligibility of vocoded speech for native and non-native speakers of English. Also examined was the intelligibility of waveform speech (uncompressed, wave-formatted speech) in several recording levels. The intelligibility of vocoded speech was significantly worse than waveform speech, with intelligibilities significantly greater for native than non-native speakers. Mean difference of vocoded speech intelligibility between native and non-native speakers was not statistically greater than that of waveform speech, but there was a strong trend toward interaction. Under the limitation of a fixed transmission bandwidth, a doubled level of resolution (sound amplitude) produced a higher intelligibility than a doubled level of sampling rate (frequency range). The characteristics of waveform speech significantly affected intelligibility for non-native speakers, but not native speakers. Vocoders with a higher bit rate may be required for safety and reliability reasons if the end users include non-native speakers.