Implicit Measurement of Intelligibility of Male and Female Voice Text-to-speech (tts) Synthesis in Noise Using a Phoneme Detection Task

ABSTRACT: Given the increasing application of TTS synthesis in commercial and clinical settings, there is a need to develop methods of evaluation from the user’s perspective. An experiment is reported that compares the effect of two factors, voice gender and signal quality, on the intelligibility of three TTS systems from the user’s point of view. It was hypothesised that male voiced TTS would be more intelligible than female voiced TTS and that low quality signals would reduce intelligibility. Accuracy scores were obtained from a phoneme detection task using Semantically Unpredictable Sentences (SUS). The results indicate that an interaction between voice gender and signal quality is dependent on the TTS system. The results suggest that intelligibility, from the user’s point of view, is modulated by several factors. Consequently there is a need to evaluate TTS systems in the context of commercial requirements and specific applications.