Effect of sentence length and exposure on the intelligibility of synthesized speech

Abstract

This study investigated the effect of the length of sentences and the amount of exposure to synthesized speech on the intelligibility of a popular text-to-speech (TTS) output device. The results show that sentences with a mean word length of 11 words are as intelligible as sentences with a mean length of approximately 5 words. In augmentative and educational applications of TTS output, it appears that relatively long sentences may be used without fear of adverse impact on intelligibility. The study also shows that there is a large and linear increase in intelligibility with exposure to synthesized speech up to the first 5 sentences heard by listeners within an experimental session. Thereafter, there is a less marked and variable amount of increase in intelligibility up to the next 15 sentences heard. Exposure beyond the first 20 sentences does not appear to contribute to increased intelligibility within a session. Exposure to the Echo II synthesizer over a 3-day period showed that subjects improve signifi...