HMM-based Speech Synthesizer for Easily Understandable Speech Broadcasting

In this paper, we report an improvement to our proposed easy-to-understand public-address system. This system broadcasts synthesized speech, which is parsed from speech-recognized text. In order to improve the performance our system, we implemented a Hidden Markov Model (HMM) based speech synthesis system dependent on the input speaker's speech. As a result, the performance of the speech synthesizer did not depend upon the number of learning sentences to generate the model, and a constant subjective quality value was measured using the MUSHRA evaluation method. We found that speech synthesized by learning 50 sentences exhibited sufficient performance.