Output details
11 - Computer Science and Informatics
University of Edinburgh
Analysis of Speaker Adaptation Algorithms for HMM-based Speech Synthesis and a Constrained SMAPLR Adaptation Algorithm
<22> Originality: Paper develops a set of novel algorithms for speaker adaptation for statistical speech synthesis, resulting in an innovative approach to building new synthetic voices from small amounts of speech data.
Significance: Prior to this, the development of a new voice required many hours of carefully annotated speech recordings from a single speaker. The new algorithms enable a large number of different synthetic voices to be easily constructed. Awarded the Acoustical Society of Japan’s Itakura Prize.
Rigour: The proposed approach was evaluated using the standard Japanese speech corpus for ASR and TTS research, using both objective and subjective measures.