キーワード：音声知覚、第二言語、知覚学習、分析合成音
Japanese speakers' identification of English /r/ and /l/ using original and synthetic speech, generated by the STRAIGHT alogorithm, was compared. No significant difference was observed, demonstrating the potential for using synthetic speech generated by STRAIGHT in second language(L2) training studies. As a first attempt to find an effective stimulus manipulation for speech training experiments, we adapted the original stimuli by lengthening the duration of /r/-/l/ and/or enhancing F3. Performance on these manipulated stimuli showed no significant difference from performance on synthetic stimuli without such manipulations, suggesting the neccesity of further effort in finding more effective L2 training stimuli.
Key Words: speech perception, second language, perception learning, synthesized speech