Speak like a native

Speaking rhythm plays an important role in speech production and the perception of non-native languages. However, conventional techniques are insufficient to control speaking rhythm. In this study, we invented a novel method for extracting the speaking rhythm from speech signals using a non-negative temporal decomposition (NTD) and controlled the speaking rhythm using the method. This algorithm decomposes a speech spectrogram into a set of temporally overlapped phoneme-dependent event functions and corresponding event vectors under speech-specific restrictions. We found that the speaking rhythm can be converted by modifying the obtained event functions. We hope that this technique will alleviate the burden involved in communication with non-native languages.