As part of my research into voice quality conversion and swapping, I attempt accent swapping between a British RP speaker and an American speaker.Analysis/synthesis is carried out on the same, or similar, utterances spoken by both speakers. The technique employs Dynamic Time Warping (DTW) of the utterances, followed by pitch-period-based Linear Predictive Coding (LPC).
LPC modelling attempts to separate the speech/voice production system into the glottal source and the vocal tract filter. Although the physiology and behaviour of the vocal tract of any speaker will be unique and idiosyncratic, the initial aims are to tap the individuality of the glottal source. The sources and filters are then swapped to produce an utterance that sounds like one speaker but with the accent of the other.
Results of attempts to improve the modelling through short-frame analysis during the closed phase of the glottis are also discussed.
To download this paper, please return to Proceedings of the 1998 Postgraduate Conference