Looking into the past: Power spectral representation of periodic signals, sampling theories and fundamental frequency estimation for remaking speech Prof. Hideki Kawahara Auditory Media Laboratory, Department of Design Information Sciences, Faculty of Systems Engineering, Wakayama University Old ideas shed new light on underlying technologies for remaking speech.A speech analysis, modification and resynthesis framework STRAIGHT was introduced in 1996 by revising the channel VOCODER made by Homer Dudley in 1939. Since then, STRAIGHT has been widely used in speech perception research, speech synthesis and speech manipulations. However, it was computationally heavy and its theoretical basis were not well established. Recently, a new procedure for power spectral estimation of periodic signals was introduced and combining it with a reformulation of sampling theory, consistent sampling enabled complete reformulation of STRAIGHT and resulted into TANDEM-STRAIGHT. That is theoretically transparent and computationally efficient. Furthermore, zero-crossing of fundamental component, a method employed in remaking speech in 1939, yielded a new F0 extraction algorithm that supersedes state of the art F0 extractors in terms of speed and accuracy. "Basso continuo" of these will also be discussed.
Cite as: Kawahara, H. (2008) Looking into the past: Power spectral representation of periodic signals, sampling theories and fundamental frequency estimation for remaking speech. Proc. International Symposium on Chinese Spoken Language Processing
@inproceedings{kawahara08_iscslp,
author={Hideki Kawahara},
title={{Looking into the past: Power spectral representation of periodic signals, sampling theories and fundamental frequency estimation for remaking speech}},
year=2008,
booktitle={Proc. International Symposium on Chinese Spoken Language Processing}
}