A novel vocoder concept based on discrete time acoustic tubes

A novel vocoder concept is presented which is based on discrete time equivalents of the uniform acoustic tube. Recently proposed models of lossy sections of different lengths are cascaded to reprensent the vocal tract. Results comparable to a standard 12-section LPC-vocoder are achieved cascading typically 5 tube sections of different lengths. Major savings are derived in terms of the hardware complexity during speech synthesis and in term of the data rate during speech transmission. The concept can also be used for modeling the nasal tract by interconnecting a discrete time acoustic tube system to the vocal tract model via a proper adaptor. Till now losses of the low-loss-type are considered. The analysis part of the vocoder is based ona standard autocorrelation LPC-analysis followed by an additional approximation stage yielding a reduction of the number of tube sections. The performance test of the proposed vocoder is done both by subjective evaluation and by comparison in the spectral and the acoustic tube domain.