A system for digital speech data transmission is described. Quasi-stationary tracts of the speech waveform are represented by a fundamental period repeated a number of times depending on a periodicity index related to the correlation between successive speech intervals having the duration of a pitch. Moreover, a fundamental period is represented by its maxima and minima and their displacements in time. Nonstationary tracts are represented only by their maxima and minima and their position in time, while silence intevals are merely represented by a labeling symbol and their durations. In synthesis, at the receiver end of the channel, the interpolation between maxima and minima is performed in accordance with the results of a sort of sampling theorem, which is analogous to considering maxima and minima as samples of a function for which the time scale has been distorted. Spoken sentences of good quality are reconstructed with a transmission rate of about 4 kbit/s.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.