Towards a variable frame rate speech transmission system with frame selection by time-domain segmentation - a status report

The results of a time domain segmentation algorithm are applied to a variable frame rate (VFR) transmission system. Three speech signal parameters (short-term absolute average of the speech signal, absolute average of the differenced signal, and the ratio of these two) are measured pitch-synchronously and interpolated to short frames of 5 to 10 ms duration. From these frames, a distance measure is evaluated leading towards a "segment length function" which indicates the time distance from a given frame to exceed a parameter change threshold. This function is directly converted into a local interval sequence for the frames of a speech transmission system. The procedure has been successfully applied to a transmission system which uses a pitch-synchronous time-domain Karhunen-Loève expansion (KLE) of the speech signal.