Ultra low bit rate voice coding
M.J. Ovens · 2000
High frequency (HF) radio is used for long haul or extended range communications in many situations. Under stressed HF channel conditions, the supportable data rate falls below that required by existing low bit rate speech coding algorithms. This paper presents research undertaken at DERA Malvern on the development of a real-time speech coding system which utilises automatic speech recognition (ASR) and synthesis technologies to achieve speech coding at data rates below 300 bps. A continuous speech recogniser is used to transcribe incoming speech as a sequence of sub-word units, termed acoustic segments. Prosodic information (pitch and duration) is combined with segment identity to form a serial data stream suitable for transmission. A parallel formant speech synthesiser is used to synthesise the speech at the receiver, using models trained to a particular talker's voice to establish talker characteristics.