Idiap Research Institute
Centre du Parc
Rue Marconi 19
PO Box 592
CH - 1920 Martigny
Switzerland
T +41 27 721 77 11
F +41 27 721 77 12
Contact form
RECOD - Low bit-rate speech coding
It is well known that bit-rates in a communication channel can be reduced by increasing the semantic level of the data being transmitted. In particular, if technology for automatic speech recognition (ASR) and text to speech synthesis (TTS) are available, then low bit-rates can be achieved by transmitting the output of the ASR. This output is then re-synthesised at the receiver, constituting a communication channel. Building on this simple scenario, there is a choice of whether to recognise and resynthesise (amongst others) phone-like units or words. The former leads to an unconstrained vocabulary, whereas the latter, with a constrained vocabulary, leads to the lowest bit-rates.