EUROSPEECH 2001 Scandinavia
Both for the transmission over GSM and IP networks, voice must be encoded at the originating end and subsequently decoded at the receiving end. This lossy coding produces a quality deterioration, that though acceptable for a human being, seriously affects the performance of Automatic Speech Recognizers (ASR) when they are not specifically designed for operating under those conditions. The authors have already introduced and tested a new robust front-end which improves ASR performances by simulating both networks environments. Here we complete some previous results including realistic GSM models, compare both scenarios and put forward and integrated scenario where mobile GSM devices require the services of an ASR facility situated into an IP network.
Bibliographic reference. Gallardo-Antolin, Ascension / Pelaez-Moreno, Carmen / Diaz-de-Maria, Fernando (2001): "A robust front-end for ASR over IP snd GSM networks: an integrated scenario", In EUROSPEECH-2001, 1103-1106.