Prosodic Parameters of French in a Speech Recognition System

Abstract

The importance of integrating prosody into different speech processing systems is nowadays widely acknowledged. Prosodie parameter modelling was first carried out for text-to-speech synthesis, since this speech processing technique could not work without prosody (Emerard, 1977; Klatt, 1979). It appeared to speech researchers that prosody would also be helpful in speech recognition (Carbonell, Haton, Lonchamp & Pierrel, 1982; Waibel, 1987; Ljolje & Fallside, 1987; Wang & Hirschberg, 1992). However, the way to use prosodie parameters in a speech recognition system is less straightforward than in text-to-speech synthesis. A good predicting model in speech recognition has to forecast different varieties of speaking styles while in speech synthesis one correct speaking style prediction, appropriate to a given application, is sufficient.KeywordsFalse AlarmSpeech RecognitionSpeech SignalAutomatic Speech RecognitionSpeech Recognition SystemThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.