COMPLEMENTARY CUES FOR SPEECH RECOGNITION

In this paper, we discuss the role of different complementary cues at three different levels: acoustic, aerodynamic and articulatory on a speech recognition task. The recognition system is a standard speaker dependant HMM recognizer. The corpus is based on CVCV sequences uttered by one speaker. It is shown that aerodynamic and articulatory cues used in combination with Mel frequency cepstral coefficients lead to a substantial increase of the system recognition score.