The Hmm Based Amazigh Digits Audiovisual Speech Recognition System
DOI:
https://doi.org/10.17762/msea.v71i4.773Abstract
In this paper, we present an Amazigh audio-visual speech recognition system that combines the information coming from the audio and visual modalities. The proposed system is considered, as far as we know, the first audio-visual system that uses Amazigh language. We develop each subsystem in different platforms. In order to building a visual subsystem, we extract the features from the region of the mouth using DCT to be modeled using Hidden Markov Models (HMM).Whereas, the audio subsystem is based on the Carnegie Mellon University Sphinx tools based on HMM. The two sub-systems use the AmDigit_AVSR (Amazigh Digit _ Audio-visual Speech Recognition System) database. The combined system obtained best performances of 93,99 % using “OR” based-rules. Our experiments show that the combination of the visual and acoustic information improves the performance of speech