15th International Congress of Phonetic Sciences (ICPhS-15)
In this paper we present a novel approach for articulatory features extraction. Our approach relies on an auto-segmental multilinear representation of AFs. Overlap and precedence relations between AFs on the different levels of the multi-linear representation can be extracted and then presented to a phonological parser for further recognition. This representation models co-articulation affect. We used parallel systems of multi-Gaussian Hidden Markov Model based recognisers to extract AFs classes. The statistical system was implemented using a novel modification of the HTK toolkit which allows it to perform multi-thread multi-feature recognition. Testing proved the overall performance is extremely promising. Among the highest accuracies achieved are 98% for vowels and 93% for rhotic sounds. Current work investigates interdependencies of extracting different feature types.
Bibliographic reference. Abu-Amer, Tarek / Carson-Berndsen, Julie (2003): "A multi-linear HMMs system for articulatory features extraction", In ICPhS-15, 739-742.