Speaker Independent Phonetic Recognition Using Auditory Modelling and Recurrent Neural Networks

Two speaker independent speech recognition experiments, regarding the automatic discrimination of the Italian alphabet I-set and E-set , two very difficult Italian phonetic classes, will be described. The speech signal is analyzed by a recently developed joint synchrony/mean-rate auditory processing scheme and a fully-connected feed-forward recurrent BP network was used for the classification stage. The achieved speaker independent mean recognition rate was 65%, for the I- set and 88% for the E-set showing rather satisfactory results given the difficulty of both tasks.