Authors : T. Santhanam , M. Nachamai , M. Muthuraman and C.P. Sumathi
Abstract: Artificial Neural Networks (ANN) is one of the approaches used in acoustic modeling. ANNs are better suited for handling complicated tasks. They are used to recognize speech and also to handle even low quality, noisy and speaker independent data with an improved efficiency i.e., ANN has displaced the most frequently employed Hidden Markov Model (HMM) for speech prob lems in all aspects in general and scalability in particular. Scalability in ANN is very less compared with HMM when provided with huge a mount of training data. This study makes use of ANN for classification of phonemes and style (stressed speech or neutral speech). Better results are obtained by considering only the pitch contour, one among the many base features of speech, as input to the network than giving multiple inputs. The Gaussian-kernel estimator function capable of mapping the data into a high dimensional space takes care of the scalability feature of ANN. This approach has resulted in enhanced recognition rates of speaking style.
T. Santhanam , M. Nachamai , M. Muthuraman and C.P. Sumathi , 2007. Stressed/Neutral Speech Classification Using Gaussian Support Vector Machines. International Journal of Soft Computing, 2: 335-338.