Neural Networks
| Hi-index | 0.00 |
Speech can be represented phonetically by a finite set of symbols called the phonemes of the language. Identification of these phonemes is the most important part of any speech recognition system. A work is described in this paper where an Artificial Neural Network (ANN) based approach is carried out to recognize consonant and vowel phonemes of two alphabet Assamese words. A Self Organizing Map (SOM) based segmentation algorithm segments the incoming speech signal into its constituent phonemes and from the SOM segmented phonemes the three constituent phonemes are identified by some sort of recognition algorithm based on two different ANN structures Probabilistic Neural Network (PNN) and Learning Vector Quantization (LVQ). Formant frequency of Assamese phonemes is used very effectively as a priori knowledge in the proposed algorithm. The SOM based segmentation algorithm shows distinct advantage in terms of recognition success rate in comparison to the conventional speech segmentation methods like windowing or Discrete Wavelet Transform (DWT).