A Wavelet Based Recognition System for Malayalam Vowels using Artificial Neural Networks

This work explores the use of a discrete wavelet transform, a feature extractor mechanism for speech recognition. Speech recognition is a fascinating application of digital signal processing offering unparalleled opportunities. The real-world applications deploying speech recognition and its implications can be varied across various fields. Speech recognition can automate many tasks that previously required hands-on human interaction. Accurate vowel recognition forms the backbone of most successful speech recognition systems. The vowel set of Malayalam, one of the South Indian languages, is used to create the database. A hybrid approach with discrete wavelet transforms and neural networks are used to form a system with improved performance. Daubechies wavelet is employed in this experiment. Features are extracted by using Discrete Wavelet Transforms (DWT). Training, testing and pattern recognition are performed using Artificial Neural Networks (ANN). The results show excellent overall recognition accuracy above 95%. The high accuracy obtained shows promising potentials of discrete wavelet transforms and neural networks in speech recognition.