A robust speaker dependent algorithm for isolated word recognition

The paper deals with a simple speaker-dependent (SD) isolated word recognition (IWR) system based on template-based pattern matching. Different algorithms for storing and calculating the distortion between models and examples of words to be recognised are analysed. More specifically, the paper proposes a new algorithm that enhances performance with a slight increase in computational load and the amount of memory needed to store the models as compared with a traditional VQ-based approach. The results obtained in tests are given in terms of recognition rate, using the TIMIT-46 database with various type of background noise and different SNRs.