Spoken Hindi paired word recognition using probabilistic neural network.

Abstract

Automatic speech recognition system plays significant role in the development of real world applications. This paper demonstrated spoken Hindi (Indian national language) paired word recognition (SHPWR) system, which has been examined with the help of probabilistic neural network as a classifier. This type of network is a combination of radial basis layer and a competitive transfer function layer which picks the maximum probabilities as a final result. SHPWR is pattern recognition (PR) problem. A PR technique encompasses two fundamental tasks; description and classification. In the description process, features are extracted from the Hindi paired word templates, and probabilistic neural network is used as a classifier. For the experimental point of view 1000 Hindi paired word templates have been recorded from individuals with different environment condition, gender, and age groups. Speaker dependent SHPWR systems achieved close to 100% while the accuracy of speaker independent SHPWR systems is quite adequate.