GPU Accelerated Speech Recognition

Abstract

Speech recognition technology is one of the quickly developing advanced technologies of engineering. It has various applications in different zones and offers potential points of interest. There is a correlation to pre-requirement for speech recognition and machine learning which is that grammar classification and a method for extraction phonemes from utterances are required. For Speech architecture, three models are utilized in speech recognition to do the preparing like a phonetic dictionary, acoustic model, language model. There are four stages of the recognition process: analysis, feature extraction, modeling and matching. Deficiency factors were brought forward which is the dataset threshold and feature extracting method. The higher dataset produces a lower word error rate (WER) which gives more accuracy in the recognition process.