Acoustic modeling in Automatic Speech Recognition - A Survey

Automatic speech recognition (ASR) systems are in evolutionary process. These systems use models which work at two stages, namely the parameterization of the input noise signal followed by training and testing of the features using any classification technique. Researchers have proposed a variety of acoustic models to accomplish this complex task. In this paper, we represent overview of Hidden Markov Model (HMM), Deep Neural Networks (DNNs) and Convolutional Neural Network (CNN) based models, which are the backbone of ASR systems. CNN is advanced method which normalizes speaker variance by using local filters in convolution layer. CNN architecture has advance features like weight sharing, local filters and pooling etc. This paper will also summarize the details of improvements in CNN s using different convolution techniques.