Publication

Abstract :

This paper mainly focuses on developing a word recognition system using the CNN structure. Several advancements have been made in the Automatic Speech Recognition (ASR) technology that enables the machine to understand the natural language. The main constrain rise is the nature of the input speech signal, which makes it difficult to retain the original information. The noisy speech signal is initially passed through the pre-processing stage and converted to the spectrogram to extract the feature. To extract the features, these spectrograms are fed to CNN's layers to feature extract and then train the model. The vectors are now cross-matched at the testing phase, and the maximum close weighted value from the fully connected layers leads to the output. The system performs with an efficiency of 88.20% in a non-isolated environment.