Limited Vocabulary Speech Recognition

Abstract

The Automatic Speech Recognition (ASR) systems play an major role in recognizing spoken language, enabling seamless human-machine interaction. This paper proposes a new ASR model which is adapted for limited vocabulary scenarios that optimizes speech recognition accuracy and performance. This model uses a special datasets of human voice recordings containing a limited subset of vocabulary. Using advanced machine learning techniques, that includes deep neural networks and possibly convolution neural networks (CNNs), the proposed system aims to capture the variations in human speech patterns. Using a datasets of different human voice or speech samples, the ASR model goes through training, validation and testing stages to predict the corresponding text. This paper mainly Focus on a limited vocabulary allows greater accuracy in identifying and transcribing certain commonly used words.