Automatic Speech Recognition trained with Convolutional Neural Network and predicted with Recurrent Neural Network

An important method for interacting between humans and computers is Automatic-Speech-Recognition (ASR). This technique enables a computer to recognize the word (acoustic) spoken into a microphone and convert it into text or other suitable formats for further processing. The w i d e applications using Convolutional Neural Networks (CNNs) towards Voice Recognizing systems becomes widespread. This study investigates the use of deep CNNs to detect potentially mispronounced English phonemes. Machine learning models built on deep learning have produced noteworthy outcomes in a variety of vision-related applications, including Speech Recognition. In this work, Recurrent Neural Network (RNN) is used to process feature maps from CNN architecture trained on Speech Command dataset. The simulation results reveal that the proposed CNN architecture with RNN performs significantly, with an accuracy rate of 91.81%and a trained Word Error Rate (WER) of 12.4% on comparing with existing works.