Build an efficient Amazigh Speech Recognition using a CNN-CTC model

Abstract

This work focuses on the potential of Convolutional Neural Networks (CNN) in combination with Connectionist Temporal Classification (CTC) to develop an efficient Amazigh recognition system. The CNN-CTC model have a simple architecture, don’t need any alignment between the input and the output produces the most probable sequences of labels. In order to evaluate the performance of the model, a series of experiments were conducted. The experimental results demonstrate that the model attained a commendable accuracy of $93.5 \%$ when trained on a +30 -age category dataset with Dropout regularization.