Dari Speech Classification Using Deep Convolutional Neural Network

Recently, speech recognition is one of the most advanced research topics in the world. Many recent research papers have proven the power of deep neural networks in speech recognition systems. The main purpose of this paper is to identify isolated words in Dari speech using deep learning algorithms. This research is one of the new studies in Dari speech recognition and focuses on one-word speech recognition. This collection uses our audio files as a database because there were no Dari language databases on the market at that time. In this paper, the Convolutional Neural Network (CNN) is implemented to detect automatically isolated words in Dari. Besides, it uses Mel frequency coefficients (MFCC) to learn the representation of features during training. This model achieved 88.2% in the test set. The results show that the model can predict samples of words seen during training with high accuracy. However, it is somewhat trying to generalize terms outside the scope of training data and very noisy examples.