Comparative study of CNN, LSTM and hybrid CNN-LSTM model in amazigh speech recognition using spectrogram feature extraction and different gender and age dataset

The field of artificial intelligence has witnessed remarkable advancements in speech recognition technology. Among the forefront contenders in this domain are Recurrent Neural Networks (RNN) and Convolutional Neural Networks (CNN). However, when it comes to their efficacy in recognizing the Amazigh language, which network reigns supreme? This article presents a comparative study of Convolutional Neural Networks (CNN), Long Short Term Memory (LSTM), and a hybrid CNN-LSTM model in the context of speech recognition systems. The main objective of this work is to identify which network architecture delivers the best performance for recognizing the Amazigh language. Our research stands out as one of the first to develop and compare three distinct deep models specifically for the Amazigh language, effectively addressing the challenges posed by a low-resource language. Through a series of rigorous experiments and evaluations conducted using the Tifdigit dataset, the study’s results underscore the superiority of CNNs in Amazigh speech recognition with 88% of accuracy when the CNN trained with female category dataset.