Sentiment Analysis of Dialectal Speech: Unveiling Emotions through Deep Learning Models
Loading...
Date
2024-04
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The British University in Dubai (BUiD)
Abstract
Dialect Speech Sentiment Analysis is an evolutional field where machine learning algorithms are utilized to detect emotions in spoken language. However, Arabic, particularly Egyptian Arabic, remains underrepresented, lacking a dedicated speech sentiment database. This thesis introduces a novel dataset specifically created for sentiment and emotion detection in the Egyptian Arabic dialect, generated from publicly available YouTube videos and annotated across seven emotional categories: anger, happiness, sadness, disgust, fear, romantic, and neutrality. The proposed solution involves leveraging a multi-stage machine learning pipeline that first extracts spectral features such as MFCC and mel spectrograms from acoustic speech waves using Fourier transformation. These features are then classified using a range of Deep Learning Models, including convolutional neural networks (CNN), bidirectional long-short-term memory (BI-LSTM), gated recurrent units (GRU), and Artificial Neural Networks (ANNs). A key contribution of this work is the development and evaluation of hybrid Deep Learning Models that combine CNN-BI-LSTM, CNN-GRU, GRU-CNN, GRU-BI-LSTM, and GRU-ANN architectures. The results demonstrate the superiority of the hybrid CNN-BI-LSTM model, achieving an accuracy of 93%, significantly outperforming individual deep-learning models such as CNN (87%) and BI-LSTM (83%). Additionally, the GRU-CNN hybrid model attained a notable accuracy of 90%. These findings establish the robustness and effectiveness of hybrid architectures in enhancing emotion recognition accuracy in Arabic speech data, presenting a novel approach for Arabic dialect sentiment analysis.
Description
Keywords
sentiment analysis, speech recognition, Arabic speech recognition, deep learning methods, hybrid deep learning approaches