Comparative Study of Malay Speech Emotion Recognition Systems using AlexNet with Comparison to English Speech Emotion Recognition Systems

Cheang, Zi Jing (2025) Comparative Study of Malay Speech Emotion Recognition Systems using AlexNet with Comparison to English Speech Emotion Recognition Systems. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Cheang Zi Jing - FULL TEXT.pdf Restricted to Registered users only Download (5MB)

Abstract

Speech Emotion Recognition is a crucial area within human-computer interaction and artificial intelligence, aimed at identifying emotional states from speech. This research addresses the lack of robust SER systems for the Malay language, which is underrepresented in existing studies predominantly focused on major global languages. The objective is to develop a comprehensive Malay speech emotion database and a corresponding SER system utilizing the AlexNet model. Methodologically, the project involves recording emotional speech samples from 19 Malaysian volunteers in a controlled studio environment, adhering to the RAVDESS standards with a total of 1140 audio files. The dataset was expanded to 4560 samples through data augmentation techniques such as noise injection, pitch shifting, time stretching and time shifting. Feature extraction involved a combination of ZCR, Chroma_stft, MFCC, RMS and MelSpectrogram to effectively capture emotional characteristics in speech. To ensure usability, a simple and intuitive GUI was developed, allowing users to record or upload audio files and visualize waveforms. The research shows that the model achieved an accuracy of 91.67% with a macro and weighted F1-score of 0.92 by utilizing the self-recorded SuaraEmo databases.