Research of Malay Speech Emotion Recognition Systems Using VGG16 with Comparison to English Speech Emotion Recognition Systems

Lee, Wei Gen (2025) Research of Malay Speech Emotion Recognition Systems Using VGG16 with Comparison to English Speech Emotion Recognition Systems. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Lee Wei Gen - FULL TEXT.pdf Restricted to Registered users only Download (3MB)

Abstract

The demand for human-interactive applications has increased interest in machine learning algorithms for emotion detection from speech. While emotion recognition algorithms are commonly applied to languages like German, English, and Spanish due to the availability of datasets, speech emotion datasets for Malay are limited. This study explores emotion recognition using spoken Malay data, where a dataset was created by inviting participants to record Malay speech in Audio Lab. The recordings were annotated with eight emotional categories: neutrality, calm, sadness, surprise, anger, happiness, disgust, and fearfulness. Spectral features, specifically Chroma_STFT, were extracted and used for classification with the VGG16 model. The data were compared and analyzed across three models that employed different feature extraction techniques. The development of this Malay Speech Emotion Recognition (SER) system is crucial for advancing human-computer interaction in the Malay-speaking community. By utilizing VGG16, the system classifies emotions such as happiness, sadness, and anger with high accuracy. The system achieved an overall accuracy of 98%, with a macro average of 97% and a weighted F1-score of 98%, demonstrating its effectiveness using the self-recorded SuaraEmo database. This research contributes to emotion-aware technology for the Malay language.