A Symphony of Sentiments using Log-Melspectrogram Techniques for Emotional Classification

This research paper investigates the topic of audio emotion categorization via a novel approach involving log-Mel spectrogram with augmentation. Though it is much lower than MFCC, the research reveals a reasonable accuracy rate of 63%. It also draws attention to the great room for improvement, though. The work uses a complex model, a 2D CNN with VGG19 architectural resemblance. The audio properties are handled in this paradigm as a 30x216 pixel image. Many features are extracted by specialised methods utilising datasets like TESS, CREMA-D, SAVEE, and RAVDESS. It is stressed how crucial it is to quickly detect emotions via audio clues. The results of the 2D CNN model are promising and can outperform traditional methods. Faster interpretation is made possible by using a specific class to hold result properties. This work closes the gap between image-based and audio-based techniques in emotion analysis by improving the growing field of audio emotion classification and providing chances for accurate and efficient identification.