Key points are not available for this paper at this time.
The proposed research presents a novel approach to Speech Audio Emotion Recognition, integrating a Multi-Layer Perceptron (MLP) classifier and advanced feature extraction methods. With the growing importance of emotion recognition in speech, particularly in human-computer interaction and sentiment analysis, the study utilizes the RAVDESS dataset and explores features like Mel Frequency Cepstral Coefficients, Mel Spectrogram, Zero-Crossing Rate, Spectral Rolloff, Spectral Flux, and Chroma Vectors. The methodology, encompassing exploratory data analysis, preprocessing, and model training, achieves a notable accuracy of 89.31% on both training and testing sets. Notably, the inclusion of features such as Spectral Roll off and Chroma Vectors proves pivotal in capturing nuanced non-linear emotion patterns. Real-time audio testing further validates the model's effectiveness, demonstrating practical applicability. The developed system not only significantly contributes to advancing Speech Audio Emotion Recognition but also holds promise for real-world applications, enhancing human-computer interaction and sentiment analysis across various domains. In summary, this research introduces a robust and accurate emotion recognition system, leveraging the MLP classifier and diverse feature sets, with potential implications for a range of real-world scenarios.
K et al. (Wed,) studied this question.