Emotion Recognition from Speech Signals Using Hybrid CNN Model

The fast growth of internet and communications networks has drastically enhanced data transport, allowing tasks like Speech Emotion Recognition (SER), an essential aspect of human-computer interaction, to be carried out. However, it is still challenging to recognize emotions in speech due to the dynamic and varied nature of speech signals. This work proposes a hybrid Machine Learning (ML) model integrating Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks based on enhancing the performance of SER. The CNN extracts the spectral features of the signal while the LSTM is capable of learning temporal dependencies in speech signals. Preprocessing involves the application of spectrograms and mel-frequency cepstral coefficients. Experiments performed on a public data set have revealed that the suggested hybrid model performs better than the CNN and LSTM models, with an impressive accuracy of 95%.The experimental results are promising for its applications in surveillance, autonomous driving, action recognition, and human-computer interaction.