A deep neuro-fuzzy framework for speech emotion recognition

Speech Emotion Recognition (SER) is crucial in fields like healthcare and education, requiring robust techniques for accurate emotion detection. This paper proposes a deep neuro-fuzzy framework combining Deep Neural Networks (DNN) and Adaptive Neuro-Fuzzy Inference Systems (ANFIS). It includes fuzzification, deep feature extraction, and defuzzification units, enhancing SER accuracy while addressing ANFIS limitations with high-dimensional data and DNN's lack of interpretability. The scheme's productivity on three standard speech databases is appraised: RML, SAVEE, and RAVDESS. The results indicate that our framework outperforms ANFIS, DNN, and pre-trained models, achieving up to 97.95% accuracy and showing strong potential for future SER research.