A deep neuro-fuzzy framework for speech emotion recognition.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Signal Processing

Background:

• Speech Emotion Recognition (SER) is vital for human-computer interaction, healthcare, and education.
• Existing methods like Adaptive Neuro-Fuzzy Inference Systems (ANFIS) and Deep Neural Networks (DNN) have limitations in handling high-dimensional speech data and interpretability.
• Robust SER techniques are needed for accurate emotion detection.

Purpose of the Study:

• To propose a hybrid deep neuro-fuzzy framework for enhanced Speech Emotion Recognition (SER).
• To combine the strengths of Deep Neural Networks (DNN) for feature extraction and Adaptive Neuro-Fuzzy Inference Systems (ANFIS) for classification.
• To address the limitations of ANFIS with high-dimensional data and DNN's lack of interpretability.

Main Methods:

• A novel framework integrating DNN and ANFIS is developed.
• The framework incorporates fuzzification, deep feature extraction, and defuzzification units.
• The proposed model was evaluated on three standard speech emotion databases: RML, SAVEE, and RAVDESS.

Main Results:

• The deep neuro-fuzzy framework achieved high accuracy in SER, reaching up to 97.95%.
• The proposed model demonstrated superior performance compared to standalone ANFIS, DNN, and pre-trained models.
• The framework effectively handles high-dimensional speech data and enhances interpretability.

Conclusions:

• The proposed deep neuro-fuzzy framework offers a robust and accurate solution for Speech Emotion Recognition.
• This hybrid approach shows significant potential for advancing SER research and applications.
• The framework's effectiveness is validated across multiple benchmark speech datasets.