Assessing random forest performance in low resource speech emotion recognition.

Area of Science:

• Human-Computer Interaction (HCI)
• Artificial Intelligence (AI)
• Speech Processing

Background:

• Speech Emotion Recognition (SER) is crucial for empathetic AI in HCI.
• Urdu is a low-resource language with limited SER research.
• Existing SER models often struggle with linguistic diversity.

Purpose of the Study:

• To evaluate the effectiveness of a Random Forest (RF) classifier for SER in Urdu.
• To identify key Mel-frequency cepstral coefficients (MFCCs) for emotion discrimination.
• To advance AI's emotional understanding in under-resourced languages.

Main Methods:

• Utilized Mel-frequency cepstral coefficients (MFCCs) for feature extraction.
• Employed a Random Forest (RF) classifier for emotion classification.
• Focused on three primary emotions: happiness, sadness, and anger.

Main Results:

• Achieved a validation accuracy of 94.53% for Urdu SER.
• Demonstrated the robustness of the RF classifier for this task.
• Identified significant MFCC features contributing to emotion differentiation.

Conclusions:

• The RF classifier is a viable and effective tool for SER in low-resource languages like Urdu.
• This research paves the way for more emotionally intelligent AI systems.
• Future work includes expanding emotion categories and exploring diverse datasets.