Speech Emotion Recognition

Problem

Speech emotion classification must account for differences in speakers, audio quality, and emotional expression.

Architecture

Audio preprocessing → MFCC feature extraction → neural-network classification → emotion prediction and evaluation.

Challenges

• Prepared audio samples and extracted MFCC features
• Explored convolutional and sequential model architectures
• Considered class balance and speaker variability during evaluation
• Compared model performance using held-out data

Benchmarks

The project references speech emotion datasets including RAVDESS and CREMA-D. Report accuracy and latency only when supported by the final evaluation results.

Lessons Learned

• Consistent preprocessing is essential between training and inference
• Audio feature quality can strongly affect classification performance
• Evaluation splits should account for speaker variability where possible