Back
Year
2024
Tech & Technique
Speech AI, Linguistics, PyTorch, Acoustic Features, Text Classification
Description
A research project in collaboration with Prof. Chiranjeevi Yarra to overcome the limitations of traditional text-only classifiers in low-resource settings.
This project introduced a novel hybrid topic classifier that combines both acoustic and textual features from audio data. The model demonstrated a remarkable **+40% accuracy improvement** over text-only and audio-only baselines, proving its effectiveness and applicability for real-world, resource-constrained scenarios.
This project introduced a novel hybrid topic classifier that combines both acoustic and textual features from audio data. The model demonstrated a remarkable **+40% accuracy improvement** over text-only and audio-only baselines, proving its effectiveness and applicability for real-world, resource-constrained scenarios.
My Role
As a research collaborator, my contributions included:
- ✅ Co-designing the hybrid model architecture.
- 🎶 Implementing the acoustic feature extraction pipeline.
- 📊 Setting up and running experiments to benchmark performance against baseline models.
- 📈 Analyzing results and contributing to the research findings.