University project
Gobbie, speech emotion recognition
Read the postAn 8-emotion speech classifier built on WavLM, HuBERT and wav2vec 2.0 embeddings. It reached 0.890 accuracy under speaker-disjoint cross-validation, up from a 0.465 baseline, and ran as a live microphone demo at UdeSA AI Fest.
2026, Python, PyTorch, Hugging Face Transformers, scikit-learn
