EAA Fusion Head (HuBERT + WavLM → Gemma-3-270M)
This repository contains the fusion/regression head and config for the EAA system.
Adapters (same overall model):
- HuBERT LoRA adapter:
marccgrau/EEA_hubert_adapter
- WavLM LoRA adapter:
marccgrau/EEA_wavlm_adapter
- Gemma-3-270M LoRA adapter:
marccgrau/EEA_gemma3_270m_adapter_hubert_wavlm
It fuses HuBERT & WavLM sequences via dual cross-attention, reduces to N tokens,
and conditions Gemma (via LoRA) to regress an aggression score in [1..10].
