asa-global-alld-best

Headline global ALLD (preference-aligned) model from the bachelor thesis Automatic Speech Assessment (Schmidt-Svejstrup & Lindahl, DTU 2026). In-domain MSE 0.202, BLEU 26.12. Fine-tuned from Qwen2-Audio-7B with the ALLD method. Code: https://github.com/DavidLindahl/automatic-speech-assessment

Downloads last month
4
Safetensors
Model size
8B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support