BSC_IWSLT26_IF_SHORT_unconstrained_contrastive

Data conditions: unconstrained

Model architecture: direct (mhubert-base-25hz speech encoder and SalamandraTA-7B-wmt25 as the LLM backbone)

Track: Short