2606.26819

Total: 1

#1 FBK's Long-form SpeechLLMs for IWSLT 2026 Instruction Following [PDF] [Copy] [Kimi] [REL]

Authors: Zhihang Xie, Marco Gaido, Sara Papi, Matteo Negri, Luisa Bentivogli

This paper describes our submission to the IWSLT 2026 Instruction Following shared task. SpeechLLMs are developed for both short-form and long-form speech instruction following under constrained settings. For the short track, strong performance is achieved on MCIF, with a SIFS score of 2.0708. For the long track, three speech segmentation methods are explored, and the HIFS score is introduced to account for unstable long-form generation. Experimental results show that fixed 30-second segmentation provides the most robust long-form performance, achieving the highest HIFS score of 2.0663. Further analysis shows that hallucination mainly manifests as repetitive insertions in generated outputs, substantially affecting ASR and SSUM, while short-form capabilities are largely retained after long-form extension.

Subject: Computation and Language

Publish: 2026-06-25 10:01:45 UTC