MUSCAT: MUltilingual, SCientific ConversATion Benchmark

In three linesMUSCAT is a multilingual benchmark to evaluate Automatic Speech Recognition (ASR) systems on bilingual scientific conversations with code-switching. The dataset contains discussions between multiple speakers in different languages and proposes an evaluation framework beyond Word Error Rate (WER). Results show current ASR systems struggle with these challenges.

Summary generated by Claude — human-verified