Transsion Cascaded Framework for Speaker-Attributed Multilingual ASR
September 21, 2026
Transsion's submission for the MLC-SLM 2026 Challenge uses a cascaded framework combining DiariZen for diarization, Qwen3-Omni for multilingual ASR, and a CTC-based alignment model. The system generates speaker-attributed transcriptions by fusing diarization segments with timestamped text.
HOW THIS AFFECTS YOU
●
builderYou can study this cascaded approach to improve speaker-attributed transcription in multilingual conversational settings.