Reproduce “Qwen-Audio-3.1-ASR Technical Report” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chuanmeng Bian, Daren Chen, Peixin Chen, Zhigao Chen, Chong Deng, Zhiyun Fan, Zhifu Gao, Bo Gong and 40 more
The Qwen-Audio-3.1-ASR system extends instruction-controlled audio recognition with speaker-labeled transcripts containing words and time intervals, while its Next variant also supports audio descriptions, localization, and questions about sound. It reports 15.09% macro-average CER/WER across seven GigaSpeechBench languages, versus 22.50% previously, and 12.93% macro-average cpWER across four diarization test sets, versus 15.13% for MOSS-Transcribe-Diarize; these use approximately four-minute chunks, not hour-long recordings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.