Reproduce “LibriSpeech-Real-QA: A Benchmark for Streaming Question Answering” 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
Zewen Zhan, Hui Lu, Xueyuan Chen, Huimeng Wang, Yikai Huang, Xixin Wu, Shiyin Kang, Bin Xia and 1 more
The authors introduce LibriSpeech-Real-QA, a benchmark for measuring whether speech models answer when queried words occur in audio. They fine-tune Qwen2.5-Omni-7B to process audio chunks, choosing an answer or silence at each boundary. With 400 ms chunks and a one-chunk grace period, lexical detection reaches F1 0.988/0.968 on test-clean/test-other. These results concern LibriSpeech tasks, not unrestricted conversation or spoken replies.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.