BAAI's CS-Dialogue lands on Hugging Face: 104 hours of code-switching speech
AdinaYakup · x · 2026-11-04
BAAI's CS-Dialogue dataset is now live on Hugging Face (paper arXiv:2502.18913): 104 hours of spontaneous Mandarin-English code-switching dialogues from 200 proficient English speakers in 100 two-person conversations, spanning Mandarin-only, code-switching, and English-only segments, with character-level manual transcriptions and annotations under a CC BY-NC-SA 4.0 non-commercial license, built to advance code-switching ASR research.
More from Research
- Neuropixels probes record hundreds of neurons at once, reshaping neuroscience — StartupYou · 2026-11-04
- Paper "Kernel-Gradient Drifting Models" accepted to NeurIPS 2026 — LucaAmb · 2026-11-04
- Weight-Space Programming: GPT-6 edits just 0.016% of Qwen3.5-0.8B weights, boosting math up to 6x — HaggaiMaron · 2026-11-04
- LoCo: local preference learning cuts reasoning tokens by up to 45% without accuracy loss — pmttyji · 2026-11-04
- Linear RNNs can recognize a periodic language under exact arithmetic — lambdaviking · 2026-11-04
- ICLR Weight Space Learning Workshop opens submissions, spotlighting the field's core research — HaggaiMaron · 2026-11-04