Skip to content

Dataset · microsoft

NOTSOFAR-1 single-device meeting recordings

microsoft/NOTSOFAR

Audio data and baseline solutions from Microsoft for remote meeting speech transcription.

What's inside

Created for the NOTSOFAR-1 challenge (part of CHiME-8) focused on speech recognition for meetings recorded on a single device. It includes between 10k and 100k audio recordings, along with baseline system code and validation split results.

What you can do with it

  • Train and test ASR systems for multi-user conference calls
  • Develop speaker diarization models in challenging acoustic conditions
Size
10K–100K rows
Downloads this month
32.2K
Likes
28

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("microsoft/NOTSOFAR")