Hugging Face's logo Hugging Face
  • Models
  • Datasets
  • Spaces
  • Buckets
  • Docs
  • Enterprise
  • Pricing
    • Website
      • Tasks
      • HuggingChat
      • Collections
      • Languages
      • Organizations
    • Community
      • Blog
      • Posts
      • Daily Papers
      • Hardware
      • Learn
      • Discord
      • Forum
      • GitHub
    • Solutions
      • Team & Enterprise
      • Hugging Face PRO
      • Enterprise Support
      • Inference Providers
      • Inference Endpoints
      • Storage Buckets

  • Log In
  • Sign Up

Datasets:
Scicom-intl
/
Whisper-Hallucination

Tasks:
Automatic Speech Recognition
Audio Classification
Modalities:
Audio
Tabular
Text
Formats:
parquet
Languages:
Malay
Indonesian
English
Size:
100K - 1M
ArXiv:
Tags:
whisper
hallucination
repetition
looping
asr-robustness
malaysian
Libraries:
Datasets
Dask
Polars
License:
Dataset card Data Studio Files Files and versions
xet
Community
1
Whisper-Hallucination
4.12 GB
Ctrl+K
Ctrl+K
  • 1 contributor
History: 32 commits
huseinzolkepliscicom's picture
huseinzolkepliscicom
wild: declare train/test splits
8126be4 verified about 2 months ago
  • bench
    Add baseline scores for whisper-large-v2 / large-v3 / large-v3-turbo over all 8 arms (11,852 clips each), plus the benchmark harness used to produce them 2 months ago
  • data
    wild: split into train (3,656) and test (4,673) by source recording about 2 months ago
  • external
    Whisper hallucination + repetition probes: reduplication and silence audio arms, 40,891-phrase lexicon, high-risk phrase targets, public Malaysian speech survey 2 months ago
  • lexicon_raw
    Whisper hallucination + repetition probes: reduplication and silence audio arms, 40,891-phrase lexicon, high-risk phrase targets, public Malaysian speech survey 2 months ago
  • licenses
    Add librispeech_test_clean (2,620 clips) as the English WER regression guard, matching the split arXiv:2609.04561 reports its cost on; document the raw-vs-filtered HR discrepancy 2 months ago
  • .gitattributes
    2.5 kB
    initial commit 2 months ago
  • README.md
    24.2 kB
    wild: declare train/test splits about 2 months ago
  • exclusions.json
    242 kB
    Mark all audio arms test-only: this is a benchmark, not training data. Add exclusions.json for building a disjoint training set. 2 months ago