SLUE (Spoken Language Understanding Evaluation)

English SLU benchmark on natural (not read) speech — Phase-1 adds ASR, named-entity recognition, and sentiment annotations over subsets of VoxPopuli and VoxCeleb; Phase-2 adds dialog act classification, QA, summarization, and named-entity localization.

• Task: asr
• Languages: en
• Hours: 27.3
• Domain: spoken language understanding
• License: see source terms
• Homepage: https://asappresearch.github.io/slue-toolkit/
• Paper: https://arxiv.org/abs/2111.10367

Recommendation

Use SLUE to benchmark end-to-end and pipeline SLU systems on real-world English speech with consistent annotations — Phase-1 for short-utterance NER/SA/ASR, Phase-2 for longer-form discourse tasks. Licensing is inherited from upstream VoxPopuli (CC0) and VoxCeleb (research-only), so downstream redistribution must respect both source terms.

Getting the data

Obtain from the dataset homepage.

Phase-1 fine-tune split totals 27.3h (VoxPopuli 14.5h + VoxCeleb 12.8h). Phase-2 paper at https://arxiv.org/abs/2212.10525. Toolkit (MIT) at https://github.com/asappresearch/slue-toolkit; HF datasets asapp/slue and asapp/slue-phase-2.

Suggested processing

A recommended VoxKitchen pipeline ships in the repository at voxkitchen/templates/pipelines/asr-training-data.yaml — run it with vkit docker run.