Dataset · kensho
Large English corpus for speech recognition (ASR)
kensho/spgispeech
Millions of English audio and text pairs for training and testing automatic speech recognition systems.
What's inside
The dataset contains between 1 and 10 million records with audio and text modalities in English for the automatic-speech-recognition task. The presented card fragment contains only a standard table of contents without a description of sources, field structures, and licensing terms.
What you can do with it
- Training and fine-tuning ASR models for the English language
- Checking speech transcription quality on large volumes of data
- Tasks
- Automatic speech recognition
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 3.1K
- Likes
- 42
- Access
- on request on Hugging Face
Models trained on it
- mradermacher/tiny-audio-GGUF 434
- oruk/oruk-loop-153MAutomatic speech recognition 377
- OpenMOSS-Team/MOSS-Transcribe-preview-2BAutomatic speech recognition 299
- mazesmazes/tiny-audioAutomatic speech recognition 210
- bosonai/Qwen3-ASR-1.7B-hf-orzeAutomatic speech recognition 73
- Musci-research/Musci-ASR-2.4BAutomatic speech recognition 47
- esc-bench/wav2vec2-aed-spgispeechAutomatic speech recognition 15
- esc-bench/conformer-rnnt-spgispeech 13
How to load it
from datasets import load_dataset
ds = load_dataset("kensho/spgispeech")