LibriSpeech
1000h read English audiobook corpus — the canonical ASR benchmark since 2015.
1000h read English audiobook corpus — the canonical ASR benchmark since 2015.
Best for training and evaluating English ASR models — every modern paper reports test-clean/test-other WER. Pricing: free.
What it is
LibriSpeech is the 1000-hour read-English benchmark derived from LibriVox audiobooks. Splits: train-clean-100 / train-clean-360 / train-other-500 + dev/test-clean + dev/test-other. The de-facto WER reference for English ASR. License: CC BY 4.0. BibTeX: panayotov2015librispeech.
Watch out for: CC BY 4.0 · read audiobook speech only · low ambient noise · single-speaker · not representative of conversational or domain audio. Cite: Panayotov et al., ICASSP 2015.
Install / use
from datasets import load_dataset; ds = load_dataset('openslr/librispeech_asr', 'clean')Features
| Speaker diarization | No |
| Word-level timestamps | No |
| Streaming / real-time | No |
| Languages supported | 1 |
| HIPAA eligible | No |
LibriSpeech vs Whipscribe
| Feature | LibriSpeech | Whipscribe |
|---|---|---|
| Category | Open source | Transcription APIs |
| Pricing | free | free beta |
| Speaker diarization | No | Yes |
| Word timestamps | No | Yes |
| Streaming | No | No |
| Languages | 1 | 99 |
| Platforms | HuggingFace, OpenSLR | Web, API, MCP |
Alternatives to LibriSpeech
Whipscribe is a managed faster-whisper + whisperX service. If you want transcripts without running infrastructure, paste a URL or drop a file in the form below — you'll have a transcript in seconds.