Libri-Light

by Meta AI / Facebook AI Research

60k hours of unlabeled English audiobook audio for self-supervised pretraining.

TL;DR

60k hours of unlabeled English audiobook audio for self-supervised pretraining.

Best for self-supervised pretraining (wav2vec 2.0, HuBERT) and limited-supervision benchmarks (10min / 1h / 10h splits). Pricing: free.

Category
Open source
License
—
Stars
—
Last push
—
Pricing
free
Platforms
GitHub, OpenSLR

What it is

Libri-Light packages ~60,000 hours of unlabeled LibriVox audio plus a limited-supervision evaluation suite (10min / 1h / 10h labeled splits). The standard pretraining corpus for English self-supervised speech models. License: CC BY 4.0.

Best for: Self-supervised pretraining (wav2vec 2.0, HuBERT) and limited-supervision benchmarks (10min / 1h / 10h splits).
Watch out for: CC BY 4.0 · derived from LibriVox public-domain audiobooks · English-only · read speech bias · 60k hours of audio is ~7 TB. Cite: Kahn et al., ICASSP 2020.

Install / use

git clone https://github.com/facebookresearch/libri-light

Features

Speaker diarizationNo
Word-level timestampsNo
Streaming / real-timeNo
Languages supported1
HIPAA eligibleNo

Libri-Light vs Whipscribe

FeatureLibri-LightWhipscribe
CategoryOpen sourceTranscription APIs
Pricingfreefree beta
Speaker diarizationNoYes
Word timestampsNoYes
StreamingNoNo
Languages199
PlatformsGitHub, OpenSLRWeb, API, MCP

Alternatives to Libri-Light

OpenAI Whisper
OpenAI

The reference open-source multilingual ASR model from OpenAI.

OSS · MIT ★ 98.1k
whisper.cpp
Georgi Gerganov

C/C++ port of Whisper — runs on anything, from a Raspberry Pi to Apple Silicon.

OSS · MIT ★ 48.8k
faster-whisper
SYSTRAN

4× faster than reference Whisper using CTranslate2 — production sweet spot.

OSS · MIT ★ 22.3k

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.