YODAS2

by CMU / WAVLab

Refresh of YODAS with long-form audio + per-language sharding — 422k hours.

TL;DR

Refresh of YODAS with long-form audio + per-language sharding — 422k hours.

Best for long-form ASR training (full videos preserved); cross-lingual alignment. Pricing: research-only.

Category
Open source
License
—
Stars
—
Last push
—
Pricing
research-only
Platforms
HuggingFace

What it is

YODAS2 refactors YODAS with full-video audio + segment-level metadata (versus 30-second clips in v1), enabling long-form ASR training. 149 languages, 422k hours.

Best for: Long-form ASR training (full videos preserved); cross-lingual alignment.
Watch out for: Same YouTube + CC BY 3.0 subtitle constraint as YODAS v1. Cite: Li et al., 2024.

Install / use

from datasets import load_dataset; ds = load_dataset('espnet/yodas2', 'en000')

Features

Speaker diarizationNo
Word-level timestampsYes
Streaming / real-timeNo
Languages supported149
HIPAA eligibleNo

YODAS2 vs Whipscribe

FeatureYODAS2Whipscribe
CategoryOpen sourceTranscription APIs
Pricingresearch-onlyfree beta
Speaker diarizationNoYes
Word timestampsYesYes
StreamingNoNo
Languages14999
PlatformsHuggingFaceWeb, API, MCP

Alternatives to YODAS2

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.