Meta Omnilingual ASR: open speech recognition for 1,600+ languages
Don’t want to install Meta Omnilingual ASR? Drop the file here.
The first 20 minutes of any recording are transcribed free: a recording up to 20 minutes transcribes in full, a longer one is previewed, and a pack from $6 (500 minutes) transcribes the rest and opens every export, and credits never expire.
Meta's open speech recognition for over 1,600 languages, hundreds never covered before, with CTC and LLM-based models from 300M to 7B parameters under Apache-2.0.
Get Meta Omnilingual ASR Official site & pricing
Omnilingual ASR is Meta's open speech recognition for 1,600+ languages, including Cebuano, Javanese and Nepali, with CTC and LLM models from 300M to 7B. Apache-2.0.
Except for the Unlimited variants, inputs must be under 40 seconds, so long recordings need chunking. The predecessor is Meta MMS.
What it is
Omnilingual ASR is Meta FAIR's open-source speech-recognition system for more than 1,600 languages, released in November 2025. It ships wav2vec2-based CTC models and LLM-ASR models at 300M, 1B, 3B and 7B parameters; the December 2025 update added v2 checkpoints with lower error rates and LLM_Unlimited variants that decode audio of any length. Meta reports a character error rate below 10 for 78% of languages with its 7B LLM-ASR system. Code and models are Apache-2.0, and the training corpus is on Hugging Face under CC-BY-4.0. The supported-language list uses language_script codes (for example ceb_Latn, jav_Latn, nep_Deva); at our check it held 1,668 entries. It is a separate release from Meta's earlier MMS models.
Watch out for: The standard CTC and LLM models accept audio shorter than 40 seconds; only the LLM_Unlimited variants take long audio. A research pipeline built on fairseq2 and libsndfile, and you pass the language code.
Long recordings, hosted, without 40-second chunks
- Drop the audio or video file into the box at the top of this page, or paste a link (YouTube, a podcast or RSS feed, Google Drive, Twitch or Kick). Nothing to install and no GPU needed.
- The language is detected automatically and the transcript stays in the language that was spoken (WhipScribe does not translate).
- Read the summary free. A pack from $6 (500 minutes) opens the full transcript and the downloads: TXT, DOCX, SRT, VTT and JSON. Packs are one-time purchases and credits never expire.
- Files can be up to 10 hours or 5 GB per file. No 40-second chunking on your side.
Related: Transcribe Cebuano · Javanese transcription · Nepali transcription · Meta MMS · Pricing
Install / use
pip install omnilingual-asrKey specs
What people actually do with Meta Omnilingual ASR-style transcription
The tool is the means. These are the jobs — each one priced at published rates, each one wired up on its own page.
Features
| Speaker diarization | Not stated |
| Word-level timestamps | Not stated |
| Streaming / real-time | Not stated |
| Languages supported | 1600 |
| HIPAA eligible | No |
Links
- github.com/facebookresearch/omnilingual-asr ↗Apache-2.0 · 0.2.0 on 2025-12-22
- Omnilingual ASR demo ↗Hugging Face Space
- omnilingual-asr-corpus ↗training corpus · CC-BY-4.0
Meta Omnilingual ASR vs Whipscribe
| Feature | Meta Omnilingual ASR | Whipscribe |
|---|---|---|
| Category | Open source | Transcription APIs |
| Pricing | free (Apache-2.0) | $6–$24 one-time packs (credits never expire) · first 20 minutes free |
| Speaker identification | No | Named in the summary, key quotes and chapters |
| Word timestamps | No | Yes |
| Streaming | No | No |
| Languages | 1600 | 99 |
| Platforms | Linux, macOS | Web, API, MCP |
Alternatives to Meta Omnilingual ASR
Frequently asked about Meta Omnilingual ASR
How many languages does Omnilingual ASR support?
More than 1,600. The list is in the repository as language_script codes such as eng_Latn, ceb_Latn (Cebuano), jav_Latn (Javanese) and nep_Deva (Nepali); it held 1,668 codes when we checked on 11 October 2026.
Can Omnilingual ASR transcribe long recordings?
The CTC and standard LLM models accept audio shorter than 40 seconds. The omniASR_LLM_Unlimited models added in December 2025 decode audio of unlimited length.
How is Omnilingual ASR different from Meta MMS?
It is a newer, separate release (November 2025) covering 1,600+ languages against MMS's 1,100+, with LLM-based decoders up to 7B parameters, and it is Apache-2.0.
How do I install Omnilingual ASR?
pip install omnilingual-asr (or uv add omnilingual-asr), plus libsndfile for audio. Then load ASRInferencePipeline with a model card such as omniASR_LLM_Unlimited_7B_v2 and pass the language codes.
Can I use Meta Omnilingual ASR in the browser?
Meta hosts a demo on Hugging Face Spaces (facebook/omniasr-transcriptions) for trying it. For your own recordings at volume, you run the models yourself.
Does WhipScribe cover the same languages?
No. WhipScribe detects 99+ languages automatically. For a language outside that set, or a low-resource language where accuracy is critical, Omnilingual ASR is one of the few open options; test a short clip on WhipScribe before committing either way.
What is Meta Omnilingual ASR?
Omnilingual ASR is Meta's open speech recognition for 1,600+ languages, including Cebuano, Javanese and Nepali, with CTC and LLM models from 300M to 7B. Apache-2.0.
Is Meta Omnilingual ASR free?
Yes. Meta Omnilingual ASR is free and open source (free (Apache-2.0)).
Is Meta Omnilingual ASR HIPAA compliant?
Meta Omnilingual ASR does not claim HIPAA compliance. For PHI, run it on infrastructure you control or choose a service that will sign a BAA.
What are the limitations of Meta Omnilingual ASR?
The standard CTC and LLM models accept audio shorter than 40 seconds; only the LLM_Unlimited variants take long audio. A research pipeline built on fairseq2 and libsndfile, and you pass the language code.
Who is Meta Omnilingual ASR best for?
Researchers and developers working in low-resource languages that mainstream engines skip, with GPU capacity for a 1B-7B model.
What should I know before choosing Meta Omnilingual ASR?
Except for the Unlimited variants, inputs must be under 40 seconds, so long recordings need chunking. The predecessor is Meta MMS.
Does Meta Omnilingual ASR run offline?
Meta Omnilingual ASR runs on your own machine — it is open source, so nothing leaves the computer unless you configure it to.
Where is the Meta Omnilingual ASR source code?
On GitHub at github.com/facebookresearch/omnilingual-asr. It is maintained by Meta AI (FAIR).
What kind of tool is Meta Omnilingual ASR?
In this directory Meta Omnilingual ASR is filed under oss engine, low resource languages as a open-source tool.
Whipscribe is a hosted transcription 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.
Transcribe your recordings
Running Meta Omnilingual ASR yourself? Send the same file here and compare the transcript. Transcribe any language, any recording · Transcription services from $0.29 an hour · Transcribe from any meeting or recording platform · How to transcribe, step by step
Explore
All transcription tools · By language · By platform · By file format · How-to guides · Transcription services · Transcription API · Pricing
Long recording to get through?
One file up to 10 hours, as an upload or a link: see how to transcribe an MP3 of any length, a podcast episode as text from its link and the SRT file generator.
Comparing tools before you pay?
If you would rather pay a service than run software yourself: see an honest roundup of the best transcription services, affordable transcription priced per hour and how our transcription services work.