Project AURIONAI Use-Cases Research & Innovation
Meta
speech recognitionfamily

Omnilingual ASR

Meta multilingual ASR family covering 1,600+ languages; code and models released under Apache 2.0, training corpus CC-BY-4.0.

MetaMetaUSA
Results table of Omnilingual ASR error rates for high, medium and low resource languages.
Levelfamily
Classspeech recognition
Parametersnot in the notes
Contextnot in the notes
Released10 Nov 2025
Opennessopen source
Runs onCPU server, embedded GPU, single datacentre GPU
Statusactive

Benchmark Results

vendor-reported
Common Voice 22 (CV22-76) 22speech
7B LLM-ASR, CV22-76, approximatevendor-reported
6.4
03.57

Score in CER. Lower is better.

Nov 2025 arxiv.org
FLEURSspeech
7B LLM-ASR, FLEURS-102, approximatevendor-reported
6.2
03.57

Score in CER. Lower is better.

Nov 2025 arxiv.org
Where this comes from
Every score names who ran it: vendor-reported, maintainer-run or independent. A rank is shown with the population it was ranked in (gap rule 3).

Lineage

finetunesuccessor
MMS to MMS fine-tuned with LM (IDSV): finetune, medium confidenceMMS to MMS fine-tuned, no LM (IDSV): finetune, medium confidenceMMS to Omnilingual ASR: successor, medium confidencewav2vec 2.0 to Omnilingual ASR: successor, medium confidenceMMSMMSMetawav2vec 2.0wav2vec 2.0MetaMMS fine-tuned with LM (IDSV)MMS fine-tuned with LM (ID…Institut Teknologi Bandung (ITB),…MMS fine-tuned, no LM (IDSV)MMS fine-tuned, no LM (IDS…Institut Teknologi Bandung (ITB),…Omnilingual ASROmnilingual ASRMeta, Nov 2025
MetaMMS is the parent (successor)medium confidencenotes
Metawav2vec 2.0 is the parent (successor)medium confidencenotes
Where this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 5 models. Confidence and evidence are the notes' own.

R&D Topics

Massively multilingual speech recognitionexemplar

Identity

Developer
Meta
Family
not in the notes
Released
10 Nov 2025 sourcesannouncement page
Weights published
–
Record checked
10 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.

Architecture and Size

Architecture
other wav2vec 2.0 encoder + CTC head or about 1.2B LLM-style decoder; in-context few-shot for unseen languages; 300M/1B/3B/7B sizes
Total parameters
– sources
Active parameters
–
Checkpoint parameters
–
Note
317M to 6.49B encoders
Context length
not in the notes
Input
audio
Output
text
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).

Openness and Access

Openness
open source
Licence
Apache License 2.0 sources
Access routes
open weights download
Where this comes from
The licence file governs over registry tags and aggregator labels.

Deployment

Runs on
CPU server, embedded GPU, single datacentre GPU
Footprint
300M edge/CPU; 7B single GPU
Runtimes
not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.

Indonesia

Indonesian support
not stated
Regional languages
not in the notes
Jakarta hosting
unknown
Languages
1,600+ languages; average 7.1 CER over 1,570 languages
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.

Industry Use Cases

No industry use case links this model.

Sources

  1. arxiv.org/pdf/2511.09690 https://arxiv.org/pdf/2511.09690
  2. GitHub - facebookresearch/omnilingual-asr: Omnilingual ASR Open-Source Multilingual SpeechRecognition for 1600+ Languages github.com https://github.com/facebookresearch/omnilingual-asr

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Omnilingual ASR", https://aurion.semai.id/models/meta-omnilingual-asr/, records of October 2026.