View organization page for AI at Meta
AI at Meta

1,030,483 followers

Introducing Meta Omnilingual Automatic Speech Recognition (ASR), a suite of models providing ASR capabilities for over 1,600 languages, including 500 low-coverage languages never before served by any ASR system. While most ASR systems focus on a limited set of languages that are well-represented on the internet, this release marks a major step toward building a truly universal transcription system. 🔗 Learn more: https://go.meta.me/ff13fa Highlights include: - State-of-the-art performance with character error rates below 10 for 78% of supported languages. - First large-scale ASR framework with in-context learning, enabling extension to new languages with just a few audio samples. - A full suite of open source models and a dataset, including Omnilingual w2v 2.0, a 7B-parameter multilingual speech representation model, and Omnilingual ASR Corpus, a unique dataset spanning 350 underserved languages.

Erik Arakelyan

Senior Researcher at Nvidia

1w

Since when are Eastern and Western Armenian vulnerable or endangered languages ?

  • No alternative text description for this image
Like
Reply
22 Reactions
Chirag Chandnani

Master’s in CS @ UIUC | ML Engineer | LLMs and RAGs | J.N. Tata Scholar 2025 | Uni Topper @ VIT Vellore CS

1w

This could be transformative, especially for a country like India which has 1000+ languages and dialects, it can address problems such as making education more inclusive through native language learning tools, and improving healthcare communication in rural areas, especially with doctors from different states of the country. Truly a step towards a more "inclusive" future.

Like
Reply
11 Reactions
AI at Meta

1w

Head over to the Omnilingual demo to explore the languages in the dataset: https://aidemos.atmeta.com/omnilingualasr/language-globe

Like
Reply
10 Reactions
Faizan Javed

Agentic AI Expert | Building & Deploying AI Agents | Empowering organizations to amplify human potential through intelligent automation

1w

1,600 languages. 500 that have never had voice recognition before. Let that sink in. For decades, AI only spoke if you were privileged enough to speak English, Mandarin, or maybe 50 other languages. Billions of people were locked out. Their voices ignored. Their languages treated as irrelevant. Meta just shattered that barrier. This is not about better technology. This is about who gets to exist in the digital world. Communities speaking Tamasheq, Lingala, and hundreds of endangered languages now have a seat at the table. The in-context learning changes everything. A few audio samples and the system learns your language. No massive datasets required. No years of development. Just access. And they made it open source. Free for researchers, educators, and the communities who need it most. This is the AI revolution we should be talking about. Not the one that makes the rich richer. The one that gives voice to the voiceless. When 78 percent of 1,600 languages achieve error rates below 10, that is not a metric. That is millions of people finally being heard. What will you build with this?

Like
Reply
2 Reactions
Denis B.

Software Engineer / Developer (SWE) | Backend, Fullstack, Cloud | Python, C++, Rust | High-Load Distributed Systems | PostgreSQL/ClickHouse | UK Work Permit

1w

Incredible work by the Meta team! The focus on low-coverage languages and the new in-context learning feature make this a game-changer for accessibility globally.

Like
Reply
2 Reactions
MyongHak J.

Building I.N.G: A real-time video platform where your curiosity becomes income. Let’s connect. (Your curiosity deserves ROI.)| Real-time platform | Shortform | AI video tech

1w

Truly impressive leap, Meta, expanding ASR to 1,600 languages isn’t just technological progress, it’s cultural inclusion at scale. Giving voice to underserved languages means giving identity back to communities that were digitally silent.

Like
Reply
2 Reactions
Muhammad Ramzan

Helping Businesses & Professionals Grow with B2B Lead Gen, SEO, LinkedIn Outreach, and Profile Optimization | Certified LinkedIn Expert | Technical Recruitment Specialist

1w

Outstanding progress toward universal, inclusive ASR. Exciting to see Meta pushing the boundaries of multilingual AI! 🚀

Like
Reply
1 Reaction
Lasse Brüggemann

Digital, Data & Transformation Lead

1w

Radomir Popovic

Like
Reply
1 Reaction
Felipe Rodrigues

Fundador da Frather & Independent Dollar Futures Trader

1w

For rich of personality, Frather ♣️🔥

Like
Reply
1 Reaction
Luck Suknimit

ลักษณ์ สุขนิมิตร, IG:@LUCK.SUKNIMIT.2024 Bangkok Thailand🇹🇭 official

1w

That's great

Like
Reply
1 Reaction
See more comments

To view or add a comment, sign in

More Relevant Posts

  • View profile for Alexandre MOURACHKO
    Alexandre MOURACHKO

    Research Engineering Manager at Facebook AI Research

    Super excited to share this huge team effort with the rest of the world today! While most ASR systems focus on a limited set of languages that are well-represented on the internet, Omnilingual ASR is the first ever to support 1600 languages out of the box, and even more through in-context learning. You can use our models today with a commercial license, as well as other research artifacts like Omnilingual ASR corpus, our commissioned dataset https://lnkd.in/eiFt_jWg We need to make multilinguality a priority for AI research, towards true "Omnilinguality".

    View organization page for AI at Meta
    AI at Meta

    1,030,483 followers

    Introducing Meta Omnilingual Automatic Speech Recognition (ASR), a suite of models providing ASR capabilities for over 1,600 languages, including 500 low-coverage languages never before served by any ASR system. While most ASR systems focus on a limited set of languages that are well-represented on the internet, this release marks a major step toward building a truly universal transcription system. 🔗 Learn more: https://go.meta.me/ff13fa Highlights include: - State-of-the-art performance with character error rates below 10 for 78% of supported languages. - First large-scale ASR framework with in-context learning, enabling extension to new languages with just a few audio samples. - A full suite of open source models and a dataset, including Omnilingual w2v 2.0, a 7B-parameter multilingual speech representation model, and Omnilingual ASR Corpus, a unique dataset spanning 350 underserved languages.

    To view or add a comment, sign in

  • View profile for Matthew Setzler
    Matthew Setzler

    Senior AI Research Scientist, Meta | PhD, Cognitive Science

    Very excited to share that the Omnilingual Automatic Speech Recognition (ASR) model is launching today! I had the privilege of working on this with my colleagues at FAIR (Meta AI), with participation from linguists and language speakers across the world. Most ASR systems focus on just the highest-resource languages that are well represented on the internet. In this project, we hugely expand the range of language coverage to more than 1600 languages. Further, we developed a zero-shot model that can transcribe unseen languages with just a few reference examples. The result is a SoTA ASR model with the largest language coverage of any Speech AI model to date (by far). In addition to releasing our paper; we are open-sourcing a suite of model checkpoints at different sizes, inference and finetuning code, and a dataset comprising hundreds of languages that have never before been incorporated into any ASR systems. Check out the blog post for more information, and links to the demo for a transcription tool and fun visualizations 🎤 🌍 Blog post: https://lnkd.in/ejy5rAYy Paper: https://lnkd.in/eYEQNW8k https://lnkd.in/efJBrGrp

    View organization page for AI at Meta
    AI at Meta

    1,030,483 followers

    Introducing Meta Omnilingual Automatic Speech Recognition (ASR), a suite of models providing ASR capabilities for over 1,600 languages, including 500 low-coverage languages never before served by any ASR system. While most ASR systems focus on a limited set of languages that are well-represented on the internet, this release marks a major step toward building a truly universal transcription system. 🔗 Learn more: https://go.meta.me/ff13fa Highlights include: - State-of-the-art performance with character error rates below 10 for 78% of supported languages. - First large-scale ASR framework with in-context learning, enabling extension to new languages with just a few audio samples. - A full suite of open source models and a dataset, including Omnilingual w2v 2.0, a 7B-parameter multilingual speech representation model, and Omnilingual ASR Corpus, a unique dataset spanning 350 underserved languages.

    To view or add a comment, sign in

  • View profile for Yu-An Chung
    Yu-An Chung

    Research Scientist at Meta AI

    Excited to share that Omnilingual Automatic Speech Recognition (ASR) for 1600+ languages is launching today! While most ASR systems focus on a limited set of languages that are well-represented on the internet, this release marks a major step toward building a truly universal transcription system. This was a truly global effort and we are so excited to share it with the world today. Key results: - We are releasing a suite of models providing ASR capabilities for over 1,600 languages, including 500 low-coverage languages never before served by any ASR system - We are also releasing the Omnilingual ASR Corpus, a unique dataset spanning 350 underserved languages that was curated in collaboration with our global partners - State-of-the-art performance with character error rates below 10% for 78% of supported languages. - First large-scale ASR framework with in-context learning, enabling extension to new languages with just a few audio samples. Links:  - Paper: https://lnkd.in/esHUPQ3U - Models and dataset: https://lnkd.in/erttjbA3 - Language exploration demo: https://lnkd.in/eCWw9Gxr - Transcription tool: https://lnkd.in/ek55wzxm

    View organization page for AI at Meta
    AI at Meta

    1,030,483 followers

    Introducing Meta Omnilingual Automatic Speech Recognition (ASR), a suite of models providing ASR capabilities for over 1,600 languages, including 500 low-coverage languages never before served by any ASR system. While most ASR systems focus on a limited set of languages that are well-represented on the internet, this release marks a major step toward building a truly universal transcription system. 🔗 Learn more: https://go.meta.me/ff13fa Highlights include: - State-of-the-art performance with character error rates below 10 for 78% of supported languages. - First large-scale ASR framework with in-context learning, enabling extension to new languages with just a few audio samples. - A full suite of open source models and a dataset, including Omnilingual w2v 2.0, a 7B-parameter multilingual speech representation model, and Omnilingual ASR Corpus, a unique dataset spanning 350 underserved languages.

    To view or add a comment, sign in

  • View profile for Shireen Yates
    Shireen Yates

    Product Leader

    Thrilled to share this work with the world! Automatic Speech Recognition (ASR) is technology that can turn speech into text, enabling tasks like transcriptions. Meta's Omnililingual ASR is pioneering research enabling ASR for 1600+ languages, and 500 language never before "heard" by AI. ASR is used for a wide range of tasks dependent on converting spoken language into text, including but not limited to: education (transcription of lectures, language learning tools), language preservation (recording spoken languages), healthcare (clinical documentation, patient engagement), emergency response (transcription of emergency callers for voice-activated systems) and so much more! We hope the open-source community can develop further on what we have started with Omnilingual ASR and eventually enable ASR and ASR dependent tasks for all spoken language in the world.

    View organization page for AI at Meta
    AI at Meta

    1,030,483 followers

    Introducing Meta Omnilingual Automatic Speech Recognition (ASR), a suite of models providing ASR capabilities for over 1,600 languages, including 500 low-coverage languages never before served by any ASR system. While most ASR systems focus on a limited set of languages that are well-represented on the internet, this release marks a major step toward building a truly universal transcription system. 🔗 Learn more: https://go.meta.me/ff13fa Highlights include: - State-of-the-art performance with character error rates below 10 for 78% of supported languages. - First large-scale ASR framework with in-context learning, enabling extension to new languages with just a few audio samples. - A full suite of open source models and a dataset, including Omnilingual w2v 2.0, a 7B-parameter multilingual speech representation model, and Omnilingual ASR Corpus, a unique dataset spanning 350 underserved languages.

    To view or add a comment, sign in

  • View profile for Jeff Wang
    Jeff Wang

    Product Management | MSL, Fundamental AI Research

    🗣️ New Speech Recognition research from FAIR today! - suite of ASR models that cover 1600 languages, including 500 never served before by any ASR system - 300M-to-7B model sizes - released under Apache 2.0 license - supports low-resource languages with less than <10 hrs of data ⬇️ Download: https://lnkd.in/gvD9vBFq 📄 Paper: https://lnkd.in/gVXNtnHW 🌎 Language Exploration demo: https://lnkd.in/gWf8jRyJ 🚨 Learn more: https://lnkd.in/ggJeQg4c

    View organization page for AI at Meta
    AI at Meta

    1,030,483 followers

    Introducing Meta Omnilingual Automatic Speech Recognition (ASR), a suite of models providing ASR capabilities for over 1,600 languages, including 500 low-coverage languages never before served by any ASR system. While most ASR systems focus on a limited set of languages that are well-represented on the internet, this release marks a major step toward building a truly universal transcription system. 🔗 Learn more: https://go.meta.me/ff13fa Highlights include: - State-of-the-art performance with character error rates below 10 for 78% of supported languages. - First large-scale ASR framework with in-context learning, enabling extension to new languages with just a few audio samples. - A full suite of open source models and a dataset, including Omnilingual w2v 2.0, a 7B-parameter multilingual speech representation model, and Omnilingual ASR Corpus, a unique dataset spanning 350 underserved languages.

    To view or add a comment, sign in

  • View profile for Skyler Wang
    Skyler Wang

    Assistant Professor, McGill University | Research Scientist, Meta

    📢📢 Very excited to share my team's new work on Omnilingual ASR: an open-source system that provides out-of-the-box automatic speech recognition for over 1,600 languages. ASR is a cornerstone of today’s most widely used technologies, from voice assistants to transcription tools, yet most languages remain unsupported. What’s especially exciting is that, unlike previous systems with fixed coverage, Omnilingual ASR can extend to unseen languages using just a few in-context examples. As always, we look forward to all the different, creative ways folks will use a system like this, as well as valuable community feedback. 🌍🗣️ Read the paper here: https://lnkd.in/giD8wydr

    View organization page for AI at Meta
    AI at Meta

    1,030,483 followers

    Introducing Meta Omnilingual Automatic Speech Recognition (ASR), a suite of models providing ASR capabilities for over 1,600 languages, including 500 low-coverage languages never before served by any ASR system. While most ASR systems focus on a limited set of languages that are well-represented on the internet, this release marks a major step toward building a truly universal transcription system. 🔗 Learn more: https://go.meta.me/ff13fa Highlights include: - State-of-the-art performance with character error rates below 10 for 78% of supported languages. - First large-scale ASR framework with in-context learning, enabling extension to new languages with just a few audio samples. - A full suite of open source models and a dataset, including Omnilingual w2v 2.0, a 7B-parameter multilingual speech representation model, and Omnilingual ASR Corpus, a unique dataset spanning 350 underserved languages.

    To view or add a comment, sign in

  • View profile for Stefan Bauschard
    Stefan Bauschard

    AI Education Policy Consultant

    This idea that GenAI is just trained on and communicates with English still floats around. This is another contrary example.

    View organization page for AI at Meta
    AI at Meta

    1,030,483 followers

    Introducing Meta Omnilingual Automatic Speech Recognition (ASR), a suite of models providing ASR capabilities for over 1,600 languages, including 500 low-coverage languages never before served by any ASR system. While most ASR systems focus on a limited set of languages that are well-represented on the internet, this release marks a major step toward building a truly universal transcription system. 🔗 Learn more: https://go.meta.me/ff13fa Highlights include: - State-of-the-art performance with character error rates below 10 for 78% of supported languages. - First large-scale ASR framework with in-context learning, enabling extension to new languages with just a few audio samples. - A full suite of open source models and a dataset, including Omnilingual w2v 2.0, a 7B-parameter multilingual speech representation model, and Omnilingual ASR Corpus, a unique dataset spanning 350 underserved languages.

    To view or add a comment, sign in

  • View profile for Francisco Matos
    Francisco Matos

    Cybersecurity & AI Counsel | Data Protection & Privacy Expert | Regulatory & Risk Advisor (GDPR, NIS2, EU AI Act, EECC, ePD)

    Incredible work from the team at AI at Meta! 🤯 The release of Omnilingual Automatic Speech Recognition (ASR) is a massive step forward for global accessibility and digital inclusion. Key highlights of this breakthrough: 🤖1,600+ Languages Covered: This is the largest ASR system to date, including over 500 low-resource languages that have never before been transcribed by AI. 🥢Community-Driven: The model is designed to be extensible, allowing people around the world to add support for new languages with only a few of their own samples. 📭Open-Source Commitment: They are open-sourcing the models (Omnilingual wav2vec 2.0) and the Omnilingual ASR Corpus, a unique collection of transcribed speech in 350 underserved languages, for non-commercial use. This work directly addresses the digital divide and pushes the boundaries of what is possible in multilingual speech technology. #AI #ASR #Multilingual #OpenSource #MachineLearning #NLP

    View organization page for AI at Meta
    AI at Meta

    1,030,483 followers

    Introducing Meta Omnilingual Automatic Speech Recognition (ASR), a suite of models providing ASR capabilities for over 1,600 languages, including 500 low-coverage languages never before served by any ASR system. While most ASR systems focus on a limited set of languages that are well-represented on the internet, this release marks a major step toward building a truly universal transcription system. 🔗 Learn more: https://go.meta.me/ff13fa Highlights include: - State-of-the-art performance with character error rates below 10 for 78% of supported languages. - First large-scale ASR framework with in-context learning, enabling extension to new languages with just a few audio samples. - A full suite of open source models and a dataset, including Omnilingual w2v 2.0, a 7B-parameter multilingual speech representation model, and Omnilingual ASR Corpus, a unique dataset spanning 350 underserved languages.

    To view or add a comment, sign in

  • View organization page for OneNine
    OneNine

    859 followers

    Meta’s new Omnilingual ASR confirms a truth we’ve built OneNine around: the next frontier in AI isn’t language coverage; it’s language quality. While Meta’s models now span 1,600 languages, 22% still face high error rates. That gap highlights a global need for verified, culturally-accurate, enterprise-grade data; the kind that web-scraped corpora can’t deliver. OneNine exists to close that gap. We provide the human-audited, high-fidelity linguistic data that transforms large-scale models into commercially reliable systems. Meta has shown the appetite for inclusive AI. We deliver the quality that makes it possible. #ONENINE #MetaASR #AIData #LowResourceLanguages #DataQuality #MachineLearning #SpeechRecognition #EthicalAI

    View organization page for AI at Meta
    AI at Meta

    1,030,483 followers

    Introducing Meta Omnilingual Automatic Speech Recognition (ASR), a suite of models providing ASR capabilities for over 1,600 languages, including 500 low-coverage languages never before served by any ASR system. While most ASR systems focus on a limited set of languages that are well-represented on the internet, this release marks a major step toward building a truly universal transcription system. 🔗 Learn more: https://go.meta.me/ff13fa Highlights include: - State-of-the-art performance with character error rates below 10 for 78% of supported languages. - First large-scale ASR framework with in-context learning, enabling extension to new languages with just a few audio samples. - A full suite of open source models and a dataset, including Omnilingual w2v 2.0, a 7B-parameter multilingual speech representation model, and Omnilingual ASR Corpus, a unique dataset spanning 350 underserved languages.

    To view or add a comment, sign in

  • View profile for Manindra Utchanah
    Manindra Utchanah

    LegalTech innovator and Barrister-at-Law

    Finally, some proper open-source speech-to-text tools for Mauritian Creole 🇲🇺! I’ve tested it, and while it still needs fine-tuning, the results are better than I expected. This could make a real difference for court trials, disciplinary hearings, and client meetings: all automatically transcribed in Kreol⚖️ And beyond LegalTech, it’s a big step toward making AI technology truly accessible to everyone: from AI tutors to business tools and public-service apps that actually understand spoken Kreol Morisien. Can’t wait to plug it into my own workflows and even more excited to see how the local developer community builds on this 🚀

    View organization page for AI at Meta
    AI at Meta

    1,030,483 followers

    Introducing Meta Omnilingual Automatic Speech Recognition (ASR), a suite of models providing ASR capabilities for over 1,600 languages, including 500 low-coverage languages never before served by any ASR system. While most ASR systems focus on a limited set of languages that are well-represented on the internet, this release marks a major step toward building a truly universal transcription system. 🔗 Learn more: https://go.meta.me/ff13fa Highlights include: - State-of-the-art performance with character error rates below 10 for 78% of supported languages. - First large-scale ASR framework with in-context learning, enabling extension to new languages with just a few audio samples. - A full suite of open source models and a dataset, including Omnilingual w2v 2.0, a 7B-parameter multilingual speech representation model, and Omnilingual ASR Corpus, a unique dataset spanning 350 underserved languages.

    To view or add a comment, sign in

AI at Meta

1,030,483 followers

View Profile

Explore content categories