Meta, the technology giant behind Facebook and Instagram, has made a major breakthrough in artificial intelligence with the launch of Omnilingual ASR, a powerful speech recognition system that can understand and transcribe over 1,600 languages worldwide. This opens up huge possibilities for people speaking less well-known languages to use AI tools in their everyday lives.

Breaking New Ground in Language Technology

Meta’s Fundamental AI Research (FAIR) team revealed the new system on 10 November 2025. This system stands out because it supports nearly 500 low-resource languages that earlier AI models ignored. In comparison, popular speech tools by companies like OpenAI support fewer than 100 languages. Meta’s Omnilingual ASR system performs very well for most languages, achieving low error rates in transcription, which means the AI can accurately convert spoken words into text even if the accent or dialect is unfamiliar.

The system is built on a very large AI model trained on 4.3 million hours of audio without transcripts. This “self-supervised” training means the AI learned patterns in speech on its own before being fine-tuned to specific languages. Meta has made the technology open source under a licence that allows anyone to use it freely, adapt it, and even develop commercial products with it.

Expanding Access Through AI

Meta’s Chief AI Officer, Alexandr Wang, called Omnilingual ASR a step towards ‘universal AI’, an intelligence that understands everyone, regardless of which language they speak. The AI uses a technique called zero-shot learning, meaning it can quickly learn to recognise new languages with just a few examples. This flexibility allows the system’s reach to grow beyond the 1,600 languages it initially supports to potentially over 5,400 languages using simple real-world audio and text samples.

By providing different-sized models, from a small 300 million-parameter version that can run on mobile phones to a large 7 billion-parameter version for powerful computers, Meta has made the system useful for various devices and situations. This wide usability is essential for India and other multilingual countries where smartphones are the main access point for digital services.

Bridging the Language Divide

One of Meta’s main goals with Omnilingual ASR is to make AI accessible to language communities that have been left behind by current technologies. To create the system, Meta partnered with language preservation groups and collected recordings from many underrepresented communities. They also released an Omnilingual ASR Corpus, a large dataset containing transcriptions of 350 underserved languages to help researchers and developers improve the AI further.

While the system shows excellent transcription accuracy for many well-represented languages, accuracy still varies for very low-resource ones. Around 95 percent of high and medium-resource languages have character error rates below 10 per cent, but only 36 per cent of low-resource languages reach that level. This shows there is still work to be done to ensure equal benefits for all language speakers.

Meta’s broader vision for this technology ties into efforts to advance AI superintelligence, aiming to push artificial intelligence capabilities much further. The open-source release is also a shift in how Meta shares its research, encouraging innovation by giving developers worldwide immediate access to cutting-edge AI tools for spoken language processing.

Long or Short, get news the way you like. No ads. No redirections. Download Newspin and Stay Alert, The CSR Journal Mobile app, for fast, crisp, clean updates!

App Store –  https://apps.apple.com/in/app/newspin/id6746449540 

Google Play Store – https://play.google.com/store/apps/details?id=com.inventifweb.newspin&pcampaignid=web_share