Meta's Omnilingual Automatic Speech Recognition (ASR) system has been named to TIME's list of the Best Inventions of 2026. The recognition was announced on October 15, 2026, nearly a year after the company's research team first released the tool to the public. TIME's Best Inventions is an annual feature recognizing standout inventions of the year.

The system's core proposition is straightforward: automatic speech-to-text coverage for more than 1,600 languages, including approximately 500 low-resource languages that, until this project, no speech recognition system had supported.

What Meta Built

Automatic speech recognition, or ASR, is the technology that listens to spoken words and converts them into written text. It powers everyday features such as dictation on smartphones, captions on videos, and the transcripts generated after every work call.

Roughly 7,000 languages are spoken around the world. Meta's Fundamental AI Research team, known as FAIR, set out to close a large share of that gap and released Omnilingual ASR as open source on November 10, 2025.

The system supports more than 1,600 languages, more than 500 of which had never been covered by any ASR solution before. On accuracy, Meta reports a character error rate below 10% for 78% of the supported languages. Character error rate measures how many individual letters the system gets wrong; a score under 10% means fewer than one character in ten needs fixing.

The Technical Choices That Matter

Rather than shipping a single model, Meta released a family of models ranging from 300 million to 7 billion parameters in size, giving developers options to match computing resources to the task at hand.

Alongside the models, the company published the Omnilingual ASR Corpus, a dataset covering 350 underserved languages that gives other researchers raw material to build on. The system can also adapt to a new language using only a small number of audio samples, an approach known as zero-shot adaptation — a valuable shortcut for languages that have little recorded speech available to begin with.

Open Source, With Real-World Users

Omnilingual ASR was released under the Apache 2.0 license, one of the most widely used open-source licenses in software, allowing both commercial and research use. Meta developed the project with global partners and emphasized open data collection and licensing throughout the process.

One of the first real-world deployments has emerged in Nigeria, where health practitioners have put the system to work improving patient care.

Background: The Long Tail of Language Tech

Earlier ASR systems, including OpenAI's Whisper, have been capable of transcribing only approximately 99 languages, leaving much of the world's linguistic diversity without adequate voice technology support. That gap has kept the dictation, captioning, and transcription conveniences common in major languages largely out of reach for speakers of most of the world's languages.

What This Means

For developers and researchers, the Apache 2.0 license stands out as the headline feature. Open weights, a public corpus, and models spanning 300 million to 7 billion parameters mean the barrier to building local-language products is lower than it has ever been for many of these communities.

There are limits worth watching. A character error rate below 10% for 78% of languages still leaves 22% of supported languages above that threshold — making per-language accuracy figures a metric to watch going forward.