TIME names Meta’s 1,600-language speech system among 2026’s best inventions

TIME included Meta’s Omnilingual ASR in its Best Inventions of 2026 coverage on October 15, recognizing a speech-recognition system that supports 1,600 languages. The entry highlights a practical use for that reach: Nigerian health practitioners are already using it in community clinics. Universal transcription remains the project’s ambition, rather than an achieved result.

Automatic speech recognition, or ASR, turns spoken words into text. TIME’s entry describes Omnilingual ASR as a suite of open-source models and says 500 of its supported languages cannot currently be transcribed by other AI systems.

The underlying work predates the recognition. Meta’s research publication is dated November 10, 2025, and describes coverage of more than 1,600 languages, including over 500 never previously served by an ASR system.

Meta frames the problem as both a coverage gap and an obstacle to participation. Its paper says most of the world’s more than 7,000 languages remain unsupported. It also argues that systems restricted to a fixed set of languages make it difficult for communities to add their own.

Beyond its initial language count, the system is designed to accept additions. Meta says communities can introduce an unserved language with a handful of their own data samples. Its published language list is therefore not intended to be a permanent boundary.

The training material combines public resources with community recordings collected through compensated local partnerships. Meta presents that collaboration as part of the design. The publication also says open-source models and tools can lower the expertise and computing barriers facing researchers and communities that want to participate.

TIME says Nigerian health practitioners are using Omnilingual ASR to improve patient care in community clinics, though its brief entry provides no measured patient outcomes.

Transcription is also only one step in the language workflow TIME describes. Paired with Meta’s Omnilingual Machine Translation system, the resulting transcripts can be translated across more than 1,600 languages. Speech recognition supplies the written words; translation is handled by the paired system.