Hi HN, We built Omnilingual ASR, a large-scale multilingual speech recognition system designed to make transcription accessible worldwide. It supports 1,600+ languages, including 500 previously unsupported low-resource ones, all with state-of-the-art accuracy.

Unlike typical ASR models limited to major languages, Omnilingual ASR is trained on 4.3 million hours of multilingual audio, achieving <10% character error rate for 78% of supported languages. It scales from 300M to 7B parameters, allowing users to balance speed and precision for any task.

Key Features

• 1,600+ Language Coverage – Extendable to 5,400+ via zero-shot learning

• Zero-Shot Adaptation – Add new languages with just a few in-context examples

• Multi-Speaker Detection – Automatically identify and separate speakers

• Lightning-Fast Processing – Transcribe hours of audio in minutes

• Flexible Integration – REST API, Python SDK, and web UI for cloud or edge use

Use Cases

Global media subtitling, enterprise transcription, multilingual e-learning, accessibility services, and linguistic research.

We’d love feedback on:

• Potential use cases for low-resource language transcription

• Integration needs (APIs, SDKs, plugins)

• Research or language preservation applications

https://www.omnilingualasr1.com/?i=d1d5k