Falcon ASR

Falcon ASR is TII's 1.6B speech recognition model with a focus on Arabic and the Emirati dialect. It also transcribes English, French, Spanish and Portuguese. The published access route is a public demo.

One model handles the supported languages

The model uses the same weights across five languages without a language flag. Its training included noise, overlapping speakers, reverberation and telephone effects. The release links to a demo, rather than a downloadable weight repository.TII: Introducing Falcon ASR

Review names and mixed-language speech

Practical guidance. For an Emirati recording, check place names, local expressions and changes between Arabic and English. Word-level timestamps can help reviewers find errors. The public clip demo does not establish limits for a future batch or streaming API.

Specifications & access

Falcon ASR specifications
DetailValue and conditions
AccessProvider product
Model identifiersFalcon-ASR
OutputTranscript in the spoken language, with word-level timestampsTII: Introducing Falcon ASR
Model ancestryBuilds on TII's Falcon3-Audio workTII: Introducing Falcon ASR
API statusPlanned in the October 2026 release; no public API is claimed hereTII: Introducing Falcon ASR
Demo scopeAudio clips up to 60 secondsFalcon ASR public demo

Related models