Chinar is not a multilingual model with Azerbaijani bolted on, and not Turkish with an accent. The language is what it was built around — which is the difference between a transcript you can act on and one that merely looks finished.
Most speech engines treat Azerbaijani as a neighbouring case of a language they already handle, and the output reads fluently enough that nobody checks it. Chinar is built for Azerbaijani itself, in three models trained on Azerbaijani phone audio: Chinar-F for every call at volume, Chinar-L for transcripts people read, Chinar-P for sites without a GPU. All three run on your own infrastructure.
Azerbaijani is the design target, not a language code passed to a general model.
Three models, one language: pick the size for the job, not the language support.
Chinar-L scores 93–94% word accuracy on clear Azerbaijani speech, measured against human-checked transcripts.
Trained on genuine call-centre recordings rather than read speech.
Runs on your own infrastructure, with no audio sent to a third party.
Chinar is built for Azerbaijani from the ground up rather than adapted from a related language.
Chinar-L reaches 93–94% word accuracy on clear Azerbaijani speech, measured against human-checked transcripts. Noisier audio scores lower.
Chinar-L when a person reads the transcript; Chinar-F when you transcribe every call; Chinar-P when the site has no GPU.
Chinar-L returns punctuated, capitalised Azerbaijani, ready for a document or a compliance record.
Chinar-F is cheap per hour on CPU or GPU — transcribe every call, not a sample.
Global services asked for Azerbaijani give back confident Turkish that looks like a working transcript.
Recordings stay on your infrastructure: no metering, no third-party retention.
See the complete product: problem, features, how it works and deployment.
Request a demo and we will run Chinar against your own Azerbaijani audio — Chinar-F for the volume, Chinar-L for the transcript people read, Chinar-P where there is no GPU.