Georgian STT for call centers, banks and voice agents
One focused product: a streaming Georgian speech recognition engine for telephone audio, deployed inside your infrastructure. It is in development and not yet available.
Status: In development · model in training · early access waitlist open
Who it is for
Contact centers
Transcribe Georgian calls for quality assurance, search and analytics.
Banks and financial services
Call transcription where audio must stay inside your own infrastructure.
Voice agent teams
Real-time Georgian transcripts as input for LLM-based voice agents.
IVR and telephony vendors
Let callers speak Georgian instead of pressing keys through menus.
How it works
Audio in
Your telephony system streams call audio (8 kHz, mono) to the engine running in your environment.
Streaming recognition
The model processes audio in small chunks and keeps context between them, so it does not wait for the speaker to finish.
Text out
Partial transcripts arrive while the person is speaking, followed by final transcripts for each segment.
Planned: a streaming API. Protocol, audio formats and output schema will be defined together with early access partners.
Deployment
On-premises
Runs on your own servers. Hardware requirements will be published with early access.
Your cloud
Deployed in your own cloud account or private network.
Audio stays with you
No audio or transcripts are sent to Graey.
Roadmap
Honest stages, no invented dates.
Data and baselineIn progress
Preparing consented Georgian telephone speech, a baseline model and a held-out evaluation set.
Training and internal evaluationIn progress
Fine-tuning the streaming model on telephone audio; measuring word error rate and latency on held-out calls.
Private early accessPlanned
Testing with a small group of design partners on their own audio, in their own environment.
General availabilityPlanned
Commercial release with documentation, deployment packages and support.
Join the waitlist to hear when early access opens.