Academic ASR training pipeline · Turkic languages
A non-commercial training pipeline for universities and researchers. Students record read speech under their instructor’s supervision; the recordings are used to improve transcription models.
01 — Pipeline
Non-commercial · for research
bilgi.tech is building a shared, non-commercial speech recognition (ASR) training pipeline for universities and researchers.
In the reading panel (data.bilgi.tech), students read assigned texts aloud. The texts to be read are approved by an instructor or reviewer, and every recording goes through an automatic quality check. Recordings that pass are used to improve transcription models.
A student reads the text in the panel in their own voice.
Each recording is checked automatically (audio level, match with the text); if needed, it is read again.
Recordings that pass are used to improve transcription models.
02 — Languages
What works today and what is planned, kept apart
These languages are not supported yet. Our goal is to add them.
03 — Join
Universities and research institutions only
Apply to take part with your institution.
data.bilgi.techReading and review panel for students and instructors. Sign-in required.
kkasr.bilgi.techKazakh and Russian transcription. Invited accounts only.
Applications are received through the form and reviewed by our team.
04 — Updates
News from the project
A trial version that turns Kazakh audio into text and separates speakers is running at kkasr.bilgi.tech. For now, only people with an account can sign in.
Details →The reading panel (data.bilgi.tech) is now open. 30 Kazakh reading texts are ready for the first phase; two universities are taking part.
Details →Students made their first recordings in the reading panel: about 19 minutes of audio across two texts, from one institution. Recordings go through an automatic quality check.
bilgi.tech also runs a Turkish-language AI news digest: news, Turkish language models, model releases, community and blogs.
Open the news page →