Credits
The tools and data behind Bhat's audio and content, and how they were made.
Audio
Lesson and conversation audio is computer-generated speech, not a recording of a person. It is being produced with the text-to-speech feature of the Google Gemini API, with a different voice for each character.
Clips that have not been regenerated yet use Piper, an open-source text-to-speech engine, with the Nepali voice model ne_NP-google-medium. That model was trained on the High quality TTS data for Nepali (OpenSLR 43) data set published by Google, licensed under Creative Commons Attribution-ShareAlike 4.0.
Lesson content
Lessons are currently drafted with AI and are waiting for review by a native Nepali speaker. Content that has not been reviewed is labelled in the app.
Your recordings
When you record yourself in a speaking exercise, the recording stays on your device. It is not uploaded or stored by Bhat, and Bhat does not score pronunciation.