A self-hosted Telegram bot that transcribes voice messages to text using NVIDIA's Parakeet TDT 0.6B v3 model (via sherpa-onnx). Supports 25 European languages. Runs on macOS, Linux, and Windows.
Privacy by design — recognition runs entirely on the machine where the bot is deployed. Voice data never leaves your server: no third-party speech API, no cloud upload, no logs stored to disk unless you explicitly enable file logging. Once the bot replies, the audio is gone.
The model is downloaded automatically on first run to ~/.voice_yap_bot/
and unloaded from memory after a period of inactivity
(1 hour by default), then transparently reloaded on the next message — so RAM
is only held while the bot is actively transcribing.
Content type
Image
Digest
sha256:a9f04488f…
Size
43.6 MB
Last updated
4 months ago
docker pull evg4b/voice_yap_bot:0.1.0