Voxta Parakeet
NVIDIA's local speech-to-text — accurate, fast, punctuates itself.
Parakeet is NVIDIA's open-source neural STT, covering 25 European languages with punctuation and capitalization it produces itself. It runs locally.
Setup
Add the service
Manage Services → + Add Services → Parakeet → Add. The model downloads on first use.
Pick a model
The standard TDT 0.6B v3, or its int8 build — smaller and faster to load, at some accuracy.
Turn-taking is tuned by end of speech silence, max segment length and partial interval. Voice activity detection uses Silero VAD.
No emotion or audio-event detection — see SenseVoice if you want that. It also handles one transcription at a time.