Clone a voice
Upload a clean 6–20s clip (mono, no music). It becomes a reusable voice on the Narration tab.
Your cloned voices
Dub a video / audio file
Transcribe → translate → re-voice, keeping the speaker's voice. Requires Whisper + a translation backend on the server.
Real-time
Streaming TTS works with Kokoro today. Full-duplex conversation uses NVIDIA PersonaPlex-7B on a GPU box (wire it in app/services/realtime.py).
Full-duplex conversational agent — PersonaPlex listens & speaks at once (~0.2s latency, handles interruptions).
Endpoint
/ws/converse is wired; drop the model in on your GPU host to activate.