NVIDIA PersonaPlex
NVIDIA's 7B full-duplex voice conversation model — customizable voices and personas
NVIDIA's 7B full-duplex voice conversation model with customizable voices and personas, backed by an enterprise vendor.
Voice AI: lifelike TTS, accurate ASR, and real-time voice agents.
Key features
- Real-time full-duplex speech-to-speech conversational model
- Persona control via text role prompts and audio voice conditioning
- Based on Moshi architecture; natural low-latency spoken interactions
- 20 pre-packaged voice embeddings (natural and variety, female/male)
Good for
- Researchers building persona-driven voice assistants
- Customer service voice agents with consistent personas
Stack & tags
PythonPyTorchspeech-to-speechconversational-aifull-duplexpersonavoice-conditioningopen-weights
Quick start
pip install moshi/.
export HF_TOKEN=<redacted>
SSL_DIR=$(mktemp -d); python -m moshi.server --ssl "$SSL_DIR"From the project README on GitHub.