nemotron-voice-agent-deploy

To deploy a real-time speech-to-speech voice agent using NVIDIA NIM, you orchestrate ASR, TTS, and LLM components via WebRTC or WebSocket. You start by copying the example environment file to .env, setting your NVIDIA_API_KEY, and running docker compose up.

Yes, you can run a real-time voice agent on Jetson edge devices. The deployment process automatically detects supported hardware like Jetson Thor and applies the correct platform-specific compose commands to facilitate local speech-to-speech processing.

WebRTC is used by default for local workstation-style access to the voice agent, providing near real-time audio. WebSocket serves as an alternative transport protocol when explicitly requested for the speech-to-speech deployment.

You configure a multilingual voice agent by setting optional multilingual flags in your environment configuration. This adjusts the ASR identifiers and model names to enable multilingual voice behavior on supported platforms.

You do not need a local GPU if you use Cloud NIMs. The deployment features automatic platform selection, routing to Cloud NIM deployments when hardware capabilities differ or a local GPU is unavailable.

Your NVIDIA voice agent stack may fail if the environment is misconfigured. Ensure you copy the example environment file to .env, correctly set the NVIDIA_API_KEY, select the proper models, and run docker compose up --build --no-deps -d python-app ui-app.