nemotron-voice-agent-deploy

name: nemotron-voice-agent-deploy description: Deploy Nemotron Voice Agent on Workstation (x86), Jetson Thor, or Cloud NIMs. Real-time speech-to-speech using NVIDIA ASR, TTS, LLM with WebRTC/WebSocket transport.

Nemotron Voice Agent Deployment

Real-time conversational AI voice agent using NVIDIA NIMs (ASR, TTS, LLM) with WebRTC (default) or WebSocket transport.

Deployment Flow

Always verify hardware first, even if user mentions a specific platform.

STEP 1: Hardware Detection

nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>/dev/null

Cloud NIMs (No GPU)

cd nemotron-voice-agent
git submodule update --init
cp config/env.example .env

Export your NVIDIA API key:

export NVIDIA_API_KEY=your-api-key  # Get from https://build.nvidia.com

Then edit .env:

NVIDIA_LLM_MODEL=nvidia/nemotron-3-nano-30b-a3b  # Cloud model name

If user requests WebSocket transport, also add to .env:

TRANSPORT=WEBSOCKET

docker compose up --build --no-deps -d python-app ui-app
# WebRTC: http://localhost:9000
# WebSocket: http://localhost:7860/static/index.html

Note: Deployment may take 30-60 minutes on first run.

If user requests Multilingual mode, also add to .env:

ENABLE_MULTILINGUAL=true
ASR_CLOUD_FUNCTION_ID=71203149-d3b7-4460-8231-1be2543a1fca
ASR_MODEL_NAME=parakeet-rnnt-1.1b-unified-ml-cs-universal-multi-asr-streaming

Remote Access: ssh -L 9000:localhost:9000 user@host or http://<HOST_IP>:9000

STEP 2: Platform Detection (if GPU detected)

uname -m  # x86_64 → Workstation, aarch64 → Jetson
cat /etc/nv_tegra_release 2>/dev/null && echo "Jetson"

Note: Multilingual mode available on Workstation with WebRTC transport only.