Apppublic

nxdev-org/nemotron-speech-streaming-en-0.6b

sourceHugging Faceupdated 9mo agoView on Hugging Face

0likes

App README

Nemotron Speech Streaming

Real-time speech recognition powered by NVIDIA Triton Inference Server.

Features

—Real-time streaming ASR: Bidirectional streaming for live transcription
—File upload support: Transcribe WAV, MP3, OGG, WebM files
—Beautiful UI: Modern React interface with NVIDIA branding
—WebSocket bridge: FastAPI server bridging browser to Triton

Environment Variables

Variable	Required	Description
`NGC_API_KEY`	Yes	NVIDIA NGC API key for authentication
`FUNCTION_ID`	Yes	NVCF function ID for the ASR model
`VERSION_ID`	No	NVCF function version ID
`TRITON_URL`	No	Triton server URL (default: `grpc.nvcf.nvidia.com:443`)
`MODEL_NAME`	No	Model name in Triton (default: `nemotron_asr`)
`PORT`	No	Server port (default: `8080`)

Local Development

bash

# Install Python dependencies
cd bridge
pip install -r requirements.txt

# Build React frontend
cd web
npm install
npm run build

# Run the server
NGC_API_KEY=your_key FUNCTION_ID=your_function_id python -m bridge.main

Docker

bash

# Build
docker build -t nemotron-speech .

# Run
docker run -p 8080:8080 \
  -e NGC_API_KEY=your_key \
  -e FUNCTION_ID=your_function_id \
  nemotron-speech

Architecture

┌─────────────┐     WebSocket     ┌─────────────┐     gRPC      ┌─────────────┐
│   Browser   │ ◄──────────────► │   FastAPI   │ ◄───────────► │   Triton    │
│  (React UI) │                  │   Bridge    │               │   Server    │
└─────────────┘                  └─────────────┘               └─────────────┘

License

Apache 2.0 - See LICENSE file for details.