OneASR Gateway Deployment & Configuration

OneASR is a high-performance open-source ASR gateway tailored for DuRT. It bundles Faster-Whisper, SenseVoice, FireRedASR, and Paraformer into standard REST/WebSocket APIs, supporting both Realtime Speech and File Transcription.

---

Core Capabilities

Capability	OneASR Support	Key Benefit
Realtime Streaming (Realtime)	✅ Supported (WebSocket)	Low-latency live captioning, serving as an alternative to Apple Speech
Media File Transcription (File)	✅ Supported (High-concurrency batch)	Fast multi-gigabyte media chunking and queue-based processing
Online Video URL Ingestion	✅ Supported	Paste Bilibili, YouTube, or podcast audio URLs directly without manual downloading
Word-Level Timestamps	✅ Supported (Millisecond precision)	Enables karaoke-style word highlighting and fine-tuning in Subtitle Studio
Speaker Diarization	✅ Supported (Speaker 0/1)	Automatically clusters distinct speakers in multi-person interviews
Dynamic Model Switching	✅ Supported	Switch freely between SenseVoice (ultra-fast), Faster-Whisper (multilingual), and FireRedASR

---

Quick Deployment

Option 1: Docker Deployment (Recommended)

docker run -d \
  --name oneasr \
  --gpus all \
  -p 8000:8000 \
  -v ./models:/app/models \
  -e ONEASR_API_KEY="sk-oneasr-secret-key" \
  codeandxv/oneasr:latest

(Omit --gpus all if running on CPU).

Option 2: Python Source Installation

git clone https://github.com/codeAndxv/OneASR.git
cd OneASR
pip install -r requirements.txt
uvicorn app.main:app --host 0.0.0.0 --port 8000

---

Connecting in DuRT

OneASR can be added to both Realtime Speech and File Transcription:

1. Open DuRT, navigate to Settings ➔ AI Capabilities;
2. Select File Transcription (or Realtime Speech) ➔ click "Add Service" ➔ select OneASR;
3. Enter connection credentials:
   • Service Name: e.g. LAN GPU OneASR;
   • Service URL: http://localhost:8000 (or your LAN IP, e.g. http://192.168.1.100:8000);
   • API Key: Enter your key if configured, or leave blank;
4. Click "Test Connection" (DuRT 3.0 supports OneASR v0.1.0 ~ v0.1.x protocol);
5. Choose your default model (e.g. sensevoice-v1 or whisper-large-v3) and save.