RunPod Voice Design Worker

Serverless GPU worker for Qwen3-TTS-VoiceDesign model.

Deploy

1. Build and push Docker image

cd runpod_worker
docker build -t gleitfreude/script-read-voicedesign:latest .
docker push gleitfreude/script-read-voicedesign:latest

2. Create endpoint on RunPod

Go to runpod.io/console/serverless:

• Click "New Endpoint"
• Docker image: gleitfreude/script-read-voicedesign:latest
• GPU: A40 (cheapest that fits 1.7B model) or RTX 4090
• Min workers: 0 (scale to zero when idle)
• Max workers: 1 (or more for concurrency)
• Idle timeout: 60s (keeps warm for 1 min after last request)

3. Add endpoint ID to .env

After creating, copy the endpoint ID and add to .env:

RUNPOD_ENDPOINT_ID=your_endpoint_id_here

4. Switch provider

In the app's Settings panel, change TTS Mode to "RunPod GPU".

Cost

• Cold start: ~30s (model loads into GPU memory)
• Warm request: ~5-10s per voice design
• A40: $0.39/hr → ~$0.003 per call (vs $0.20 on DashScope)