Quick Start
Deploy Whissle Gateway
One command to run the full voice AI stack on any NVIDIA GPU:
docker run -d --gpus all -p 9000:9000 \ -v whissle-trt:/tmp/trt_engines \ whissleasr/whissle-gateway:standard
The gateway is ready at http://localhost:9000 after ~2 minutes (TensorRT engines build on first run, cached after).
What's Inside
| Component | Details |
|---|---|
| ASR | 23 languages, 440ms TTFT, TensorRT accelerated |
| LLM | 3B params, 265 tok/s on GPU, OpenAI-compatible API |
| TTS | Human-quality Orpheus, EN + Hindi, 230ms TTFB |
| Diarization | ECAPA-TDNN speaker encoder, multi-speaker separation |
| Metadata | Emotion, intent, age, gender, entities per utterance |
Verify
# Health check curl http://localhost:9000/ # Quick ASR test curl -X POST http://localhost:9000/asr/transcribe \ -F "file=@audio.wav" -F "language=en"