feros
Deploy

GPU and Speech Inference

Optional self-hosted STT/TTS inference stack for feros.

GPU and Speech Inference

The inference module supports self-hosted STT/TTS services.

Build and Run

From repository root:

make inf-build-stt
make inf-build-tts
make inf-stt
make inf-tts

Default ports:

  • STT: 9001
  • TTS: 9002

Typical Usage

Use this stack when you want local GPU inference instead of external provider APIs.

Hardware Notes

  • NVIDIA GPUs and NVIDIA Container Toolkit are required.
  • Running STT and TTS on separate GPUs improves stability under barge-in traffic.

On this page