Kokoro-FastAPI

Use it when

• You need an OpenAI-compatible text-to-speech endpoint running on your own hardware so audio never leaves the machine and there is no per-character bill
• You want to narrate long-form text into audio with word or chunk level timestamps for captions or read-along playback

What it solves

Not the fit when

• Speech recognition or transcription
• Training or fine-tuning new voices; limited to bundled Kokoro-82M voicepacks and weighted mixes
• Apple Silicon GPU acceleration inside Docker (native UV run required for MPS)
• speech-to-text transcription
• cloning a specific person's voice from samples
• hosted multi-region audio infrastructure

Install

docker run -p 8880:8880 ghcr.io/remsky/kokoro-fastapi-cpu:latest (or --gpus all with ghcr.io/remsky/kokoro-fastapi-gpu:latest for NVIDIA)

Invoke

POST http://localhost:8880/v1/audio/speech, or use the OpenAI Python client with base_url http://localhost:8880/v1 and model kokoro; WebUI at /web

Alternatives

No reviewed alternatives recorded yet.