VibeVoice + RVC Speech Generation Service

Description

Local CLI and REST API for text-to-speech generation using VibeVoice and optional processing via RVC (Voice Conversion).

---

CLI

Example of execution:

 cli.py --text "Hello world" --reference-audio reference.wav --model-path 1.5B --out output.wav [--rvc-model rvc.pth]

Parameters:

• --text — text for synthesis or --text-file to read from file
• --reference-audio — reference voice (wav)
• --model-path — path to the VibeVoice model
• --out — output wav file
• --rvc-model — (optional) path to RVC model
• For other parameters, see --help

---

REST API (FastAPI)

Starting the server:

docker build -t vibevoice-rvc .
docker run -p 8000:8000 vibevoice-rvc

Endpoints

• POST /generate — speech generation (VibeVoice)
• POST /convert — WAV processing via RVC
• POST /pipeline — full pipeline (VibeVoice → RVC)

For request examples, see Swagger UI: http://localhost:8000/docs

---

Requirements

• Python 3.10+
• CUDA (for acceleration, optional)
• ffmpeg

---

Tested on Win 11 + RTX 3090 and Ubuntu Server 24.04 LTS + RTX 3060 (4-bit Quant)

---

Notes

• All computations are local, no internet connection required.
• To support different voices, use different reference audio and/or RVC models.
• We dont need LoRA support because we use RVC (But it will be useful for supporting other languages.)