Skip to main content

OBS Omniversify TTS

OBSPluginTTSC++
View on GitHub

Overview

A high-performance OBS Studio plugin designed for real-time chat preview and local Text-to-Speech (TTS) integration, optimized specifically for AMD GPUs on Arch Linux.

Provides streamers with a low-latency, GPU-accelerated TTS solution that offloads processing from the CPU to ensure smooth gameplay in demanding titles.

Features

  • Real-time Chat Preview — View and manage chat messages directly within OBS
  • Local TTS Processing — Fully local execution, no cloud latency or privacy concerns
  • AMD & Cross-GPU Optimization — Leverages ROCm or Vulkan for hardware-accelerated inference
  • Zero CPU Hogging — Minimizes CPU usage during intense gaming sessions
  • Simulation Mode — Built-in debugging tool to test voices and emoji handling before going live
  • Emoji Support — Visual preview with emojis; TTS engine intelligently ignores them
  • Engine Selection — Toggle between ROCm/Vulkan engines and voices from the OBS Dock

Architecture

┌──────────────────────────────────────────────────┐
│              OBS Plugin (C++/Qt)                  │
│  ┌─────────────┐  ┌──────────────────────────┐   │
│  │  Chat Dock   │  │  Engine Manager          │   │
│  │  (Qt Widget) │  │  ROCm / Vulkan / Sim     │   │
│  └─────────────┘  └──────────┬───────────────┘   │
│                              │ IPC (WebSocket)    │
├──────────────────────────────┼───────────────────┤
│                 TTS Backend (Python)              │
│  ┌─────────────┐  ┌──────────────────────────┐   │
│  │  ROCm/ONNX  │  │  Vulkan/ONNX             │   │
│  │  (Piper)    │  │  (Piper/Silero)          │   │
│  └─────────────┘  └──────────────────────────┘   │
└──────────────────────────────────────────────────┘

Tech Stack

CategoryTechnology
PluginC++17, OBS Studio API, Qt6
ROCm BackendPython, onnxruntime-rocm, torch-rocm
Vulkan BackendPython, onnxruntime-vulkan, ncnn
TTS ModelsPiper, Silero
IPCWebSockets / Local Sockets
BuildCMake, build.sh

Project Structure

src/
├── plugin-main.cpp     # OBS plugin entry point
├── tts-dock.cpp        # Qt dock UI (chat preview, controls)
└── tts-dock.hpp        # Dock header

backend/                # Python TTS inference backends
data/                   # Model files and configs
build.sh                # Automated build script

Roadmap

Phase 1: Simulation & Mocking (Current)

  • Basic OBS Dock UI in C++/Qt
  • Simulation Mode with text input and test button
  • Backend-agnostic TTS handler
  • Emoji stripping and rendering

Phase 2: Local TTS Engine Integration

  • Python environment via uv
  • ROCm-accelerated TTS model integration
  • IPC between C++ plugin and Python backend

Phase 3: Real Chat Integration

  • Connect to Twitch/YouTube via OBS hooks or WebSocket
  • Message queuing for high-traffic without overlapping audio

Phase 4: AUR Package & Auto-Launch

  • Package plugin and backend as an AUR PKGBUILD for easy installation
  • Integrate systemd user service or OBS startup script to spawn the TTS backend automatically on login
  • Automatic backend discovery: OBS plugin connects to backend on load without manual command execution

Quick Start

git clone https://github.com/phaylali/obs-omniversify-tts.git
cd obs-omniversify-tts
./build.sh

Requires OBS Studio, ROCm or Vulkan drivers on Arch Linux.

Design Decisions

  • Emoji Handling — Rendered in the preview dock but stripped from TTS input
  • AMD Focus — ROCm/HIP explicitly targeted; Windows/Nvidia abstracted for future
  • C++ for Plugin — Maximum stability for OBS integration
  • Python for Backend — Leverages rich AI ecosystem on Linux