royshil/obs-localvocal

What it solves

LocalVocal is an OBS plugin that provides real-time speech-to-text transcription and translation. It allows streamers and content creators to generate live captions on screen or save them to files without relying on cloud services, ensuring privacy and eliminating network latency or costs.

How it works

The plugin integrates OpenAI's Whisper model via Whisper.cpp for efficient local execution on CPUs and GPUs. For translation, it utilizes CTranslate2. It supports a wide range of hardware acceleration backends, including CUDA (NVIDIA), hipBLAS (AMD), Metal (Apple Silicon), and Vulkan, allowing it to run on various hardware configurations without requiring a high-end GPU.

Who it’s for

It is designed for OBS Studio users who need live, private, local transcription and translation of their audio streams into multiple languages.

Highlights

• Local Processing: All data stays on the machine, requiring no cloud costs, network connection, or GPU (though GPU acceleration is supported).
• Multi-language Support: Transcribes audio in 100 languages and translates captions in real time.
• Flexible Output: Captions can be displayed as OBS text sources, sent to .txt or .srt files, or sent via RTMP stream to platforms like YouTube and Twitch.
• Hardware Versatility: Supports a vast array of CPU and GPU backends across Windows, macOS, and Linux.
• Customizable Models: Users can download built-in Whisper models or provide their own GGML Whisper model files.