LocalVocal

LocalVocal adds live speech transcription and translation to OBS for streamers and people recording video. It runs Whisper on your own computer, so local captioning works offline and keeps audio processing on your machine. The plugin is open source under GPL-2.0.

Captions can appear in OBS text sources or travel with an RTMP stream to YouTube or Twitch. For recordings, it can save transcripts as TXT or SRT files and synchronize captions with OBS recording timestamps. Partial transcriptions show text while speech is still being processed, and caption filters can remove or replace selected wording.

Translation can stay local through Whisper's built-in translation or neural translation models using CTranslate2. It also supports cloud translation through DeepL and OpenAI, plus translation with a local LLM. Choosing a cloud provider sends translation requests outside your machine and requires an internet connection.

LocalVocal runs on Windows, Linux and macOS, including Intel and Apple Silicon Macs. A GPU isn't required: whisper.cpp supports CPU processing, with optional acceleration on supported NVIDIA and AMD GPUs or through Apple's Metal backend. Vulkan also supports some integrated GPUs.

You can use external GGML Whisper models, including fine-tuned models from Hugging Face, rather than relying only on the bundled Tiny.en model. That gives you a choice of speech models for the languages you caption.