Local Sub

This is a Python tool to quickly and simply transcribe some video / audio file into an SRT file using local models with Windows.

TEN VAD is used to aggressively detect speech segments and their timestamps. Then a custom tuned ASR model is used to transcribe each speech segment into text.

That's it. No cloud services needed.

Quickstart

1. Clone the repo and install the dependencies
2. Download the latest llama.cpp Vulkan binaries and place the contents in the src/binaries/llama folder.
3. Download my tuned ASR models here and place them in the src/binaries folder.
4. Run with e.g: python cli.py "YOUR_SAMPLE.wav"
5. After some time, the processed SRT file will show up in the executable folder as YOUR_SAMPLE.srt

Notes

Expect 4 GB VRAM for fast results. Note this ASR model could still be tuned further.