• 

  what the subs look like in VLC

• 

  CLI

Autosub

Andy Jarean

Track

Best Social Impact \ Helps people learn languages more efficiently by transcribing speech into text

Overview

A tool used to automatically generate .srt subtitle files (viewable in video players such as VLC) for videos \ This was originally intended for language learners, hence why it's Japanese-focused. I had originally planned to make this integrate with Anki to create flashcards for difficult words, but that became out of scope in favor of the subtitling function.

Usage

supply your own model and silero_vad.onnx file and put them in a /models/ folder \ the model I'm using is sherpa-onnx-zipformer-ja-reazonspeech-2024-08-01\ also supply your own videos

call it like this and replace ./my_video.mp4 with your video\ adjust the params however you like (look in main.py for params)

python ./main.py `
--vad-threshold 0.3 `
--min-silence-duration 0.15 `
--min-speech-duration 0.1 `
--max-speech-duration 3.0 `
--segment-padding 0.2 `
--force-max-duration 6.0 `
"./my_video.mp4"

Flaws/Plans

• Originally this was planned as a language learning tool, with Anki (flashcard) integration and a GUI but I didn't really have time for that
  • hence why I chose Japanese
• the subtitles aren't perfect
  • some lines are cut out, it performs poorly when many people are speaking at once, etc
  • for the most part, some of these issues can be remedied by playing around with the parameters. honestly I haven't found the optimal settings yet but the above seems to work okay?

Built With

• ffmpeg
• python
• sherpa-onnx