Photo of DeepakNess DeepakNess

Groq subtitle generator

Unproofread notes

Only a week ago, Groq has released an open-source tool to generate subtitles for videos in multiple languages, and it works like a charm. It has a web interface to upload videos, select languages, generate subtitles, and preview videos with subtitles.

I also found an informative post explaining how it works, and the post also shows examples of the quality of the subtitles. It's perfect, and 10x better than the YouTube's default auto-caption tool.

Backend tech-stack is:

And for frontend:

The best thing is, it extracts the audio only (by using FFmpeg) when you upload the video, and then the .wav audio is sent to Groq to process. It uses the OpenAI’s Whisper Large V3-turbo model but the Groq-hosted version, and then Qwen3-32b model by Groq is used for translating into the selected language.

One issue is, as of now it creates burned-in subtitles only, but still very impressive.

Update: I tweeted asking whether one can create soft subtitle and download the file as a .srt file, and they actually added a new branch that does this.

It would be really fun to use a local LLM to do this, and also build a desktop app to do this locally. I am already thinking about this, and might build one.

Comment via email