Hacker News new | past | comments | ask | show | jobs | submit login
Show HN: Language Learning from YouTube content (fluentsubs.com)
40 points by ph4evers 10 days ago | hide | past | favorite | 6 comments
Hi!

I'm living as an expat in France and I had a hard time following the news. So I've been working on Fluentsubs that turns YouTube content into a language learning experience.

YouTube videos up to 20 minutes can be transcribed and listened to. At the same time I made it easy to lookup certain words and automatically add spaced repetition cards using the given context. This means that you can also rehearse with a real voice and not with AI dubs.

This works especially well for the somewhat smaller languages that have a small track on Duolingo, and now are desperately looking for content (Norwegian, Finnish, Swedish, etc.)

I'd love to hear your thoughts and feedback: https://fluentsubs.com






Always excited to see others working in the language learning space! (Korean & English in my case)

Signed up and played around a bit, but was unable to get a video transcribed. It kept popping up with a "Error loading video. Please try again later." issue. I tried entering a URL and was given that the audio quality was too low.

Separately, I'm curious how you're approaching quality control in your translations for the words. While building my app, the GPT models regularly get translations wrong for Korean and miss nuance and require human-in-the-loop before sending to the user.

Great work shipping! It looks great!


Thanks a lot for trying!

I tried to upload some Korean videos and indeed it fails on transcription quality. I've built in a check to filter out low confidence videos. I'm using AssemblyAI under the hood but I think it is too strict. I'll investigate.

"Quality control" is now simply a confidence score check and a warning that translations can be wrong. As long as it is mostly right I believe there can be value. However, the experience quickly degrades if it is wrong on every sentence.

EDIT:

I'll try to deploy a new version today. I'm too scared to do it right now because I already brought the server down by building a Docker container. I'm not hosting on a super large node.


Congrats on your release! I'm working on a parallel text project where LLMs are used for a number of translation and transformation tasks (www.abal.ai) so it's nice to see what others are coming up with.

I'm curious how you landed on your current set of supported languages? Notably, I see that South Asian languages are not supported beyond Indonesian. Have you already given ones such as Hindi and Bengali a try?


Thanks! abal.ai looks great!

I'm using AssemblyAI and DeepL under the hood. I recently switched from a provider that didn't support Hindi. I'll update it soon to support it.


No support for Arabic?

I'm working on it! The biggest blocker is the reading direction but that should be solvable. Both AssemblyAI and DeepL support it.



Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: