Where Are the New Languages?
I keep getting asked, "Where are the new languages you promised?"
Fair question.
Short answer: I started working on them, then ended up rebuilding the whole core pipeline.
I was close to finishing it when I hit that point where you look at something and think: yes, it works, but it's still not good enough.
So I tore down the processing system and rebuilt it properly.
Some of you had pointed out problems like:
- phrases getting cut too early
- awkward splits
- recognition being slightly off
Those were real issues. The whole thing depends on automatic audio processing, and that's just a hard problem.
The new system handles those cases much better. Fast speech, interruptions, overlapping dialogue - it keeps phrases a lot cleaner and more natural now.
I was testing it on movies like Pulp Fiction, where the dialogue is messy, fast, and very real, and after a few minutes it was obvious the output just felt better.
There is a catch, though.
To get the benefit of the new system, I can't use it only for new languages. I have to reprocess everything.
So right now I'm rebuilding the entire database from scratch:
- all videos
- all phrases
- all languages, including English
That's why the new languages are taking longer than I originally expected.
But sometimes you realize that what you've built over months still isn't at the level you want, and the only way forward is to redo it properly.
The upside is that the whole project should come out much better, not just the new languages.
If you want to see how things are evolving:
I'll share examples soon. The difference is very noticeable.