I just give an idea for the LingQ’s Team. (probably you thought about it already)
If Speechify is very good as they say: they have 15 languages available.
If you pay for Speechify and for gpt-4, you can create content for hundreds of students in a blink of an eye.
It’s a lot less expensive because you pay once but you can create a ton of content compared to if we pay for it individually.
You can create stories in one language only, then just copy/paste and translate the same story in all the other languages and then use Speechify for the audio.
And if you want to, you can create videos for those or others with SYNTHESIA. For all other languages, Synthesia has 120 languages!
YOU CAN CREATE:
new mini-stories
different mini-stories for different levels
new courses for different genres (sci-fi, romance, adventure…)
summarizations of long Youtube videos
summarizations of tons of books.
news from creative commons blogs.
I mean, seriously, in one month you could create a lot of content.
Does it make sense?
4 Likes
There are multiple issue:
translating between languages is nuanced and depending on how similar the languages are and how difficult the sentence is, it could leave many variables on the table that AI will cut out/misinterpret between translations
Secondly, I don’t know anything about speechify, but I wouldn’t trust fledgling AI voice generators to accurately capture the nuances of many different languages. I guess these problems could be solved have by having native human oversight on approving translations, and then the same people approving of the generated voice, but that requires time, and of course, money. Maybe they could add this in some “experimental” tab where native users could approve of AI generated content for translation and sound accuracy before publishing them to be used as learner content?
3 Likes
It doesn’t matter the nuances on translation, you don’t have to be perfect with the stories translated, the purpose is not creating bilingual texts but just new stories for each language. So, the person creating a story doesn’t have to create a new similar story for each language on chatgpt all the time, he just creates one story and then translate it for different languages automatically.
A student learning French is gonna read it in French, and a student learning Spanish won’t bother about another version on another language and so on.
I use already TTS on many articles in different languages and they are becoming quite good. But Speechify they say it’s a lot better. When I listen to some TTS on M.Edge they are quite better than the ones I use.
For beginners it shouldn’t be a problem, I guess.
Yes, verifying from the librarians could be a good thing to do. Actually, librarians could create these courses by themselves if LingQ would give them access to the platforms.
3 Likes
translating between languages is nuanced and depending on how similar the languages are and how difficult the sentence is, it could leave many variables on the table that AI will cut out/misinterpret between translations
You should try it out. I’ve just tested this with (the previous) ChatGPT by asking him to write a horror story in English, Polish, and Swedish, and it did great, with a few mistakes in Polish (none of which were grave). ChatGPT is generally very good at both translation and writing as long as you don’t expect a great prose, and if you stick to popular langauges than there’s even less reason to be worried.
Secondly, I don’t know anything about speechify, but I wouldn’t trust fledgling AI voice generators to accurately capture the nuances of many different languages.
I think they have 30 or so languages, though I’ve never tested them. There are 4 samples on top of their homepage, one of them sounds very familiar and I don’t like it, the female one is very mediocre, and the remaining 2 are very good.
Remember that purpose of this would be to create more content for beginners and intermediates. Savvy people will be able to figure it out themselves if they really want to use this set of tools, especially the ChatGPT part, without waiting for LingQ to implement anything.
/edit I take back what I said about Speechify. They let you play around with some voices and they’re… all bad 
1 Like
Why use speechify (which based on what you, krampusx and Gigusek have said so far, sounds at best average) when you can use the “edge” voices. You can use it outside of Edge, both with their api and below someone has a python program with it (link below). I haven’t heard anything that sounds better than the Edge voices at this point personally. While LingQ’s service seems to have gotten better, it still sounds very unnatural at times.
1 Like
Yes, I know nothing about the technical stuff but one or the other doesn’t matter. They can just create a ton of content with these new tools.
Speechify, they say, it’s very natural with natural voices and even actor voices. But another better wouldn’t hurt.
1 Like
yes, and you probably didn’t test with gpt-4 that’s way better than gpt3.5. So, I suppose translations are even greater as gpt-4 has trillions of data.
I believe a combination of Speechify, Edge and synthesia would be the best. Probably each one of them are stronger in few languages then others.
1 Like
Yeah, that’s why I mentioned using the old model (though I don’t know how many people know how much discrepancy there is between the two, but it’s huge)!
I think the main focus should be on implementing content generation using GPT, and the LingQ team could be pretty creative with it if they wanted to, such as letting users generate their own stories on demand and using suitable vocabulary at their level.
I wouldn’t go much beyond that though, to be fair, though that might be just me. I wouldn’t bother with the AI video and voice,
1 Like
I actually find that reading and listening with TTS helps me a lot more than just reading. And I’m talking as advanced user by reading articles or books in French, Spanish and English. It helps me increasing the focus, for example, before going to bed at night. It’s quite handy to have it.
The fact is that LingQ’s TTS or iOS TTS is less natural than Edge or maybe Speechify (I haven’t tried the actors feature). So it would be good to have it already done. Imho.
Admittedly, I’ve never heard of speechify so I couldn’t say one way or the other, but yes, whichever is better or more natural!
1 Like