Hacker News new | ask | show | jobs
by davidzweig 1454 days ago
We made something a couple of weeks ago, you might like it, needs login but is 100% free. Endless (depending on language) listening exercises, sentences are selected (from tatoeba) based on vocabulary (no selection based on grammar yet). It's also integrated with a flashcard system (also free). If you set a small vocab (100 words) and do lots of cards, you start to get a feel for the grammar of a language. Something like 40 languages are supported.

https://dev.languagereactor.com/phrasepump (this is the dev site, will go to prod in a couple of weeks)
2 comments
yorwba 1454 days ago
As a contributor to Tatoeba, I'm always interested in checking out how people use our data.

The interface needs work to be usable on mobile, but I guess you know that.

I'm not sure whether the TTS you're using is really good enough to be usable for learning a language. I tried it in German and noticed quite a few TTS bugs, like pronouncing "Tu" as if it were an acronym "TU" instead of the verb "Do", as well as switching to an English pronunciation next to English names (pronouncing "Hallo Roger" more like "Hello Roger").

Tatoeba also has many recordings in quite a few languages (e.g. German https://tatoeba.org/en/audio/index/deu ) You should try using them if possible. For download instructions, see https://tatoeba.org/en/downloads

link
davidzweig 1454 days ago
Oh, thank you for your contributions! Mobile, we're planning a RN app (site is MobX/React), but we got our hands full at the moment (who wants a job?). TTS, I used to be of the opinion that TTS is something that should be kept far away from langauge learners. But, it got so damn good: https://azure.microsoft.com/en-us/services/cognitive-service... At least for many langauges. Didn't check the German much, interesting. For Hungarian, it screws up question intonation. But often it just seems completely human-like. I was concerned about the license for the audio recordings, and I assume recording quality could be mixed, but I'll take a more careful look.
link
sanqui 1454 days ago
That looks great so far! Pair it with open cloze with translation once you determine that the user should know the word and you've got magic going. You could even "beep" the word out of the recording (I'm assuming they're TTS).

Edit: typo

link
davidzweig 1454 days ago
We use microsoft TTS, was trying to figure out if they can give time ranges for beeping out words.. didn't see it. Holy grail would be just generating audio files to listen to, no front end at all. :)
link