hi,
I am currently learning Georgian, a language with little material available.
I would like to import short videos from youTube into LingQ, but as these are seldom with subtitles, I would like to be able to generate the text from the audio.
does anyone know a good website that can convert audio to text from multiple languages? preferably without having to create any account or anything like that…
regards e
LingQ?
Not sure if I understand your question.
If subtitles are autogenerated, LingQ will convert the audio to text from your Youtube video with Whisper.
If there’s nothing at all and for some reason you can’t import the video, you convert it to mp3 with any free online tool, and import it to LingQ. LingQ will convert the audio to text always using Whisper.
davideroccato is correct. Just use LingQ! Use the LingQ browser extension and click that to import the video. It’ll transcribe behind the scenes. The transcription takes a little bit of time so it won’t be available immediately. If you go to the LingQ website you’ll see it in the pending lessons tab. Once it’s done it will land in the imported lessons tab.
You can also import from the LingQ app on your phone or table…just share to the Lingq app from Youtube app.
davideroccato also has a good fallback in case it isn’t importing for any reason.
hm, ok thanks, I wasn’t aware of that.
I have managed to extract the audio as an MP3 file, I will see if I manage as you both describe. Normally I just use the subtitles generated by youtube, but these videos don’t have any.
Will test it again now more thoroughly as I didn’t get it working earlier.
regards e
Found it…
Is there any settings for the script to be used? The “Audio Transcribe” seems to give me a transcript with the Latin script only, not with the script of the actual language…
EDIT: after a few tests I got it working, I got the Georgian script…
Now I just have to figure out what I did …
Thanks for the help
regards
On import make sure you are choosing the correct language that the lesson will land in? Odd that it would produce Latin…is that where the lesson landed initially? Or was it in the Georgian section, but with Latin script?
Hi, correct, I am in the Georgian section, but it generates text with latin script.
Still experimenting with it…
I managed to get Georgian script by selecting “simplified lesson AI” however the text doesn’t seem to correspond with the audio at all…
Yeah, after you simplify, the simplified version is not going to match the audio.
Maybe there’s a bug for Georgian section that is passing “latin” as the language for the Whisper library translation???
Whisper doesn’t support Georgian, does it?
From OpenAI.com: " We currently support the following languages through both the transcriptions and translations endpoint:
Afrikaans, Arabic, Armenian, Azerbaijani, Belarusian, Bosnian, Bulgarian, Catalan, Chinese, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, Galician, German, Greek, Hebrew, Hindi, Hungarian, Icelandic, Indonesian, Italian, Japanese, Kannada, Kazakh, Korean, Latvian, Lithuanian, Macedonian, Malay, Marathi, Maori, Nepali, Norwegian, Persian, Polish, Portuguese, Romanian, Russian, Serbian, Slovak, Slovenian, Spanish, Swahili, Swedish, Tagalog, Tamil, Thai, Turkish, Ukrainian, Urdu, Vietnamese, and Welsh.
While the underlying model was trained on 98 languages, we only list the languages that exceeded <50% word error rate (WER) which is an industry standard benchmark for speech to text model accuracy. The model will return results for languages not listed above but the quality will be low."
Thanks all for the feedback. I assume that is the reason then… not supported by whisper.
Anyway, I was not aware of this function anyway, so now I am happy to use it for Russian, hope it will add support for Georgian in the future
yes, Georgian is supported by Whisper but the accuracy is not good yet. I wasn’t able to find that list. However, I saw many paid services offering Georgian by Whisper, so I suppose they don’t care.
That could explain why the transcription was doing latin instead of Georgian for that source. If the audio is not clear, if there is music or noises in the background, the AI could have some difficulties to work, especially if the accuracy is low.
You can try with different audio sources and see how it goes.
I was also looking at other resources, for example Microsoft, and they support Georgian as well: Language support - Speech service - Azure AI services | Microsoft Learn
I’m not familiar with those tools, but would Microsoft Edge work with Georgian? If yes, you could use that to transcribe some audio for free. @ericb100 might know more on how to do that by playing the audio.
