What have people's experiences been with the whisper function on LingQ?

What have people’s experiences been with the whisper function on LingQ? My partner (a native speaker of Turkish, the language I’m learning) has given me a list of YouTube channels he believes I will enjoy. Many of them don’t have subtitles, or they have autogenerated subtitles only (which I find difficult to follow in Turkish given the - for me - complicated word order).

I thought I’d try out LingQ’s Whisper functionality. This is my first time using any sort of automatic transcription service, so I’m really curious to see if it can manage to make real sentences, with punctuation.
I’ve exported the audio of two videos to MP3 format, and I’ve added them to be transcribed. Whisper is currently transcribing them. Both videos are about 10-12 minutes long, and they have been transcribing for about two hours now. No part of the lesson for either video is available, but the lesson ‘placeholder’ is available in the ‘imported lessons’ and ‘continue studying’ tabs.

Is that normal? Should I only be uploading one video at a time?

If this is to be expected I will make a list and upload what I want to study in advance, but really just checking I’m doing it right!

1 Like

Hi, @charrich87!

I have used the whisper functionality with one audio I downloaded from a video that didn’t have subtitles, and based on that experience, the experience you described sounds normal to me.

  • Uploads
  • Takes a while to transcribe but has a lesson ‘placeholder’ available in ‘imported lessons.’

I usually assume it takes a while and go and do something else (i.e., I import those videos before I need to study them and study something else). I’m generally not “waiting around” and timing how long it takes, so I can’t say that about two hours is/isn’t normal (maybe someone else can chime in there), but it isn’t instantaneous (for lack of a better word).

Uploading one at a time might make it go a bit faster, though.

I am choosing videos with embedded subtitles or alternative sources to transcribe the most accurate subtitles (meaning an import source that isn’t supported with the link extension because it’s not Youtube/Netflix). This is because there are still (understandably) transcribing mistakes by the Whisper functionality.

There are not a ton of mistakes (as far as I can tell), but videos with a single person talking at a time tend to transcribe better than those with lots of people talking at the same time or a lot of background noise (again, as far as I can tell based on my limited interaction).

I think the punctuation is a hit or a miss. (I’m using it to learn Korean, by the way.) Sometimes it’s right; sometimes it’s not there; sometimes it added it where it’s not needed/in the embedded subtitles or the official subtitles from the other source I use.

Overall, it makes transcribing great content I love (but would otherwise not be able to use due to the lack of subtitles on Youtube overall, especially in Korean—though you can find some great stuff when you dig for it) easier to do.

Hope this helps! Happy Learning! :slightly_smiling_face:


Thanks for your reply!

Looks like I may have uploaded the same file twice! :smile:
Either that or each account should only be uploading one file at a time. Might test it out again to be sure, but for now I’m assuming human error on my part.

Thanks for confirming the process seemed normal.

Had a quick look and listen to the first five minutes or so. The narrator of this video is very clear and enunciates very well, and it looks like the transcription has benefited from that. It has missed at least one entire sentence (no idea how!), and has transcribed some words incorrectly (seems to have transcribed ‘popular’ phrases rather than the less common phrase the narrator has used. Interesting. Even when the narrator clearly used a different word - and I mean clearly - the AI chose to transcribe the popular phrase).

But these are really minor points. It looks like this has made an otherwise incomprehensible video comprehensible. That’s really exciting. What a time we live in!
Just need to sort through the videos I want to watch and make sure I upload them in advance. A bit of pre-planning never hurt anyone!

Right. I’m off to learn about ant colonies. :smiley:

1 Like

It is very much language dependent. For Italian, it works great. If you are a beginner, I wouldn’t recommend it, as you need to be able to recognise the mistakes. But if you are an intermediate or advanced, for some languages, it can work great.

Here’s the graph, which you can find on the Whisper GitHub.

Regarding uploading times, I would recommend you import the content before you wish to study them. It can take anywhere for minutes to hours to generate the transcript, based on server load, etc. I would recommend doing an uploading spree, then studying the lessons at a later date.


Thanks for this. I see Turkish is fairly high up the list… I suppose it has something to do with how much content is out there for the AI to pull from.

Having now done a lesson using the generated transcript, I’d say it was good. It was certainly a million times better than the YouTube generated subtitles. There were a couple of mistakes, but you’re right, I’m advanced enough (not advanced but advanced enough!) to be able to notice the errors and ignore them. I would have struggled to take in the video without the transcript so for me it just made the video accessible.

I like your idea about an uploading spree. I now have a list of channels to pull content from, so I will invest the time in going through them for lesson worthy content! Thanks for the recommendation.


Do you have to export the YouTube audio using a third party app or can LingQ do it directly?

@kraemder Export the YouTube audio using a third-party app/service. I think it was decided by lingQ that doing it any other way would be an infringement on YouTube guidelines (or something), but I am not sure.

1 Like

it is quite easy to do honestly, there are many tools online to do that. But of course, you always need to pay attention about what you do and where, to avoid bad websites.


I prefer Narakeet, it’s much faster albeit a paid service.

1 Like

I am using Whisper for the first time today. It’s great, only a couple of mistakes.

I am using it on an Italian vlog with a speaker from Milan using really clear standard Italian, so I imagine this helps!

1 Like