I’ve had great difficulty trying to import SRT/ASS files into a lesson and having it sync with an audio mp3 file.
For context, this is for Chinese. The mp3 file is ~19 minutes long, the text has ~250 sentences of ~8 characters each. The audio is from a children’s tv show, so there is some music, sound effects, a lot of conversations, etc.
After I create the lesson, add the SRT/ASS file, and generate lesson. The initial load of the lesson usually works. The sentences are properly synced with the audio. The words are recognized by Lingq.
However, shortly after, the lesson updates itself, breaking it. Two different errors either happen
The audio timestamps are updated such that the timestamps are continuous. With this change, the existing timestamps are overwritten, incorrectly, and the audio is no longer synced with the text.
(pictures below)
90% of the words become unrecognized, greyed out, unable to select or interact with them. However, the audio remains synced.
(pictures below)
I’ve spent around 2 hours and have found no work around to this. Does anyone know a remedy for this?
This is core functionality so maybe this is some type of regression on the development side.
I have also tried just importing the audio. However, the Whisper model does not perform that well for my audio. Additionally, most of the time there are huge sections of the audio without punctuation. This results in minutes long single sentences which make lessons impractical for me.
I am experiencing a similar problem. It happened to 3 of the lessons that I created today. In my case, when I add a TITLE paragraph (these are not supposed to have a start and end time), the system erroneously decides that something is wrong and decides to regenerate the timestamps of nearby paragraphs, breaking paragraph synchronisation with the audio.
I have experienced similar problems in the past when adding new paragraphs, and, while I am still in the middle of editing, the system decides that it doesn’t like some timestamp and so goes and regenerates timestamps in the lesson with new bad ones.
But: This has never happened before when adding TITLE paragraphs. That’s new. It’s gotten a lot more aggressive.
Similarly, I have had all my corrected paragraph translations frustratingly automatically replaced with auto-generated ones with translation errors.
LingQ: Stop automatically replacing timestamps and translations, thinking that you are somehow improving lessons with errors. While perhaps for some sloppy people who put no effort into creating their lessons, this may help, for those of us who put effort into creating our lessons, you are just breaking our lessons and causing endless amounts of wasted time and frustration.
Instead: State that you may have detected something wrong with the timestamps/translations on paragraphs X, Y, and Z and ask if we would like to have those paragraphs corrected. Tell us the paragraph numbers so that we can correct them ourselves if we choose to do so (which I always will).
I just spent over an hour fighting a variation of this bug. This seems to be what @seanmchale is experiencing.
Over and over again, I created my lesson, starting by importing an .srt file for a YouTube video, finding that at some point during the editing all of the timestamps have been regenerated, very badly, and so I had to delete the lesson and start all over again.
In this case, it seems that importing an audio file somehow forces the timestamps to be regenerated. It doesn’t matter if I specify it when the lesson is first created, or if I add it in later. Once an audio file is attached to the lesson, all the timestamps get clobbered soon afterward.
For the lesson that I wound up sharing, I had to abandon the idea of also including an audio file. And so far the timestamps haven’t been clobbered.
If you import from YouTube via the browser extension, adding an audio file does not trigger this problem. But if you import a lesson starting with a subtitle file, then it can be affected by this, but perhaps not for all lessons. I imported a different YouTube lesson earlier today starting with an .srt file, and was able to add an audio file without timestamps getting messed up.
I came across another variation of the problem today and spent a lot of time fighting it and trying to work around it.
I’m wondering if it got introduced today with some code changes, because I also noticed another problem that I haven’t seen earlier.
New variation of the problem: This time I imported a lesson from YouTube with the LingQ browser extension. Twice all the timestamps got replaced with bad ones and I had to delete the lesson and start all over again. I did add an audio file, but unlike with the previous variation of this problem, this was not a subtitle file import, but a YouTube import via the LingQ browser extension, which until now has not resulted in this problem. For the 3rd attempt, however, despite wanting to add an audio file, I did not, and so far the timestamps have not been replaced.
This particular lesson has 14.6 seconds at the start before talking begins. And that may be related to triggering this problem, and it also exposed the other new problem that I noticed today:
In my 3rd attempt, when I did not add an audio file, LingQ would not play back the part of the lesson before talking starts, nor the part of the lesson after talking ends! i.e. It was not capable of playback before the first timestamp nor after the last timestamp! This has never happened before. And I highly suspect a code change that was just pushed through. This was when using LingQ via a web browser. Please have this fixed also. My workaround for this was to add a new “PARAGRAPH 1” before the real 1st paragraph, and set its start time to 00:00.0 and end time to 00:14.6 (I set the text to the lesson’s title - But don’t use a “TITLE” paragraph for this, because that has a bug too that triggers an undesired replacement of timestamps!)
I hope that all of the bugs that we have mentioned in this thread get fixed soon.
But again: Just stop replacing timestamps! It just causes problems and frustrations and wastes the time of lesson creators! Or if you are going to insist on it, then provide us a means to TURN THIS OFF!
The same goes for replacement of translations - that happened to me today also. I had to correct the same translations twice because of it.
I’ve also modified this SRT file so the timestamps are continuous (no gaps between phrases) as well as starting at second 0. However, this also still leads to the above error I described.
@seanmchale So, just to confirm here, after you import the text and manually upload an audio file too, the lessons look good and are synced properly. However, after some time, the timestamps breaks?
Hello @seanmchale The issue No2 with the grey text are in progress of fixing now. And I wanted to clarify with you the issue No1 with the timestamp regeneration. I have generated lesson with the SRT and MP3 files and timestamps are correct. What are the next steps to reproduce the issue? I tried adding paragraph, changing the timestamps - and the remaining timestamps were not changed. Thank you for your help in advance
Thanks for looking into this. I just tried to reproduce the SRT timestamp changing bug, however, I was no longer to replicate it.
Originally, I would go into the lesson, play some sentences, leave the lesson, go into the lesson again and see the timestamps are incorrect. I clicked edit lesson and saw the mismatch in timestamps.
Hello @seanmchale all two issues - with grey text and with unexpectedly changed timestamps should be good now. Could you please confirm from your side? Thank you
