For lessons with audio where the generated timestamps are way out, is there some way just to have TTS in sentence mode? Correcting all the timestamps is just too time consuming.
Or/also is there a way to delete / not have timestamps because even in page view when listening to the audio it will highlight the current sentence incorrectly.
You can disable the underlining in the TTS settings.
If it’s something you imported, you can try editing the lesson and clicking “generate timestamps”. It’ll redo it. In the past it often helped a lot. Not sure if it does nowadays as they are generally pretty good in my experience in the last year or so.
If that doesn’t work, and it’s your imported lesson, you can just delete the audio and put a link to it for reference. Then it’ll use the tts…and if you want to listen to the real thing you can just go to the source url. (update listening statistics manually if you care).
If it’s not your lesson, you’re out of luck I think. You’ll need to try to contact the person and have them regenerate the timestamps. Or maybe Lingq team can do it.
Thanks, that solves one problem nicely!
Yeah, I’ve tried regenerating but it’s no better. It works well on some lessons, but this is audio from a TV series so I think it gets thrown off by some of the background noise and music.
Deleting the audio and having a link is kind of swaps one problem for another, I still want the playlist in the app to work when listening etc.
You could try an external TTS system. Whisper is available for free and comes with a ton of settings. On setting is to specify the language model used (so the size of the data the model is based on). A bigger model means that it recognizes words better for the price of more time needed.
Faster-Whisper-XXL, which is what I have been using from time to time, allows batching. So you can tell it to transcript a folder with several files in it, and during that do something else.