In recent weeks, there have been a lot of errors in the Ukrainian audio transcripts! Here is an unedited screenshot on the left, and on the right, written in bold, are all the words I had to correct. For a text lasting about 25 minutes, it takes me at least 30 extra minutes to correct the transcription errors! Does the LingQ software have hearing problems?
@zoran
(A few months ago, there were Russian transcriptions, but they finally disappeared after quite a long time.)
1 Like
The transcription is done using ai, and the quality of the result highly depends on the amount of data the language model was trained with. This leads to big differences depending on the language that is supposed to be transcribed.
Below is a picture of the word error rates of whisper ai transcription based on two standarized test.
As you can see, even though Ukrainian isn’t the worst performing language, there are quiet some errors to be reckon with. You can try other transcription methods, I personally use Faster Whisper XXL, which you can download to your pc. It allows you to choose the language model used, too, so maybe choosing a larger model helps. In the standarized tests the differences don’t seem to be too big, and larger model take way longer for transcription, but depending on the content transcribed the results may differ.
I personally try to use material that comes with man-created subtitles and if I use automatic/ai-created ones, I keep ignoring the wrongly spelled words. The LingQ surface isn’t intented for efficient later corrections, despite this beeing criticised for god knows how long, and investing that much time in making corrections seems like a waste of time, imho.
1 Like
Thanks for these informations : I already have whisper on my Mac. I don’t understand how to use it so that the high level audios of Ukrainian teachers would be correctly transcribed, nor how the videos could be correctly transcribed? I am not computer savvy enough to master this.
Ask ChatGPT, it can guide you through the process. There should be an explanation on the page you downloaded whisper from, too.
Faster Whisper XXL is executed from the console. I can’t speak for Mac, but on Windows it is possible to have this process be automatized via an executable, so that it for example automatically transcribes all the audio files in a specific folder. ChatGPT can help you creating such an executable, too.
Thank you for your explanations, I will try, despite my natural incompetence on these technical and IT subjects which are not my areas of expertise.