Tell me the code of the movie, I'll try to find it and see if there is any problem that makes whisper not detect the audio correctly, it will be tomorrow when I upload the result, I almost used the file limit.
I'm in the same situation as the soloporhoy666 user, but with the bban-087 subtitle, it shows these strange characters. Could you do the same with this one?
I also try the whisper using Pazera mp4 to mp3 converter using format lowpass filter 16000 hz and volume +3 the result was very promising, but the problem only limitation for using GPU freely, is there any option to make it alwasy free?
In my case, the transcription usually takes 30 minutes per film, the platform gives me the opportunity to use it up to 5 times, it seems like a long time, but on the weekend I need a little more, so I use two different accounts and with that I do it for free, the payment option does not improve at all according to the developers themselves, it is only being able to use the virtual GPUs without limits.
According to what I read, it has no solution, since it was generated in a specific environment and with characteristics typical of a certain country, so only another user from that country could solve the file. the user just provided me with another file that he already had.
I label my subtitle projects as "Fan Fiction" because I intentionally change the story to make it more interesting and to fit my own personal kinks.
I think it may be the graphics card, the requirements are very demanding, something like a 2060 onwards.
SE by default uses Whisper.cpp build which is CPU only ---no GPU dependency. I wonder if in your environment it is picking up a wrong build. That can happen I guees if you have installed (python version of) Whipser prior to SE.
I couldn't get it to work, I had to try a beta and portable version, but it was taking a long time and I couldn't get any results, so I'm using the virtual Collab instead and it's going great, I'm finally getting results, which makes the story go better.
Yes, a Collab, through drive, I don't know exactly what it is, I just know that some developers create codes to use Whisper easily without using their own resources, in my case I don't know anything about programming or codes it was easy to understand and use whisper+VAD and create very good subtitles and lots of lines of dialogue, so far the record for good lines is 1700 lines with good quality. It's not perfect but much better than the other methods I've used.
