I Made an App that Accurately Produces Subtitles Using Whisper 

Is it accurate? even if you select the audio quality and the model size to be high, there may be a lot of errors depending on the voice heard, comparing to a person who talks clearly and a person who usually slurs stuff (especially anime chars). But this is really cool, amazing job! I see it really useful for productive language learning/consumsion.

Hi, Jarod! I've just tried this with a local file which is probably going to be my standard use case. I'm using the Whisper API on the server, not locally on my PC. The mp4 I used was 1.6G, it was a 45 minute video. At first, I left the audio quality at 128; this generated an mp3 file just over 43Mg, and I eventually got a Whisper API error that the audio file exceeded the size limitation for transcription. I then selected 16 audio quality. This time the mp3 file generated was almost 11Mg. An srt file was created! I looked at the subs, and they looked really good (Japanese) for the first 26 minutes. The srt file kept going for the full 45 minute program, but after 26 minutes the subs were just 「どん」repeatedly until the end of the file! 🤣 This is actually a big improvement over the beta version of the transcriber I tried out, so this is a really promising tool, and I'm quite excited. Do you have any thoughts on why the srt file crapped out half-way? Thank you very much for providing this language-learning game changer!

Great video

Some other video I saw didn't utilize an openAI key to get whisper running. Is there some kind of difference or rather benefit with utilizing the openAI Key?