Happy to see someone who can understand the maze that is FFmpeg. I wrote my own video editor too and getting any decoding/encoding to work was a nightmare
1:14:00 ish For the VLC not playing stuff back correctly. Dont drag the cursor back if you want it to restart. Simply play next so the playlist wraps around. One, The VLC slider is not absolute. Two, Pressing next forces VLC to start from the start of the file and buffer its entire contents again instead of using whatever buffer it had remaining
2:24:49 if you said AI instead of heuristicts I bet you'd reach a mil viewers by now haha jokes apart. amazing job real masterclass for us. seeing you gdb troubleshoot, line of thought, calc and paint draws thank you
I have an mpv plugin that does auto subtitles with whisper. The way they approach missing words is just padding 1s when passing the timestamp to ffmpeg. Whisper produces some duplicated words, but that works just fine.
Are you still subsampling the audio by only picking 1 every three samples? That works, but you'll also get audio aliasing, where frequencies higher than half the sample rate are folded back into the spectrum. (sum and difference between sample rate and original frequency, because you're multiplying compositions of sine waves. it'll be worst in sounds with lots of high frequencies, like the letter "s". To properly debu, you really ought to export the subsampled audio instead of the original. To better subsample, you ought to merge subsequent samples in some way. A simple slightly solution is to take the average of 3 subsequent samples. This is a form of low pass filter. The ideal, brick wall low pass filter has an impulse response (= the weights for how to add subsequent samples) takes the form sin(x)/x, 1 for x=0, scaled in such a way that sin(x) is 0 on neighbouring samples. This impulse response goes on forever, so you'll have to cut it off after some distance.
![Remble - Colors (ft. Mozzy & Stoneda5th) [Official Music Video]](http://i.ytimg.com/vi/KF-3AU9g9xg/mqdefault.jpg)
whats up, whats up, whats up
Happy to see someone who can understand the maze that is FFmpeg. I wrote my own video editor too and getting any decoding/encoding to work was a nightmare
1:14:00 ish
For the VLC not playing stuff back correctly. Dont drag the cursor back if you want it to restart. Simply play next so the playlist wraps around. One, The VLC slider is not absolute. Two, Pressing next forces VLC to start from the start of the file and buffer its entire contents again instead of using whatever buffer it had remaining
2:24:49 if you said AI instead of heuristicts I bet you'd reach a mil viewers by now haha
jokes apart. amazing job
real masterclass for us. seeing you gdb troubleshoot, line of thought, calc and paint draws thank you
I have an mpv plugin that does auto subtitles with whisper. The way they approach missing words is just padding 1s when passing the timestamp to ffmpeg. Whisper produces some duplicated words, but that works just fine.
Are you still subsampling the audio by only picking 1 every three samples? That works, but you'll also get audio aliasing, where frequencies higher than half the sample rate are folded back into the spectrum. (sum and difference between sample rate and original frequency, because you're multiplying compositions of sine waves. it'll be worst in sounds with lots of high frequencies, like the letter "s". To properly debu, you really ought to export the subsampled audio instead of the original.
To better subsample, you ought to merge subsequent samples in some way. A simple slightly solution is to take the average of 3 subsequent samples. This is a form of low pass filter. The ideal, brick wall low pass filter has an impulse response (= the weights for how to add subsequent samples) takes the form sin(x)/x, 1 for x=0, scaled in such a way that sin(x) is 0 on neighbouring samples. This impulse response goes on forever, so you'll have to cut it off after some distance.