Isn't it more correct when the txt timestamps more closely align with the actual time the syllable is sung in the audio? It's a common point of reference that all software and hardware try to match, even if they don't always succeed. It's better to sync all songs to the same correct timing than to have them all synced to different delays of different software and settings, don't you agree?
I made an experiment by generating a separate click track from the note timings in the txt with ffmpeg and playing it together with a fresh downloaded audio in Audacity. This way I could really see the audio waveform and timestamps and be sure that the clicks play at the correct time.
Results: The 6640 timestamp is indeed correct (or maybe 6630) and 6710 is wrong. So beware that YASS (at least in your setup) seems to have a 70ms delay in audio. Interestingly it also tells us that the MIDI playback in Ultrastar Play (at least on Ubuntu 22.04 with Ogg Vorbis audio) is really accurate and could be used to correctly time a song.
If you're interested in seeing a video of the Audacity playback send me a PM.
Tuupertunut: at least in all ultrastar versions i tested, midi is not properly synced. also the clicks in ultrastar are not 100% on point. use either play + visual inspection whether the notes light up as expected or space to play a single note and hear whether it starts and ends correctly.
I'm using Ultrastar Play editor on Ubuntu 22.04 to listen if the midi piano sounds align with the song. I'm not aware of any settings that would affect the midi-piano-to-song delay, other than that different editors may have different internal delays.
Tuupertunut: at least in all ultrastar versions i tested, midi is not properly synced. also the clicks in ultrastar are not 100% on point. use either play + visual inspection whether the notes light up as expected or space to play a single note and hear whether it starts and ends correctly.