Hey Peanut,
im experiencing the same issue in delay, recording a video for a png animation lip sync seems to sync up better than in stream.
a Temporary fix would be adding less mouth shapes to your png for the software to speed up and search for less mouths, however you lose most of the complex sounds and the mouth appears to mostly open and just shut.
I have quite alot of expression and exagerated mouth shapes when i stream and the delay is quite jarring due to the way that the sound is being matched with the correct mouth. when there is no lag the amount of mouths flow quite organically with the speech but the live calibration of the software searches for the sound and matches the mouth shape which breaks immersion most times. this is also dependent on internet conecvtivity as well it seems
If there was a way to set up a calibration process where the user could record a sentence in HONK! and the mouth shapes would be calibrated and saved before going live which would reduce lag time before going live in a streaming program or recording that would be fantastic. it would be awesome to see an editing panel where you could input the recording of the sound and see the mouth animation in an editing mode interface and make slight adjustments within editing mode. For instance sometimes my png will pick up sounds of my keyboard and mouse clicks if i could pre-record my keyboard or mouse clicks and set an attribute of 'silent' into the mouth layers that would reduce unwanted lipsync
the E sound viseme is calibrated to more of an EH sound depending on your accent , and native tongue you may need to re-use some mouths to adjust to your accent
here is an example of my png rig on twitch with the same issue
https://clips.twitch.tv/BashfulHardIguanaMcaT-4-SlWuVsmiVLH_u5?tt_content=url&tt...
here is also an example of where the mouth sprites flow quite nicely
https://clips.twitch.tv/SavoryBombasticFloofFunRun-o03d9H3zxHkZCAYf?tt_content=u...
Hope this helps improve Honk! as i am quite fond of it