Duplicated audio does not sound the same #83997
Operating system: Windows-10-10.0.19041-SP0 64 Bits
Graphics card: GeForce GTX 1070 Ti/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 456.38
- 2.92.0 Alpha, branch: master, commit date: 2020-12-19 06:25, hash:
Worked: (newest version of Blender that worked as expected)
Short description of error
Duplicating the same audio strip several times in a timeline, gives a different sound.
Exact steps for others to reproduce the error
- Download the attached blend-file with a sample audio strip.
- Duplicate the audio strip a few times.
- Render the output and import the mp4-file into Audacity or Blender to see the wave form. They look different.
- Check the sound. The exported audio file from Blender sound different.
Rendering the audio only (menu Render > Render audio) and decreasing the accuracy will improve the audio.
Changed status from 'Needs Triage' to: 'Needs User Info'
I've tested this only Linux and can't hear any different between the duplicated strips. That the waveform of the export to mp4 may not look exactly like the one from the original .wav file can be due to lossy encoding. Could you please provide a file with the audio duplicated where the strips sound differently? How often did you have to duplicate the audio strip until the issue occurred?
I included the original sound-file (bounce.wav), a blend-file with the sound-strip + 2x duplicated. You can hear the difference between the swooshes (but not very clearly). I included also the wave-form of the render (of the three (identical?) soundstrips. You can clearly see that the waveform of all three has small differences. After rendering the audio only, in which case you can set the accuracy (low value = high accuracy), the differences disappear. I included also the Audacity-profile of this audio-only output..
Changed status from 'Needs User Info' to: 'Needs Triage'
Changed status from 'Needs Triage' to: 'Confirmed'
After the export the audio is significantly different, before the export they all sound the same to me, but that might just be my ears not picking up subtle differences.
Yes, you're right. Before the export they sound indeed the same. Perhaps I wanted to hear the difference because I knew about the profiles. So, I did the test and recorded the sound (with volume set to 5) with a microphone. The waveform looks identical in Audacity. So, it's a problem of the render (to video). The render audio can do it right, if accuracy is set low.
It could be related to the fixed value of AUD_DEFAULT_BUFFER_SIZE. https://developer.blender.org/diffusion/B/browse/master/extern/audaspace/config/Audaspace.h.in$103
It seems to be the reading-buffer-size which is set too high and cause inaccuracy. Maybe this value should be lowered or exposed so users can set this value manually?
@neXyon Maybe you could have a look at this?
This issue was referenced by
That one was a bit harder to find. Running down the wrong rabbit hole, I also found another inaccuracy! :D
@tintwotin While the behavior depends on the size of the mixdown buffer, this is not the precise cause for the bug. During audio rendering, users can actually choose the value manually.
No due date set.
No dependencies set.
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?