Bouncing Send FX + mixer channels sounds different from baked-in Send exported channels

[cocoazenith]

So I want to bounce all mixer channels including Send FX channels in order to make a new file solely for Mixing purposes.
I am using the "Apply all mixer settings" option. Now here comes the problem. The new project sounds different from the original project, especially for the channels which use the Sends. I must mention that the original project and the new mixing project have the master section at default values, without anything activated on it.

Thinking that I must be hearing things I went on to do a comparison using only one track + one send fx. Here are the steps I followed:

1. Created a new project and made a new track with a simple Piano patch with no effects.
2. Made a new send effect.
3. Activated send fx in the main mixer on the Piano track.
4. exported song as audio file (file A)
5. bounced mixer channels: Piano track and Send fx track.
6. dragged these two new wave files in a new project
7. exported song as audio file (file B)

Now here comes the strange part, shouldn't file A sound the same as file B? They sound different. I even compared the waveforms and they are different. File B is louder. (see image, the lighter shade of behind the dark waveform is the new waveform)

I always thought that exporting separate stems for the Sends does not affect the sound of the new MIX project made out of stems.
Is the above behaviour normal? How can I correct it so that it sound the same as the Sends baked in?

Many thanks in advance to the knowledgeable!

[selig]

Yes, they should sound the same. So I followed your instructions and they do in fact sound exactly the same.
Not sure what is different, check and make 100% sure you’re not starting with a ‘template’ that includes anything at all.

[cocoazenith]

Yes, they should sound the same. So I followed your instructions and they do in fact sound exactly the same.
Not sure what is different, check and make 100% sure you’re not starting with a ‘template’ that includes anything at all.

Thanks.

Indeed I tried it again with only ONE piano note played and waveforms sound and look the same. I'll have to double check what happens in that song.

[cocoazenith]

Yes, they should sound the same. So I followed your instructions and they do in fact sound exactly the same.
Not sure what is different, check and make 100% sure you’re not starting with a ‘template’ that includes anything at all.

Thanks.

Indeed I tried it again with only ONE piano note played and waveforms sound and look the same. I'll have to double check what happens in that song.

Tried it again for multiple notes and it sounds the same. But I compared the waveforms in a graphic design software and they differ a bit in amplitude when zoomed to microscopic levels. Maybe it has to do with the piano I used, maybe not, bit it seems that when multiple notes are played I never get that perfect match that I get from that single note.

[selig]

Thanks.

Indeed I tried it again with only ONE piano note played and waveforms sound and look the same. I'll have to double check what happens in that song.

Tried it again for multiple notes and it sounds the same. But I compared the waveforms in a graphic design software and they differ a bit in amplitude when zoomed to microscopic levels. Maybe it has to do with the piano I used, maybe not, bit it seems that when multiple notes are played I never get that perfect match that I get from that single note.

The piano: probably multi-sampled where a different sample is triggered each time you play the same key. So in that there are probably at least three or more samples for every single articulation, triggered round robin, and there’s a chance it won’t play the same exact samples each time.
The reverb: if it’s algorithmic as opposed to convolution based, it will 100% be different each time.

If you remove those variables, most easily done by printing the piano to audio before you do anything else, and using a convolution reverb, you should get 100% null-able results each and every time.

[cocoazenith]

Tried it again for multiple notes and it sounds the same. But I compared the waveforms in a graphic design software and they differ a bit in amplitude when zoomed to microscopic levels. Maybe it has to do with the piano I used, maybe not, bit it seems that when multiple notes are played I never get that perfect match that I get from that single note.

The piano: probably multi-sampled where a different sample is triggered each time you play the same key. So in that there are probably at least three or more samples for every single articulation, triggered round robin, and there’s a chance it won’t play the same exact samples each time.
The reverb: if it’s algorithmic as opposed to convolution based, it will 100% be different each time.

If you remove those variables, most easily done by printing the piano to audio before you do anything else, and using a convolution reverb, you should get 100% null-able results each and every time.

Scratching my head with all of this. I used a basic Sawtooth waveform (using Vital VST3, no FX, no nothing) + Reason's RV7000 MK2 Convolution Reverb as Send FX.

The waveform from the baked-in export differs from the waveform of the file with two bounced channels (instrument + send). (img1)
And this is only the first notes of the "song".

P.S. Moreover the waveform of the file with the two bounched channels seems to be starting later. But that's not essential to my problem. (img2)

[dioxide]

Here is a demonstration about how a wave's visual appearance can be different depending on the phase. Unless you are importing a WAV file to an Audio Track and re-exporting you are unlikely to get a waveform that looks identical to the original and even then there may be differences.

Also:

[selig]

You’re still introducing a variable. Bounce the synth to audio so it is BIT FOR BIT the same each time it plays. Then you should see what I saw, which is what you expected to get in the first place.

Remember, I got it to pass a null test with my setup, which indicates there is no difference between the two waveforms being compared.
This test relies on the actual audio signal rather than comparing graphic representations of waveforms to determine for sure if there’s a difference or not.