Feedback: Cakewalk's built-in plugin upsample does not address latency

[LittleStudios]

I thought this made more sense in it's own thread.  I've been conducting a lot of tests regarding Cakewalk's built-in plugin upsample feature.  While running effects in parallel I discovered that Cakewalk isn't addressing the latency introduced by the plugin upsample feature.  I wondered if  Cakewalk's feature wasn't reporting latency back the same way plugins do, especially when plugins have their own oversampling feature.  I decided to test DDMF Metaplugin and compare phase relationships while running in parallel. 

I figured it would be best to show an example.  I've uploaded a video demonstrating this.  You may have to pause to read the captions explaining the steps taken.  Please watch to the end as I prove that Metaplugin's oversampling is in fact operating as expected.

Here's a link to the video: https://youtu.be/LgupFqtLDHc.

[LittleStudios]

I've done some more testing.  Basically the same test as linked in the video in my last post.  The difference this time is instead of sending the signal to an AUX track, I simply duplicated the track and added MCompressor in an oversampled state to the duplicated track.  I then bounced the original track to a new track and did the same for the duplicated track that has the oversampled MCompressor inserted on it.  The oversample settings were left to the default value, twice the project sample rate.  The project sample rate is 48 KHz.  The resulting latency, measured in samples, was 140 samples.  The track with the oversampled plugin was 140 samples late.  I also ran the test through DDMF Metaplugin set to 2X oversample.  When I bounced to a new track there was zero latency.  The latency was compensated for.

Just to be clear, this test was performed using Cakewalk's built in Upsample Feature. 

The point of my posts regarding the Upsample Feature isn't to complain.  Instead I hope this provides some useful information to the developers and to the users.  The cherry on top would be if this was addressed in an update.  If not, oh well.

[David Baay]

I could be wrong, and would have to test myself, but I suspect there's a logical problem in using a send to an aux track instead of a separate signal source. What I'm thinking is that any plugin delay compensation applied to the source track to keep it in sync with the upsampling delay will delay the send signal as well such that it's basically no possible to sync the two outputs unless the timing of the send is divorced from the delay of the track output signal. I haven't completely thought through how this squares with the plugin's internal upsampling being properly compensated, but i'd be interested to see if the problem reproduces when both tracks have an indepenndent-but-identical audio source

[LittleStudios]

36 minutes ago, David Baay said:

I haven't completely thought through how this squares with the plugin's internal upsampling being properly compensated, but i'd be interested to see if the problem reproduces when both tracks have an indepenndent-but-identical audio source

This is what I did.

4 hours ago, LittleStudios said:

The difference this time is instead of sending the signal to an AUX track, I simply duplicated the track and added MCompressor in an oversampled state to the duplicated track. 

I don't want to quote my prior post, but there I describe that I did use independent but identical tracks, one track had an upsampled plugin, basically doing nothing except giving the signal something to pass through.  Bounced the result of the oversampled track to a new track and counted samples (really fun btw).  The oversampled track was arriving at the speakers 140 samples late.

[David Baay]

Oops, missed that second post. I did a quick test and confirmed the 140-samples of uncompensated delay using independent audio clips  and Channel Tools with upsampling enabled for both playback and render. I also encountered some issues bouncing the master bus to a track with 64-bit DPE enabled. I reported it all to the Bakers for investigation.

Incidentally. setting your three nudge values to something like 1, 12, and 24 samples makes measuring sync errors a lot easier; just count the nudges, and add/subtract as you go until the zoomed waveforms are aligned or the audio is nulling.