Volume drop with upsampling

I’m getting what I perceive to be a volume decrease with PCM upsampling. I haven’t measured it, but it’s reasonably obvious.

Is this expected? I don’t remember noticing it previously, but I only recently started using convolution, so either I can just hear more easily the difference upsampling makes, or it’s somehow related to convolution perhaps? I exported filters for all sample rates from REW.

I’m using custom upsampling, and what I upsample too seems to be irrelevant to the volume change. I have convolution, headroom, parametric EQ and speaker setup enabled, but are all unchanged when I try different sample rates.

I’m curious because if what I’m hearing is the upsampling, then it’s far less subtle than I previously imagined.

Anyone else have any ideas?

I raised a similar issue with early 1.3. I thought this was recently fixed?
Here is the thread:

I’m getting decreased gain. Not huge, but enough to notice. Could be related though?

Upsampling should not produce a gain change unless you are up-sampling to DSD and have SDM Gain Adjustment set to something other than 0dB.

There was an issue with convolution that occurred when filters were upsampled (unrelated to whether the audio stream was being resampled), but we believe that to be fixed.

Are you seeing this when you isolate to just flipping on/off sample rate conversion with no other DSP in the pipeline?

Measuring it by ear, or by recording output of a DAC and comparing measured loudness?

Haven’t disabled all the other DSP will try that tonight.

Measuring by ear but I planned to measure next time I have the mic out just to check it’s not imaginary. I guess I was more interested in whether it might be a technical thing, or whether there’s a perception of volume change just due to upsampling. The truth us I’ve never really been able to tell any difference in upsampled playback - through Roon or HQPlayer - and as this coincides with my setting up convolution wondered whether I’m now finally hearing that now I’ve cleaned up my sound.

Assuming it’s not imagined, it could potentially be differences in REWs filters for each sample rate? Will load them up later too.

Could be…or maybe our gain adjustment for convolution filters isn’t perfect. It’s not a totally simple problem to solve

Absolute volume != perceived volume–we are normalizing filters to absolute because perceived volume requires an editorial choice about the perceptual model to use. The best way is to let REW generate filters at every rate and avoid all filter resampling, but I think REW has limitations.

Lets get convolution out of it and see how it goes.

Pretty sure I only hear it when using convolution.

That said I did the A/B comparisons too many times and started doubting myself about it happening at all! Will try again tomorrow.

Just to follow up here, I’ve done quite a bit more exploring here - using couple if different sources. It’s definitely no imaginary. It’s a volume drop of perhaps a couple dB (by ear) or thereabouts.