[pipewire] Perceptible difference in sound quality compared to pulseaudio

[shindere]

I recently switched from the pulse to the pipewire output method.

With the voice I use (Vocaalizer's Thomas Pro french voice, Voxin 4.3), the
sound is of less good quality when output via pipewire as what I had with
pulseaudio.

I don't know exactly how to describe the change, it's as if the voice is
less clear and distinct with pipewire than with pulseaudio.

Would it help to provide audio clips from the two outputs?

[alex19EP]

have you read for example https://wiki.archlinux.org/title/PipeWire#Sound_quality_(resampling_quality)

[shindere]

Thanks for the tip! It indeed looks very relevant, so I tried the change. I couldn't observe any effect but at the same time I am not sure the change I have done was actually taken into account.

[shindere]

So I created /etc/pipewire/client.conf.d/resample.conf with the following
content:

stream.properties = {
  resample.quality = 10
}

but I do not observe any change even after having restarted the
computer. And I had not more luck with the smae content put in
~/.config/pipewire/c:ient.conf.d/resample.conf.

[sthibaul]

Well, I don't know. I guess this should rather be discussed with pipewire people. Apparently the spa-resample tool can be used to try to run the resampling by hand, to check the difference.

[shindere]

Thanks. I will continue experimenting. Is there a way to compare outputs which would be more objective thanmy pour human ears? Because I start to realise that I do not trust myself anymore to know whether there actually are differences or not. Another thing which is unclear to me is that I do not know what Voxin delivers as output. If nobody else knows I'll try to figure out but if somebody already knows I'd appreciate it quite a lot to hear about it.

[TheMuso]

I think Voxin's output sample rate is 11025 Hz. Espeak is 22050 Hz.

[sthibaul]

for people who would like to check the resampling quality, one way to make sure that the resample.quality parameter is taking place is to run pw-dump , look for Accessibility (the name under which the speechd pipewire pluging registers), and check for the resample.quality parameter there.

We investigated a bit with @shindere , it seems the quality difference could be in an upgrade of voxin itself, more investigation is ongoing