My pipewire seems to have issues with crackling audio and severely dampening my mic and I have no clue why.
Pipewire’s default quantum (buffer size, effectively) is incredibly low, this is good for low latency audio but anytime your system is too busy to keep the buffers filled you get crackling.
If you look at pw-top you’ll see all of your devices and nodes. The quant column is probably 1 or a very small number for the devices.
You can increase the quantum with this command. This only lasts until pipewire restarts:
pw-metadata -n settings 0 clock.min-quantum 512
At a sample rate of 48000, this is roughly a 10ms buffer. 1024 is 20ms, etc. You want it as low as possible without getting crackling. Start with 512 and adjust from there (you don’t have to use a power of 2, a quantum of 1234 works just as well).
severely dampening my mic and I have no clue why.
By default pipewire doesn’t do any ‘mic boost’, as Windows calls it. You can get the same effect by raising the maximum volume.
In your sound control panel you should be able to turn the mic up higher than 100%. In KDE Plasma, you can do this in System Settings -> Sound -> Configure Volume Controls… [top right button] -> Raise maximum volume.
Alternatively, you can use EasyEffects to add a compressor. This will boost your mic volume and also prevent it from getting too loud
Compressors basically reduce the dynamic range of an audio signal by attenuating loud sounds and boosting quieter ones, this would provide a better mix.
Other useful plug-ins are noise canceling, (kills background noise) and echo canceling (lets you play sound out of your speakers which won’t get picked up by your mic). Sometimes apps, like Discord, will do this signal processing for you while others, like Signal, do no signal processing.
I absolutely wasn’t expecting a helpful response under a meme, so thank you very much for taking the time to write it!
anytime your system is too busy to keep the buffers filled you get crackling
I’d have to test that more thoroughly, but I do think that lines up with the timing of the issues.
You can increase the quantum with this command. This only lasts until pipewire restarts:
Can I put that in some config to make it stick?
I’m admittedly very junior to pipewire config, so most of what I have is copied from the internet, tweaked for node names / descriptions, but I generally like working with config files and slowly learning what all that stuff in there means.
I have two loopbacks (I like having music and games each grouped separately from other audio), an echo-cancel and a noise cancel (filter-chain with a single rnnoise node), all configured via .conf files. As an aside, is there a “best order” to chain echo cancel and noise cancel?
Echo cancel seems to have a quantum/rate of 480/48000 across the board. Loopbacks, rnnoise and alsa_output (my headset) all have 0/0. I imagine it makes sense for the Loopbacks and rnnoise, but should it be something else for the main output?
By default pipewire doesn’t do any ‘mic boost’, as Windows calls it. You can get the same effect by raising the maximum volume.
Well, it seemed to work just fine without echo cancel, if I capture the mic directly, but putting it through echo cancel (with or without noise cancel) seems to reduce the gain significantly.
I’m gonna mess with the volume sliders and see which ones I can crank up to fix that issue.
But I’m confused why that issue would occur in the first place and if I have something misconfigured.
Alternatively, you can use EasyEffects to add a compressor. This will boost your mic volume and also prevent it from getting too loud
Sounds like a compressor would be a good idea to have anyway. Is that also doable through the config? I’m not opposed to graphical tools, I just feel like working with the config directly is more educational. It’s also more prone to screwing things up, but that’s just bonus lessons on what not to do.
Sometimes apps, like Discord, will do this signal processing for you
Curiously, the reason I looked at echo-cancel in the first place is that Discord’s own echo fucks with things, cutting me out at times while also not cancelling the echo at others.
I have two loopbacks (I like having music and games each grouped separately from other audio), an echo-cancel and a noise cancel (filter-chain with a single rnnoise node), all configured via .conf files. As an aside, is there a “best order” to chain echo cancel and noise cancel?
Echo cancel seems to have a quantum/rate of 480/48000 across the board. Loopbacks, rnnoise and alsa_output (my headset) all have 0/0. I imagine it makes sense for the Loopbacks and rnnoise, but should it be something else for the main output?
I have been doing echo-cancel -> rnnoise. That way echo cancel gets a clean stream to do what ever correlations that it needs to do and then rnnoise de-noises what is remaining.
As far as the latency, I think it is because echo cancel needs a bit of a default wait in order to actually hear the sound coming out of your speakers (speed of sound is slow, smh) which is why I think that delay is there (though this is complete speculation, if someone knows better I’d love to know).
The quantum of all of the devices is propagated through the chain so if you have echo cancellation in a graph then all everything will use at least its quantum (if there are not higher quantum objects in the chain). If a device doesn’t have a quantum, it’ll either use the min-quantum or the highest quantum of any node in the graph.
Sounds like a compressor would be a good idea to have anyway. Is that also doable through the config? I’m not opposed to graphical tools, I just feel like working with the config directly is more educational. It’s also more prone to screwing things up, but that’s just bonus lessons on what not to do.
I agree, learning the underlying system pays dividends.
Any LADSPA filter will work as a node in pipewire. So the world is your oyster!
Curiously, the reason I looked at echo-cancel in the first place is that Discord’s own echo fucks with things, cutting me out at times while also not cancelling the echo at others.
Yeah, I’ve had similar experiences. I typically disable any sound processing done by the application and depend on my own plugins to handle that.
I adapted the normalizer Sink from that post to be a normalizer Source, replaced the local paths with the system lib names, made a mistake, fixed it, got everything hookes up as I think it should be. How it performs under load remains to be seen, and which order noise cancel and nornalizer go will be tinkered with, but at least the service restarts and everything looks fine.
May life be as kind to you as you have been to me.
Pipewire’s default quantum (buffer size, effectively) is incredibly low, this is good for low latency audio but anytime your system is too busy to keep the buffers filled you get crackling.
If you look at
pw-topyou’ll see all of your devices and nodes. The quant column is probably 1 or a very small number for the devices.You can increase the quantum with this command. This only lasts until pipewire restarts:
At a sample rate of 48000, this is roughly a 10ms buffer. 1024 is 20ms, etc. You want it as low as possible without getting crackling. Start with 512 and adjust from there (you don’t have to use a power of 2, a quantum of 1234 works just as well).
By default pipewire doesn’t do any ‘mic boost’, as Windows calls it. You can get the same effect by raising the maximum volume.
In your sound control panel you should be able to turn the mic up higher than 100%. In KDE Plasma, you can do this in System Settings -> Sound -> Configure Volume Controls… [top right button] -> Raise maximum volume.
Alternatively, you can use EasyEffects to add a compressor. This will boost your mic volume and also prevent it from getting too loud
Compressors basically reduce the dynamic range of an audio signal by attenuating loud sounds and boosting quieter ones, this would provide a better mix.
Other useful plug-ins are noise canceling, (kills background noise) and echo canceling (lets you play sound out of your speakers which won’t get picked up by your mic). Sometimes apps, like Discord, will do this signal processing for you while others, like Signal, do no signal processing.
I absolutely wasn’t expecting a helpful response under a meme, so thank you very much for taking the time to write it!
I’d have to test that more thoroughly, but I do think that lines up with the timing of the issues.
Can I put that in some config to make it stick?
I’m admittedly very junior to pipewire config, so most of what I have is copied from the internet, tweaked for node names / descriptions, but I generally like working with config files and slowly learning what all that stuff in there means.
I have two loopbacks (I like having music and games each grouped separately from other audio), an echo-cancel and a noise cancel (filter-chain with a single rnnoise node), all configured via
.conffiles. As an aside, is there a “best order” to chain echo cancel and noise cancel?Echo cancel seems to have a quantum/rate of 480/48000 across the board. Loopbacks, rnnoise and alsa_output (my headset) all have 0/0. I imagine it makes sense for the Loopbacks and rnnoise, but should it be something else for the main output?
Well, it seemed to work just fine without echo cancel, if I capture the mic directly, but putting it through echo cancel (with or without noise cancel) seems to reduce the gain significantly.
I’m gonna mess with the volume sliders and see which ones I can crank up to fix that issue.
But I’m confused why that issue would occur in the first place and if I have something misconfigured.
Sounds like a compressor would be a good idea to have anyway. Is that also doable through the config? I’m not opposed to graphical tools, I just feel like working with the config directly is more educational. It’s also more prone to screwing things up, but that’s just bonus lessons on what not to do.
Curiously, the reason I looked at echo-cancel in the first place is that Discord’s own echo fucks with things, cutting me out at times while also not cancelling the echo at others.
Make a file in pipewire.conf.d: ~/.config/pipewire/pipewire.conf.d/min-quantum.conf
With this, replace the quantum with one you’ve found works best:
context.properties = { default.clock.min-quantum = 512 }Restart pipewire for it to take effect.
I have been doing echo-cancel -> rnnoise. That way echo cancel gets a clean stream to do what ever correlations that it needs to do and then rnnoise de-noises what is remaining.
As far as the latency, I think it is because echo cancel needs a bit of a default wait in order to actually hear the sound coming out of your speakers (speed of sound is slow, smh) which is why I think that delay is there (though this is complete speculation, if someone knows better I’d love to know).
The quantum of all of the devices is propagated through the chain so if you have echo cancellation in a graph then all everything will use at least its quantum (if there are not higher quantum objects in the chain). If a device doesn’t have a quantum, it’ll either use the min-quantum or the highest quantum of any node in the graph.
I agree, learning the underlying system pays dividends.
Any LADSPA filter will work as a node in pipewire. So the world is your oyster!
This explains how to set it up better than I can here: https://forum.endeavouros.com/t/pipewire-filter-chains-normalize-audio-noise-suppression/31661
Yeah, I’ve had similar experiences. I typically disable any sound processing done by the application and depend on my own plugins to handle that.
I adapted the normalizer Sink from that post to be a normalizer Source, replaced the local paths with the system lib names, made a mistake, fixed it, got everything hookes up as I think it should be. How it performs under load remains to be seen, and which order noise cancel and nornalizer go will be tinkered with, but at least the service restarts and everything looks fine.
May life be as kind to you as you have been to me.
I’m glad you got it figured out!
Just pay it forward, there’s always people with questions that need answered :P