Headless streamer with convolution recommendation?

[dennnic]

Henrik tested NEON performance in CamillaDSP on RPi4. Since only float32 are notably accellerated, camilladsp using float64 does not gain from A72 NEON https://www.diyaudio.com/forums/pc-...rossovers-correction-etc-236.html#post6798125

And how about Fusion plugin instead of Camilla? I don't intend to use any crossover slopes or fir filters, merely an equaliser.

 

[somebodyelse]

And how about Fusion plugin instead of Camilla? I don't intend to use any crossover slopes or fir filters, merely an equaliser.

Volumio3's Fusion plugin uses Camilla to do the processing.
https://community.volumio.org/t/fusiondsp-for-volumio3-beta/50474/2

 

[voodooless]

And how about Fusion plugin instead of Camilla? I don't intend to use any crossover slopes or fir filters, merely an equaliser.

Your original question was about convolution, which is an FIR filter. The normal equalizer stuff uses IIR filters. You can equalize using FIR with use of external tools.

 

[dennnic]

Your original question was about convolution, which is an FIR filter. The normal equalizer stuff uses IIR filters. You can equalize using FIR with use of external tools.

Right, I didn't know that.
When we're at that subject, I'd like to ask - what sample rate in rew should be chosen for measuring and eq filter generation?

I've read that there's an automatic upsample in DSP if lower setting is chosen. Does 48 kHz sample mean that EQ will have fewer taps and be (theoretically) less accurate, compared to sample of 192 kHz?

 

[dennnic]

Henrik tested NEON performance in CamillaDSP on RPi4. Since only float32 are notably accellerated, camilladsp using float64 does not gain from A72 NEON https://www.diyaudio.com/forums/pc-...rossovers-correction-etc-236.html#post6798125

If we take a look at Volumio - it uses armv7, so it's 32 bit. In it's upcoming release, fusion DSP plugin would be added (based on Camilla).

Correct me if I am wrong, but in that case, Camila DSP via it's plugin would be forced to use float32, that's accelerated by neon?

 

[phofman]

If we take a look at Volumio - it uses armv7, so it's 32 bit. In it's upcoming release, fusion DSP plugin would be added (based on Camilla).

Correct me if I am wrong, but in that case, Camila DSP via it's plugin would be forced to use float32, that's accelerated by neon?

CamillaDSP uses f64 for its internal format and DSP calculations by default. IMO it does not matter if the system is 32 or 64bit. Using f32 must be requested at compile time by its option 32bit https://github.com/HEnquist/camilladsp#customized-build .

It's up to the Volumio team what options they chose. The plugin I found https://github.com/balbuze/volumio-...e1/plugins/audio_interface/CamillaDsp4Volumio has camilladsp precompiled (no idea about its compile options), but the included default camilladsp config captures only at 16bits from the snd-aloop (inside https://github.com/balbuze/volumio-..._interface/CamillaDsp4Volumio/Dsp4Volumio.zip ).