Improve ConvolverNode

[b-ma]

Hey, it's been quite some time, hope you are well !

I just had this very rapid test with the ConvolverNode to use the fft-convolver crate, and it seems quite interesting in terms of performance (numbers are speedup vs. realtime in the benchmarks example)

main

Convolution reverb: 16.92

feat/improve-convolver

with let _ = convolver.init(RENDER_QUANTUM_SIZE, &samples);

Convolution reverb: 20.91

with let _ = convolver.init(RENDER_QUANTUM_SIZE * 2, &samples);

Convolution reverb: 43.12

with let _ = convolver.init(RENDER_QUANTUM_SIZE * 4, &samples);

Convolution reverb: 81.54

with let _ = convolver.init(RENDER_QUANTUM_SIZE * 8, &samples);

Convolution reverb: 134.55

Don't really know all the implications of this partition size parameter (probably some trade off between speed and quality given the layout of the numbers...) but all unit tests are passing on my side and I didn't ear anything really chocking

I just hacked it very rapidly (i.e. tail time needs to be re-implemented, etc.) and I won't have much time to work on it until at least January.

But wanted to have you feeling about that, do you you think it worth it to continue?

[orottier]

Hello, it has been a while indeed. All is good here, I'm actually slowly working on a new release recently (just maintenance stuff).
The crate looks really promising. Real-time, partitioned and efficient is something I did not manage to find in pure rust two years ago.
Hence, let's continue exploring this solution. The tests in the rust code are quite light, so I wonder how it holds up to the wpt suite.
I'm also quite busy this month but I will try to take it a bit further. Would be really nice to have this working since the ConvolverNode may be the one true thing that still needs improvement in our lib.

[b-ma]

Ok cool

The tests in the rust code are quite light, so I wonder how it holds up to the wpt suite.

I checked the wpt, no difference, it even improved a bit for some reason

[orottier]

We could also choose to implement the multi-channel features later. I think the PR as is is good enough to be included in a release this year

[b-ma]

We could also choose to implement the multi-channel features later. I think the PR as is is good enough to be included in a release this year

Sounds good indeed, I will try to clean it up today or tomorrow

[orottier]

Mono processing looks good. Ready to merge right?

[b-ma]

Yup let's merge as is! We can continue later