hej luciano,havent tested the last version yet,but will let you know… the previous version worked already great for me, the only thing: i load the plugin and do “learn” , afterwards i hear the cleaned up audio. when i shut down ardour / mixbus and re open again, the audio is noisy again and i have to do “learn” again… I m not sure if this is avoidable within the architecture of the plugin, but would be great.
calimerox, state save/restore was added a few months ago. The learned noise pattern should be remembered with a recent version.
Yes lucianodato and Paul, please pkg with Ardour! 
@Ricardus Agreed. @lucianodato, there is no reason users can not update to your latest version even with if bundled within Ardour. It is kind of silly to expect the Ardour team to update their dist based on plugin updates. Ardour is the main project after all.
If it does not already exist, here would be a very good feature/extension to add to lv2 plugin spec: The ability to update (if desired) on their own, independent of the host. Would probably not be easy to implement however under the current lv2 API.
I’m pretty sure that currently if a LV2 plugin exists twice in the paths that Ardour looks in, then the biggest version is loaded (with a warning). I noticed that when I used Mixbus on Linux with x42 plugins both bundled with Mixbus and installed with the debian package.
Thus, if people want to get a more recent version of noise-repellent than the one bundled (if/when it is bundled), they can do that by downloading it and putting it in ~/.lv2 or similar.
Hi everyone! Sorry for the delay but I’m now trying to integrate noise-repellent with TravisCI in order to get binaries for other OSes. Would Ardour need binaries for OSX and windows multi-arch? What is the requirement in terms of architecture for Ardour to include it?
We would just merge the code (as we have done for the other a-* plugins.
Should I work in a pull request then? Would a git submodule make sense in this case?
definitely no submodule. we want to wait till you feel that NR is stable, because we’ll likely do a bit of deeper integration, such as “learn noise from selection”.
1 Like
I feel that dsp-wise is fine now. Improving current algorithm would imply a big architecture refactoring and for that I prefer to go the DPF route and do a GUI at the same time. There are few quirks now that I’m working on but I’m planning a fix release for this week so once that’s done you are clear to integrate it. Of course I can help in whatever you guys need.
Thinking a little bit deeper making a small library out of the dsp part of noise-repellent might be better for integration with Ardour.
It needs a rewrite in order to be clean for it to be integrated with Ardour. I will release another fix version really soon just to fix the state saving issues.
I’m already rewriting most of it’s code in the development branch and taking care of a more cleaner and extensible architecture design using OOP like abstractions. Also I will remove all the “adaptive” features from it and do a separate plugin for speech denoising using Xiph’s RNNoise library as a vocal activity detector and noise repellent algorithms to reduce the noise (this gives me flexibility to variate the processing block size which is needed for better denoising in those cases). So with both plugins all cases can be covered better. If anyone has a strong opinion about this I’m open to discussion. Luckily Ardour 6 is far off son maybe I can have something before they release (not every DAW ships with a denoiser, it can be an attention grabber).
So this is far off now? Bummer.
@Ricardus:
Nothing has changed, Ardour 5.12 is already released, so too late to include with 5.12. Paul has stated many times that Ardour 6.0 will be released much later, there are many changes coming, so luciano continuing to improve Noise Repellent will neither slow down nor speed up the release of Ardour 6.
And just as last week, you can download Noise Repellent and use it with the version of Ardour you already have. Nothing to be bummed about at all.
@ccaudle addressed it perfectly. I forgot to link speech-denoiser the plugin that uses RNNoise for noise reduction. If you want to try it in it’s current state here it is --> https://github.com/lucianodato/speech-denoiser
wow, will try that version soon on speech (and that s where i usually need a denoiser anyways…)! thanks again for this outstanding work! !
lucianodato,
Any docs for these plugs at all?
There is some instructions and recommended uses in the HOWTO file in github (see README too). I will move that to the github wiki soon though. What is your concern? If you had used any reduction tool, noise-repellent works almost the same. If you want the nerdy details there is doxygen documentation for the code and there I explain in each funcion what papers and methods I have used. Yes there is no manual yet but I feel that is not necessary to begin using it. I can work on a video tutorial too if you feel that is something needed.
I need to read something, because the interface is far from obvious. I have used any number of NR things in the past, and I could not get this to do anything.
Did it worked after reading the howto? You are probably not learning the noise profile. You have to loop a noise only section and press the learn profile for a second or two and then turn it off. Then the reduction will happen.