Hello again everyone! I’ve been updating noise-repellent and I’ve implemented a masking feature similar to what other commercial noise reduction tools have had for years. I will appreciate if anyone could test it and give me some feedback.
I can’t compare with commercial solutions, but based on the test i’ve made, it’s pretty impressive ! Good job ! Far better than Audacity’s noise remover.
great lucianodato, I ll test this asap!
Thanks guys but to be honest, I did a pretty lousy job with my masking implementation, it does not work correctly yet. But I have implemented a nonlinear spectral subtraction that achieve somewhat similar results to what masking is supposed to achieve. I will fix masking as soon as I figured out what I’m doing wrong (that is in DEV branch). Since I’ve posted last time I was able to fix lots of small bugs, so now it should perform better and using less CPU (no masking but fat better results than simple power subtraction). Maybe I will do a release soon if everything is working properly. Let me know what you think!
@x42: Thanks for the tip on the spectral editor. I finally had time to sit down and learn the Audacity spectral editing tool, I think it is probably the right tool for the job. I’m really wishing for a way to setup a “template” of a set of frequencies to remove and durations the effect is active. This alarm requires about 9 or more edits per occurrence. I’ve been working on the first three or four minutes of audio for the last two hours, just incredibly tedious. It does allow salvaging what I thought would be unusable without being able to use something like iZotope tools, so it’s worth it I think. I don’t use Audacity for music mixing, so I did not realize that it had such useful editing tools in the latest versions, good to know for special occasions like this.
Attempting to try out noise repellent now, but does not load with Ardour built from git as of 13 Feb 2017. Anyone tried the most recent noise repellent with the most recent Ardour?
Addendum: does work in 5.5. No time to help debug at the moment, just posting this as info in case someone else has time to look.
Noise Repellent is amazing! Thank you a lot, lucianodato! Even in the current condition (git master branch) it is the best open-source spectral noise eliminator out there. At least for Linux.
Hello friends! I’m still working on noise-repellent. Soon I will do a new release. It will include saving the noise profile with the session(thanks to a contributor) and major changes to the reduction algorithm. I’m getting closer to comercial stuff in terms of quality so don’t hesitate to try it! (Excluding izotope of course)
wonderful, looking forward to the new version! keep us posted
I’m really close to release but some testing from any of you guys would be useful. This release is a major rewrite of the reduction algorithm. Hopefully it will sound much better than previous versions. You can find the explanation for every slider and button in the readme file. If you find anything wrong don’t to hesitate to contact me on IRC (arguy) or to file a bug on github.
I just grabbed your latest version of this. I have been playing around with some Beatles Sgt. Pepper bootlegs of the (semi) raw tracks, and WOW, I am floored by how well this removes all the tape hiss and other things as well without sacrificing the audio quality! I was counting on these tracks to be a good test with your plugin because it’s impossible to say how many generations down these boots are, and they were pretty damn noisy but not now
You are correct, this plugin is very close to commercial quality, if not already. Keep up the fantastic work on this. Great job!!
sounds great Lexridge! I will test also as soon as i get my hands on some noisy dialog again…
Sorry I am on the road for the next couple of weeks, doubt I will get much chance to test, but looking forward to hearing it when I get back!
“If you want lower FFT sizes only to avoid transients distortion” - yes, exactly, ) the musical noise is one really bad thing when cleaning field recordings (I record bird calls). “If you have an example where previous versions of noise repellent were working better than current one” - the older versions are MUCH worse than the current one. You’re doing a great work!
Oops, transients distortion is another thing. But anyway, yes, I need lower FFT sizes for quicker response.
Bird calls don’t have too much low frequencies. Makes sense. Alright I’ll see what I can do.
Thanks for your feedback guys! It’s out in the wild now. I have just released it. Hope it’s useful for you!
More on 3:
If you are referring to manually setting the FFT size for increase or decrease the resolution of the processing than wouldn’t be too hard to implement. The thing is that often lower FFT sizes sounds weird when processing sounds with lower frequencies such as speech. A certain minimal frequency resolution is needed in order to avoid low end distortion because FFT filter banks are fixed width. An approach to avoid this is using something like a constant q transform or processing in the mel or bark domain. That is doable too. If you want lower FFT sizes only to avoid transients distortion I wouldn’t use that strategy since would distort the low end of your audio. Using adaptive multi-resolution STFT would be a good way or some transient protection criteria. The other thing that it worth mentioning is that instead of using an FFT size is better to select a window size in ms as this would preserve the resolution in higher sample rates. Again working on that.