everything just fades into noise at the end

When I faced artificial reverberation algorithms to the very first time I just thought why not just dissolve the audio into noise over time to generate the reverb tail but it turned out to be not that easy, at least when just having the DSP knowledge and tools of that time. Today, digital reverb generation has come a long way and the research and toolsets available are quite impressive and diverse.

While the classic feedback delay network approaches got way more refined by improved diffusion generation, todays computational power increase can smooth things out further just by brute force as well. Still some HW vendors are going this route. Sampling impulse responses from real spaces also evolved over time and some DSP convolution drawbacks like latency management has been successfully addressed and can be handled more easily given todays CPUs.

Also, convolution is still the ticket whenever modeling a specific analog device (e.g. a plate or spring reverb) appears to be difficult, as long as the modeled part of the system is linear time invariant. To achieve even more accurate results there is still no way around physical modeling but this usually requires a very sophisticated modeling effort. As in practise everything appears to be a tradeoff its not that much unusual to just combine different approaches, e.g. a reverb onset gets sampled/convoluted but the reverb tail gets computed conventionally or – the other way around – early reflections are modeled but the tail just resolves into convoluted noise.

So, as we’ve learned now that everything just fades into noise at the end it comes to no surprise that the almost 15 years old epicVerb plugin becomes legacy now. However, it remains available to download for some (additional reverb) time. Go grab your copy as long as its not competely decayed, you’ll find it in the downloads legacy section here. There won’t be a MkII version but something new is already in the making and probably see the light of day in the not so far future. Stay tuned.


  1. This is exciting! Thank you in advance!

  2. heavymetalmixer says:

    I’m definitely happy to see you want to create something new. Just like you did years ago, will you be explaning how your project/plugin works in a series of blog posts?

  3. nicolas herz says:

    quite interrested what you’ll come up with, herbert.
    while i think you’re spot on with what you wrote, it has to be said that convolution in terms of reverb has another drawback, which i think is way more crucial than latency:
    it’s the missing modulation in the feedback tail. you can’t modulate an impulse response in its feedback, as there simply is no feedback – you’re essintially playing back a captured sample. natural reverbs that actually sound good not only in terms of realistic fashion, but also provide what i call the “idealistic imagination” of a natural reverb in terms of musicality are mostly modulating – even if it is only a tiny bit (air pressure differences, etc). the reason for this to be a crucial factor is simple, especially for music where you often use sounds that don’t modulate their pitch in itself. send a plain waveform of whatever kind into a non-modulating reverb – after the initial delay pattern is run through once (just hold the sound you’re feeding into the reverb for 3 or 4 ec), you’ll hear no room information whatsoever anymore. that’s because reverbs are just delays. and if these are not modulating in time, they are placed phase shifted directly onto the input – statically. so all you end up is with a phasey release of the sound with close to no room information or no room information at all.
    this is the problem with non-modulating algorithmic reverbs and especially with _every_ convolution reverb. even if it has a “modulation” parameter, it is simply placed in the signal path right before the wet reverbation, so it is _not_ the delays in the feedback being modulated, but just the input that goes into the reverberation – which can’t do suffice – at least not as satisfying like delays that are modulated within the feedback path.
    just wanted to pont this out, because i think that _this_ is the most significant issue that just can’t be solved in an impulse resonse reverb – it’s simply systemic, so to say (if you’re not doing .dynamic convolution, which in a reverb very fast gets very expensive on the cpu).
    so it doesn’t help at all to create an impulse response of a modulating reverb tail – as soon it’s an impulse, it’s static – there’s no modulation. that is why impulse responses of an l480 with f.e. the random hall algorithm (which made the l480 so famous) with a lot of spin and wander can _never_ sound as good as the real thing – even if it was captured from it.

    again – curious what you come up with! take care!

    kind regards,
    nico from bigtone.

    • Hi Nico,
      glad to have you back here 🙂 and you are absolutely right that this can be seen as another drawback of working with impulse responses. While that sort of modulation isn’t really a property of a natural room (as you pointed out already) its just amazing how we got used addicted to that “random hall” stuff introduced during the Lexicon area. Originally, its purpose was just to further increase the modal density which was the challenge in using delay networks. From todays perspective I see the whole epicVerb design being a hommage to exactly those phenomena 🙂 I just would execute it differently today and do not follow that “jack of all trades” device approach anymore. Also the production aestetics changed over time but this might be something interesting to further look at in another article and discussion?

  4. can’t wait. if it’s even better than epicverb than it must be good.

What do you think?

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: