When I first read about transient shaper, I was like “what’s the difference with a compressor? Is there one?”. And I tried to see how to get these transient without relying on the transient energy, with a relative power (ratio between the instant power and the mean power) filter, or its derivative, but nothing worked. Until someone explained that the gain was driven by the difference between a fast attack filtered power and a slower one. So here it goes.

First, let’s have a look on the filter graph:

I’ve surrounded the specific transient shaper part in with a dotted line. This is the difference with a compressor/limiter/expander: the way the signal steered the gain computation is generated.

Let’s start from a kick. The fast envelope follower can then be generated (red curve) as well as the slow envelope follower (green curve). The difference is always positive (if the two follower have the same release time value), so we can use it to compute a gain through our usual GainCompressorFilter.
Depending on whether you want to increase the transient or attenuate it, the ratio will be below or higher than 1 (respectively), which is what is shown in the last two plots here:

In the end, it’s all about the right algorithms. If you have a proper framework, you may already have everything you need for some filters. In the case of a transient shaper, I already had all the filters in my toolbox. Now I just need to make a plugin out of this simple pipeline!

The code for the last plots can be found on Github here: https://github.com/mbrucher/AudioTK/blob/master/Examples/dynamic/display_transient_shaper.py