I am interested in comparing 0.5-second 44100/s signals when different ADSR envelopes are applied to them. However, I don't want the comparison to be as fine-grained as simply calculating a 5000-dimensional envelope - ideally I'd like a <100 dimension vector that told me everything I needed to know about change in amplitude over time, and which was not influenced by the actual frequency. What are some 1-10 dimension statistics that could capture meaningful information about how the dynamics are changing over time?
An ADSR has exactly five parameters, and you need these five to characterize it - every set of parameters yields a different envelope.
And frankly, the ADSR envelope is pretty easy to visualize, and to understand:
So, calculate the envelope of your signal, find the
- peak amplitude,
- the time it takes to reach that peak
- the plateau level
- the time it takes to reach that plateau from the peak, and
- the time it takes for the envelop to decay after leaving the plateau.
These are your five degrees of freedom, and they can be quite directly read from a slightly averaged envelope.