One Way Forward

Could probably deal with the volume problem by programming on-off receptors.
Could probably also do an automatic removal of lower formants neurally.
It would be good to figure out what the brain really does, even if only small parts are implemented. (and I'm sure the biologists know a lot.)
I'm working on learning especially in multi-layer systems.
So, try learning formant separation, and then recombination for A1.
Then ground it like here (biologically implausible).
Then try other sounds (e.g. consonants, stops).
Stick volume control in some where.
Then try speech recongition.
Combine with higher level language processing.
Then try psychophysics simulations.
Might then compete with HMMs.
Could also combine with other neurally implemented behaviour.