Someone will come up with a decent accelerator eventually. Right now we're in the pre-soundblaster days where audio sythesis is being done by the CPU but for neural networks.
We're repurposing graphics hardware, these are primitive times.
Also, there, I explained it for you.