I am getting started with pd and have built a simple patch which interprets midi files using mrpeach's midi file object. It plays them on an array of simple square wave oscillators and then routes the output to the dac and a wav file.

Waiting 20 minutes for the Ode to Joy to render to wav inspired a question: is there a way to use pure data to render midi to synthesized wav asynchronously, as fast as the CPU can handle? Given how many pd objects rely on times given in ms, it seems like this might be difficult.

hmm, you can change the samplerate to a much higher one with [block~] or [switch~], but i'm not sure how well [writesf~] is going to work with the upsampled data. You might need to write your output audio into a table (array) first, and then save your .wav file from that array.

by upsampling (check the help file for block~ to see how to do this), you'll still have to use trial and error to see how fast your cpu can process the data, but for simple midi to square wave synthesis, you'll probably be able to get quite a few times faster than realtime. It might be possible to get 20 minutes down to 1/16th of that time??

as you said, the time-based objects are going to make things difficult for you. You will need to scale any time values by the same fraction that you upsample by.

yeah...it's kinda difficult, but should be possible.

Have a go if you want. You can post your patch if you get stuck somewhere.