DBraun
commented
3 weeks ago Thanks. I think it's an improvement over the one from Faust Libraries, but still not as good as Dexed etc.
In my experience, once you can get to a batch size of 512, the GPU parallelism outweighs the slower clock rate.
For just DawDreamer, it would be like this
import dawdreamer as daw
from scipy.io import wavfile
SAMPLE_RATE = 44100
engine = daw.RenderEngine(SAMPLE_RATE, 512) # 512 block size
faust_processor = engine.make_faust_processor("faust")
faust_processor.num_voices = 1
code = open("custom_dx7.lib", "r").read()
i = 1 # between 1 and 32
code += f"\nprocess = dx7_algorithm({i});"
faust_processor.set_dsp_string(code)
print(faust_processor.get_parameters_description())
engine.load_graph([(faust_processor, [])])
faust_processor.add_midi_note(60, 100, 0, 3)
engine.set_bpm(120.)
engine.render(4.) # render 4 seconds
audio = engine.get_audio() # shaped (C, N samples)
wavfile.write('faust_dx7_demo.wav', SAMPLE_RATE, audio.transpose())
print('All done!')Note that the parameters are not typical [0-1] VST parameters.
It should be possible to adapt this to the multiprocessing example: https://github.com/DBraun/DawDreamer/tree/main/examples/multiprocessing_plugins
update: made the faust code mono.
The improved dx7.lib is great!
But even with a GPU, JAX rendering is I think slower/more expensive per output than multicore CPU.
I've tried to get CPP rendering to work with sndfile architecture, but that isn't working :(
I've tried stripping JAX from
dx7_render.pyand not transforming the values, but I end up with empty audio files (or just direct sine-tones always at a high pitch).Any help would be greatly appreciated!