how do I render a midi file and load the audio data to memory?

[ghost]

I'm using python. and what I've been doing lately is: using the CLI render the midi file to file x and then read the file and load it to memory. I know that the writing-reading routine is quite slow and in my use case efficiency is crucial. So I want to skip that part and just load the output to memory. How do I do that?

[mawe42]

Hi!

Assuming you are on Linux or a similar platform: If you specify - as filename, then FluidSynth will render the audio to stdout instead. Combine that with -q to make sure no extra output messes up your audio stream:

fluidsynth /path/to/font.sf2 /path/to/file.mid -q -F-

Then just execute that using Python and capture stdout, for example with subprocess.check_output.

Cheers

[ghost]

I did see the docs. I looked for any function that returns the audio samples from a midi file and found none. Can I open an issue requesting it to be added?

[mawe42]

Not sure what you mean... MIDI files don't contain audio samples. If you mean a way to render a MIDI file to audio via the API (which is what you are currently doing via the CLI), then the above referenced example code is what you want.

[ghost]

yeah, I know. but like I said, the reading-writing routine is quite slow. I would like for a function that takes a given midi file, from it generate audio samples with the help of some soundfonts loaded with some settings and NOT write them to a file but return them as an array (or two in the case of stereo audio).

[mawe42]

If your main performance problem really is writing the synthesized audio to a file/pipe and reading the data into Python, then I suggest to ditch Python and go straight to C and use the audio rendering API to retrieve the samples: https://www.fluidsynth.org/api/group__audio__rendering.html

However... my guess is that you haven't done any profiling to find out where your performance bottleneck is. I would be extremely surprised if file I/O is your problem here.

[ghost]

tbh, I know that the file I/O is slower than just loading to memory, I just don't know how slow. but I need it to be as fast as possible so I can spend a few hours creating python bindings for some C functions (A language which I never used, but ditching python isn't an option). and for the record merely loading takes about .2 seconds while the rendering takes like 1.6 seconds so I'm hoping to get at least 12.5% faster rendering.

Anyways, thank you so much for helping me out. I really appreciate it. I hope you have a lovely day. :D