gensound creates temp audio files which don't get clean up afterwards

[kodcdangky]

Hello there,

Really loving this project, every feature I have tried so far has been an absolute wonder!

I have only 1 problem so far that I hope someone can help me fix.

import time
import gensound as gs

audio = gs.WAV('D:\\System\\Desktop\\water_alarm.mp3')
audio.play()
time.sleep(2)

When I run this code on PyCharm, it plays normally leaving nothing behind (or at least none that I can find anywhere). But if I export this code using pyinstaller into an .exe file and run that, a folder called gensound_temp is created where I have the .exe file, and a .wav temp file is generated every time play() is called. This folder and all its files are kept permanently, or until I manually delete them. Is it possible for these files to be cleaned up afterwards?

If this is unwanted behavior, please let me know what info can I provide to help diagnose the problem.
I currently have gensound 0.5.2, numpy 1.22.1, pygame 2.1.2, pyinstaller 4.8 installed, if that helps.

[Quefumas]

Welcome, and thanks for the feedback!

My suggestions:

1. If all you're doing is playing from already-existing WAV/MP3 files, I suggest not using gensound at all, since it would be much simpler with PyGame or any other alternatives. I can help navigate those, if needed.
2. It's possible (maybe?) that due to pyinstaller, gensound can't find PyGame and then resorts to more primitive I/O capabilities. See if you can run from gensound.io import IO; IO.status(True) in both environments, and copy the output. This should tell you which I/O capabilities gensound is aware of in both environments, and we can test this hypothesis and see what we can do.
3. Try to load a WAV rather than MP3 file (you can use the temporary one for example), and see if playback creates another new file or not.
4. Try to load an MP3, but instead of play(), use export('filename.wav'), and see if this creates any redundant wave files (other than the desired output).

Steps 3&4 will let us know if the temporary file is created because of playback limitations, or due to reading MP3 files, which are not supported natively by Python. If step 3 does not create an extra file, a temporary solution would be to just use WAVs (gensound can help convert), although that's far from ideal of course. Another non-ideal solution is to call IO.cleanup() before exiting the application, which attempts to clear these temporary files, with varying amounts of success.

[Quefumas]

And two more things:

1. if the application plays the same file/signal more than once, this is much more efficient:

wav = gs.WAV('D:\\System\\Desktop\\water_alarm.mp3')
audio = wav.realise() # create Audio object once
audio.play() # run this as many times as you want
time.sleep(2)

This is because every time you run play() on the original WAV object, this reprocesses the entire audio stream. This way, the processing is separated from the playback (see more in the IO Guide). Perhaps I should make it so that calling play on a pure WAV object with no modifications should skip processing and play directly from file.

2. Use time.sleep(2) with some care. It will act differently according to the playback interface used. I think PyGame plays the audio in the background (though maybe we can pass an argument to change that), but some other interfaces will pause until playback ends.

[Quefumas]

Thanks, you're raising a lot of good issues here.

Before I start getting wordy, a fix on my suggestion:

from gensound.io import _IO_pygame

audio = gs.WAV('D:\\System\\Desktop\\water_alarm.wav')
processed_audio = audio.realise(sample_rate=44100)
processed_audio._prepare_buffer(2, None) # <----- added this
_IO_pygame.playback(processed_audio)

[Quefumas]

Wait! I think this should be working too:

import gensound as gs
from gensound.io import IO

IO.set_io("play", "pygame") # call this once; this forces gensound to use pygame

gs.WAV('D:\\System\\Desktop\\water_alarm.wav').play()

The method set_io was meant to switch between available interfaces, but I don't see why it shouldn't work even for devices that it doesn't know are available.

[Quefumas]

The boring part:
The reason your workaround was working was that pygame was always available (importable), but the mechanism that gensound uses for detecting which interfaces are available is not working well inside a pyinstaller environment (not surprising). Gensound could tell what's available by trying to import it, but that would take time and fill up memory for absolutely no good reason.

This mechanism is lacking in other areas too, but if using set_io works for this, that's already something.

[kodcdangky]

IO.set_io worked, and for .mp3 as well as .wav, thank you so much!
Also I always like to try to understand the underlying mechanism, so no worries
Cheers! And thanks again

[Quefumas]

Thank you for letting me know about this. I'll try to figure out how to fix this in future versions, and now at least we know of a workaround.

(On an aside, you asked about console window opening - I haven't really used any of PyInstaller or PyGame, so I can't help you much, but I would try to see whether using any other feature of PyGame has the same effect. Perhaps PyGame is used to running in a console by default, so it creates one for itself)

[kodcdangky]

#28 (reply in thread) <- The answer followed by the explanation