Possible f32 rendering issue?

[Hooperla]

Hello there! Found out about this release a bit ago and saw that you made a windows release. Decided to try it out and I'm very happy with the results, seems like a pretty easy way to make sure I have some SC-55 goodness!

However, I noticed that, for whatever reason, using f32 produces different results compared to f16 or s32. In the latter the track will end on the last note as described by the 'cut' default option, but f32 seems to 'double' the track with a long period of silence afterwards for whatever reason. I say 'double' because it seems to match the length that has audio with a section of silence. My best guess is that it either just isn't cutting off correctly or that it could be looping the track without really 'playing it' again due to midi data telling it to do so or something? Not the best of guesses but it's what I have with my limited knowledge.

I've attached a zip to this with the console output of me doing all three renderings on the same midi + the midi to maybe help in figuring this out.
If all of this is me misunderstanding how f32 is supposed to work and this is 'correct' behavior than I apologize in advance.

Stuff for Issue Submit.zip

[jcmoyer]

Hi, thanks for reporting this. I found a silly mistake and I think I've fixed the problem. I'll generate a new release later tonight after running the test suite. As a workaround you can open the wav file in audacity and re-export it to correct the playback length.

[jcmoyer]

I published 0.4.1, let me know if this doesn't fix the issue.

[Hooperla]

Seems to have gotten rid of the 'doubling' issue I was describing, thank you very much for the quick response! However, I did notice after making the renders with the new version that f32 seems to make the audio louder than s32 or s16. Again, don't know if that's intentional or not, but I figured since I was going to respond to this that I might as well throw in some Spek Spectrograms to see if this is anything as well.

0.4.1 Render Test Stuff.zip

[jcmoyer]

Yes it's intentional. You can see some discussion about it in #5 and nukeykt#59 (@Falcosoft contributed the f32 normalization). I would recommend rendering to f32 for the increased dynamic range and then using another program like ffmpeg to scale the audio if necessary.

[Hooperla]

Ah, thank you for the explanations!