Hello,
I have finally completed the bidirectional conversion using JS. The data in the audio buffers before and the conversion to a Jitter matrix and after the back conversion appear to be the same. However, the audio time domain reconstruction at the output of fluid.bufstft~ has some clicks.
Could some STFT guru have a look and suggest some possible cause?
MSPJitter V2_1.zip (7.7 KB)
Turns out that Discourse is really weird about letting one download attached js files: it thinks one is trying to execute them instead (perhaps it’s to do with the web server settings 
). If you’re able, could you edit the post to just zip everything up instead? That will also be more reliable for preserving the names of files.
Done.
Thanks. Still missing jit2buf_RBV1.js?
In any case, I can make two general reccomondations if you want to resynthesise:
Otherwise reconstruction will be (possibly very) approximate…
Ah, sorry. This is corrected now.
I know about the hop size but it would at least reduce the sound duration by 2, so it’s not possible at the moment. I think it can have an acceptable sound but if you can have a look, it would be nice.
I’ve tried with a hop size of 512 but I still got a bad reconstruction from the Jitter matrix.
Also, I really can’t use the Nyquist band because my picture must be 512x512 pixels. But I add it with a zero magnitude to the matrix data when converting back to STFT data.
I didn’t get it to work yet, before having to move on to something else, but there’s definitely an off-by-one error in the counting somewhere in either of the js files. The 0th channel doesn’t seem to get filled in the reconstructed buffers.