sevagh / free-music-demixer Goto Github PK

running macOS 14.2.1 and a MacBook Air M1

Unhandled Promise Rejection: RuntimeError: Out of bounds memory access (evaluating '(_free=Module["_free"]=Module["asm"]["x"]).apply(null,arguments)')

This is from Safari, FF shows similar errors. File is not processed and no progress bar shows up. No errors in dev logs:

[Javascript 21:09:14] Fetched and cached model files
[Javascript 21:09:14] Selected model: demucs-6s
[Javascript 21:09:40] Initializing 4 workers!
[Javascript 21:09:41] Beginning demix job
[Javascript 21:10:35] Worker 2 is ready!
[Javascript 21:10:35] Worker 3 is ready!
[Javascript 21:10:35] Worker 0 is ready!
[Javascript 21:10:35] Worker 1 is ready!

Hi,i have a old capture record from TV in the 90, was record in a one of the first dual audio vcr of the epoque, so they save in the tape the original track in v.o+dubbed track. Unfortunly these vcr uses a fine separation line in the tape to separate the audio Channels, and the tape get this line deteriorated over the years, so today the vcr can't separate the audio and read the the two audio tracks at the same time, resulting in a audio track with the v.o+dubbed track playing at the same time.

My questions is if with the help of a Ia can be possible isolate and separate these audio voice tracks to their normal state.

Is there a compiled version somewhere?

3080 GPU

  0%|                                                                                                                                                                                         | 0/1 [00:06<?, ?it/s]
Traceback (most recent call last):
  File "/home/user/.virtualenvs/demixer/bin/umx", line 8, in <module>
    sys.exit(separate())
             ^^^^^^^^^^
  File "/free-music-demixer/vendor/open-unmix-pytorch/openunmix/cli.py", line 167, in separate
    estimates = predict.separate(
                ^^^^^^^^^^^^^^^^^
  File "/free-music-demixer/vendor/open-unmix-pytorch/openunmix/predict.py", line 78, in separate
    estimates = separator(audio)
                ^^^^^^^^^^^^^^^^
  File "/home/user/.virtualenvs/demixer/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/free-music-demixer/vendor/open-unmix-pytorch/openunmix/model.py", line 314, in forward
    targets_stft = targets_stft.permute(0, 5, 3, 2, 1, 4).contiguous()
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 2.56 GiB (GPU 0; 9.77 GiB total capacity; 5.71 GiB already allocated; 1.37 GiB free; 6.99 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Dear @sevagh ,
I wanted to "downlad weights" but it was aborted with the error message in javascript console:

CompileError: WebAssembly.instantiateStreaming(): Compiling function #21 failed: Wasm SIMD unsupported @+3930

I don't understand much of WASM. Could you please help?

My System: Firefox or Chromium
Ubuntu 22.04.3 LTS
AMD® Phenom(tm) ii x4 955 processor × 4

First of all, thanks for this! I've been doing tabs for many years, and learned different ways to trying to separate tracks as best as possible, but this takes it to another level.

It would be nice to have the melody potentially separated by different guitar/keyboards/etc tracks. I hope this can be achieved eventually.

Hello,

I've been attempting to try this online tool in order to see how capable it is to demix some of my favourite tracks.
So, I've used an 8.8 MB .mp3 file as input, and the website seems not able to process the result.

I've noticed the dev log keeps stuck on this line:

[WASM/C++ 17:28:32] Getting waveforms from istft
[WASM/C++ 17:28:34] Copying waveforms

And if I take a look at JS console, this message is printed:

Uncaught (in promise) DOMException: Failed to execute 'postMessage' on 'DedicatedWorkerGlobalScope': Data cannot be cloned, out of memory.
    at https://freemusicdemixer.com/worker.js:25:13

I think I'll have to try running this locally without a web interface so that it can get enough memory to work properly.
Just wanted to let you know, I think UI should handle this error to avoid keep user waiting in vain.

For some reason, the only result I'm getting (at least for the file I've tried) is just garbage noise. I'm not sure if there's anything I'm doing wrong in terms of the format I'm uploading (I've tried 24-bit FLAC and 16-bit WAV, both to no avail) or whether it's just a song that doesn't work well, but something clearly isn't working right for this song.

If there are any other details I can provide that might help with debugging this, please let me know.