Can I use Demucs-GUI ... as a CLI application?

[schweini]

This might seem like a silly question, but Demucs-Gui seems to be the simplest (only?) prepackaged cross platform binary of Demucs available.

Is there a way to easily run it completely headless and from the command line interface? I would love to once in while have a webserver of mint stem some tracks, but would really rather not get into the whole Python woods just for that. Calling Demucs-Gui from the CLI would be ideal form my use case.

[CarlGao4]

I've packed a pre-installed environment of demucs CLI.
https://1drv.ms/u/c/ea7bcc9ccd2da37c/ERL9OQVmIoVEmsHGscU3NWYBPmA3repwjmUReh4nmK6vRg?e=MynyPp

[dts350z]

@CarlGao4 I get:

Traceback (most recent call last):
File "demucs-standalone.py", line 5, in
File "demucs\separate.py", line 168, in main
File "demucs\api.py", line 307, in separate_audio_file
File "demucs\api.py", line 224, in _load_audio
NameError: name 'ta' is not defined. Did you mean: 'th'?
[39720] Failed to execute script 'demucs-standalone' due to unhandled exception!

and Claude says: Based on that GitHub discussion and the error you're encountering, it appears you're trying to use a standalone CLI version of Demucs-GUI that has a packaging issue. The 'ta' is not defined error strongly suggests that torchaudio (commonly abbreviated as ta in Demucs code) wasn't properly included in the standalone build.

[CarlGao4]

Did you extract all the files from the archive before running?

[dts350z]

Yes I used "extract all" in windows file explorer, and then launched from a cmd in the extracted folder:

D:\Glenn\Downloads\demucs\demucs

D:\Glenn\Downloads\demucs\demucs>demucs.exe --two-stems drums "D:\Glenn\Music\Love Alive.flac"
Selected model is a bag of 1 models. You will see that many progress bars per track.
Separated tracks will be stored in D:\Glenn\Downloads\demucs\demucs\separated\htdemucs
Separating track D:\Glenn\Music\Love Alive.flac
Traceback (most recent call last):
File "demucs-standalone.py", line 5, in
File "demucs\separate.py", line 168, in main
File "demucs\api.py", line 307, in separate_audio_file
File "demucs\api.py", line 224, in _load_audio
NameError: name 'ta' is not defined. Did you mean: 'th'?
[39720] Failed to execute script 'demucs-standalone' due to unhandled exception!

[CarlGao4]

This is quite strange. Can you make sure that all files are there? I've attached a screenshot of the _internal folder:

[dts350z]

I'll try unzipping at the top of drive, so there are no spaces in the directory path...

No luck. I also unzipped using winwar, vs. Windows.

D:\demucs\demucs>.\demucs.exe "N:\Correlation Tourture Tracks\Dirty City.flac" -n mdx_extra --float32 --clip-mode clamp -o "N:\Correlation Tourture Tracks\output_folder"
Selected model is a bag of 4 models. You will see that many progress bars per track.
Separated tracks will be stored in N:\Correlation Tourture Tracks\output_folder\mdx_extra
Separating track N:\Correlation Tourture Tracks\Dirty City.flac
Traceback (most recent call last):
File "demucs-standalone.py", line 5, in
File "demucs\separate.py", line 168, in main
File "demucs\api.py", line 307, in separate_audio_file
File "demucs\api.py", line 224, in _load_audio
NameError: name 'ta' is not defined. Did you mean: 'th'?
[51456] Failed to execute script 'demucs-standalone' due to unhandled exception!

D:\demucs\demucs>

Also tried running from the _internal directory:

D:\demucs\demucs\_internal>..\demucs.exe "N:\Correlation Tourture Tracks\Dirty City.flac" -n mdx_extra --float32 --clip-mode clamp -o "N:\Correlation Tourture Tracks\output_folder"
Selected model is a bag of 4 models. You will see that many progress bars per track.
Separated tracks will be stored in N:\Correlation Tourture Tracks\output_folder\mdx_extra
Separating track N:\Correlation Tourture Tracks\Dirty City.flac
Traceback (most recent call last):
File "demucs-standalone.py", line 5, in
File "demucs\separate.py", line 168, in main
origin, res = separator.separate_audio_file(track)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "demucs\api.py", line 307, in separate_audio_file
return self.separate_tensor(self._load_audio(file), self.samplerate)
^^^^^^^^^^^^^^^^^^^^^^
File "demucs\api.py", line 224, in _load_audio
wav, sr = ta.load(str(track))
^^
NameError: name 'ta' is not defined. Did you mean: 'th'?
[66776] Failed to execute script 'demucs-standalone' due to unhandled exception!

[dts350z]

Unzipping with 7zip (vs. winrar or Windows native) also didn't help.

I will also try on a "clean" install of windows 11:

Different problem (first anyway):

C:\Users\glenn\Downloads\demucs\demucs>.\demucs.exe "C:\Users\glenn\Music\Love Alive.flac"
Downloading: "https://dl.fbaipublicfiles.com/demucs/hybrid_transformer/955717e8-8726e21a.th" to C:\Users\glenn/.cache\torch\hub\checkpoints\955717e8-8726e21a.th
Traceback (most recent call last):
File "urllib\request.py", line 1348, in do_open
File "http\client.py", line 1286, in request
File "http\client.py", line 1332, in _send_request
File "http\client.py", line 1281, in endheaders
File "http\client.py", line 1041, in _send_output
File "http\client.py", line 979, in send
File "http\client.py", line 1458, in connect
File "ssl.py", line 517, in wrap_socket
File "ssl.py", line 1108, in _create
File "ssl.py", line 1379, in do_handshake
ssl.SSLCertVerificationError: [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:1006)

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "demucs-standalone.py", line 5, in
File "demucs\separate.py", line 123, in main
File "demucs\api.py", line 119, in init
File "demucs\api.py", line 204, in _load_model
File "demucs\pretrained.py", line 78, in get_model
File "demucs\repo.py", line 160, in get_model
File "demucs\repo.py", line 139, in get_model
File "demucs\repo.py", line 139, in
File "demucs\repo.py", line 68, in get_model
File "torch\hub.py", line 746, in load_state_dict_from_url
File "torch\hub.py", line 611, in download_url_to_file
File "urllib\request.py", line 216, in urlopen
File "urllib\request.py", line 519, in open
File "urllib\request.py", line 536, in _open
File "urllib\request.py", line 496, in _call_chain
File "urllib\request.py", line 1391, in https_open
File "urllib\request.py", line 1351, in do_open
urllib.error.URLError: <urlopen error [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:1006)>
[1800] Failed to execute script 'demucs-standalone' due to unhandled exception!

[CarlGao4]

Then this is a network error, it failed to download pretrained model. Maybe you can try manually downloading https://dl.fbaipublicfiles.com/demucs/hybrid_transformer/955717e8-8726e21a.th to C:\Users\glenn/.cache\torch\hub\checkpoints\955717e8-8726e21a.th

[dts350z]

Actually, I think you are missing a cert package, and also some import torch audio as ta lines are commented out.

FYI I worked with Claude Sonnet 4.0 (and MCP) to make a clean working version:

https://drive.google.com/drive/folders/19aM9_g34pv4PoJFrCmISUSX6I0eX6owQ?usp=sharing

In that folder are the step by step instructions to build the package, a build script, and the resulting package (zipped).

This has been tested on Windows 11 machines with and without GPUs.

Thanks for nudging me to get this done. I need this for my up-remix script project.

[CarlGao4]

This is really strange as you are the first user to experience this issue...
As you've seen, the error is torchaudio package not found, but you've located it inside the package

[neuronetv]

I would like to try this demucs standalone package however all the links I've found are only downloadable through a browser. is there a direct link so I could get it with wget?

[CarlGao4]

https://demucs-gui.carlgao4.workers.dev/files/demucs.7z

[ianier]

Is there a way to easily run it completely headless and from the command line interface?

I had exactly the same question and couldn't find any suitable solution, so I went for my own standalone build. I wrote this blog post with step-by-step instructions to build a standalone demucs CLI executable for Windows. Hope it helps.

[neuronetv]

thankyou for this blog post, I tried it in windows 11 and it ended up working.
I dont know if you want to update your blog but one brick wall I ran into is that it needs python 3.10.x to work. current python is 3.14 and it causes a build error with torchaudio<2.1,>=0.8_
in your blog you said 'copy both ffmpeg.exe and ffprobe.exe to the same directory where demucs_cli.exe is'. the second error is that one needs to copy all the files in the ffmpeg folder, not just ffmpeg.exe and ffprobe.exe.
your blog is very much appreciated, great work and thankyou

[ianier]

Thanks for your comments! Glad it helped.
In the blog article I linked to Python 3.9 in the prerequisites, which is what I used and worked. Regarding ffmpeg, ffprobe.exe and ffmpeg.exe should indeed be enough if you use the dist I linked to in the article.
It may be worth adding a disclaimer that the step-by-step method works when you rely on the actual versions of the software listed in the prerequisites, and deviating from there may require some tweaking, but can be made to work too :-)

[ianier]

FYI, I also compiled it with nuitka, and got 3 different versions:
-Single-file standalone exe, CPU-only, ~142 MB . This is the one that I'm bundling with Chronotron.
-Single-file standalone exe, with CUDA support, ~1.6 GB
-One-directory exe, standalone, with CUDA support, ~4.7 GB
If anyone is interested, I can share the executables somewhere.