Where communities thrive

  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
  • Jun 05 23:28
    henry-kanengiser commented #413
  • Jun 05 15:13
    ateammeech labeled #413
  • Jun 05 15:13
    ateammeech opened #413
  • Jun 05 14:19
    Vnicius commented #354
  • Jun 05 13:08
    romi1502 closed #412
  • Jun 05 12:56
    Faylixe synchronize #412
  • Jun 05 12:56
    Faylixe synchronize #412
  • Jun 05 11:43
    romi1502 synchronize #412
  • Jun 05 10:28
    romi1502 review_requested #412
  • Jun 05 10:28
    romi1502 review_requested #412
  • Jun 05 10:28
    romi1502 review_requested #412
  • Jun 05 10:28
    romi1502 opened #412
  • Jun 05 02:02
    stewbawka commented #383
  • Jun 04 00:47
    junh1024 commented #411
  • Jun 03 13:09
    JavaShipped commented #411
  • Jun 03 13:07
    JavaShipped commented #411
  • Jun 03 11:03
    junh1024 commented #411
  • Jun 03 10:05
    JavaShipped commented #411
  • Jun 03 05:12
    junh1024 commented #411
  • Jun 03 03:00
    ez-dev commented #5
Ok, doing a fresh install seemed to work for my problem. Now the problem I'm facing is that when i try to run the 4stem feature on an mp3 of around 6-9MB in size (around 3-4 mins in length), it never completes.
This is on a machine with intel i3 and 4GB RAM. Can anyone please confirm what kind of specs are necessary to split a single mp3 file into 4 stems?
This info would be really helpful in helping me determine if I ain't up to spec, or something needs to be tweaked.
Many thanks!
2 replies
edit: when i say never comlpletes, it is 'killed' or 'dumped' after about 5 mins of attempting the conversion
as it appears to max out the memory
Hi everyone. We have released a new version of spleeter== 1.4.4 if fixes some of the bugs that have been reported and brings some functionalities that you asked.
Also using this opportunity to point again to the FAQ : that we have completed these last days
Tom Watts
@Tom-Watts_gitlab has had over 70,000 songs uploaded now (for free!), the biggest request by far has been for frequncies above 11kHz
Whoever has a model for 44kHz wields a lot of power over the universe. Do you know if anyone is working on this? Wondering because the original work with UNet architecture you used (Spotify) was lower than 11kHz, hoping the next progression is > 22kHz. Was the choice purely for efficiency, or were there other challenges?
2 replies
Rayan fernandes
im trying to deisgn a custom website
where u can seperate youtube songs by searching its name
3 replies


Hi everyone, I propose to use replies to this thread to reference external services that are using spleeter
1 reply
Hi guys I need help
I installed Anaconda, then I runned the command: conda install -c conda-forge spleeter
Then I do spleeter and it says: Fatal error in launcher: Unable to create process using '"d:\bld\spleeter_1574775894867_h_env\python.exe" "C:\Users\TheIn\Anaconda3\Scripts\spleeter.exe" separate -h ht'
1 reply
hi all
Nicholas Ngai
Hi Guys i have been using spleeter for awhile now and just now when I tried to extract vocals from a song it didnt work
i get an ffprobe error, is there any solution to this?
2 replies
Angel Hiraldo
Guillaume Vincke

Hi ! First, I want to say a big thank you to Deezer's team for publishing their work. That's great to see such a technology available. I hope it will leads many companies into doing the same.
I'm developing a C++ port of spleeter as a side project (right here: My goal is to give people the opportunity to use your technology within plug-ins.
At the moment, I've been able to use the pre-trained models with waveforms as input/output to tensorflow. My next step is to use them with an stft frame as input and a mask as output.I did so (almost) successfully but realized that performances (CPU and quality wise) are the worse. Getting into your code a bit more, I came across your 'T' parameter ( If I understand it properly, it means that the current pre-trained models are meant to process batches of 512 frames (~12sec of audio) right ?
If I'm correct, you can imagine that such a latency in a plugin isn't acceptable :).
So, my questions would be:

  • Did you already try different values for that parameter ?
  • If so, how much does it affect the result with let's say 32 or 64 frames ?
  • I guess spleeter wasn't developed with real-time constraints in mind. Do you have leads on how to get there (maybe a specific network architecture) ? Or do you think it is just not possible.

Anyway, thank you once again for your great work !

8 replies
Douglas Kastle

Hi all, this looks like an amazing tool. However I am looking for something that could cleanup audio for a podcast, so those that should like the were in a submarine.

Is there any way to just separate out clean audio, if there is no accompaniment music? And the holy grail, could it separate out two different speakers into separate channels?

I am guessing the available stems don't allow this, but if this network was trained would it be possible, if not why not. and if not, is any one else doing any source separation for spoken audio?

2 replies
Any way this is still amazing work I will play with what is there!
Hi everyone, we will soon release a new version of Spleeter with changes that may interest some of you :) Basically we noticed that using tensorflow STFT/iSTFT was very efficient on GPU but not so much on CPU, and sometimes led to a large (unnecessary) RAM consumption. So we're adding the possibility to use librosa instead for these transforms.
The PR is deezer/spleeter#282 if you want to take a look and maybe give it a try. We will bump the Spleeter version to 1.5 once we finish thoroughly testing it. All feedbacks welcome in the meantime
I'm proud to announce the first spleeter inspired music genre...''spleetercore '' or '' X-spleet '' we cant decide...
here is a link to the 1st compilation album using spleeter to generate horrifying new music...
Hi everyone, we just released version 1.5 today, with some changes on how we load the audio signal in a CPU setting. According to our tests, it could drastically reduce the memory consumption. Try it out and don't hesitate to give us your feedbacks
Also obviously, stay safe, stay home and split stuff :)
FĂ©lix Voituret
Please note that v1.5.0 is only available at the moment on PyPi, we are currently updating Conda recipe for librosa integration.
2 replies
Hey guys! What is best place, in your opinion, to host/wrap spleeter lib. GCloudFunctions? GAppEngine? VPS?
Hi guys. I'm having an incredible amount of issues with Spleeter, especially on the GPU end. Would anyone like to have a chat about it with me? I run Spleeter as a web service (not stating which) and I keep seeing my traffic increasing but I can't for the love of me satisfy the demand due to issues with Spleeter. I'd love to scale up, but I just don't see that happening right now due to the issues. Please I'm asking for a hand here, t saddens me that there's huge potential for my site to grow but my configuration not allowing me :(
The latest conda version of spleeter-gpu does not work for me at all
the old one works, but the the gpu mode is very unstable and crashes all the time
running Spleeter GPU through the command line works on the old version works, but not through my own pythin script using the api
running the old Spleeter CPU on command line processes my 3 minute song in 30 seconds, but using the API it takes 2 minutes (???)
I an very confused as to why it is like this
I have googled everywhere
I have read almost all issues in the github repo
I have reinstalled my system, my anaconda my python, my conda envs
absolutely everything
and I've tried to get it to work for 2 weeks
nothing, absolutely nothing
only old version spleeter CPU is working stable
but it's relatively slow
Alright, I gave old version of Spleeter another shot using GPU and now it works.