Where communities thrive

  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
  • 02:58
    aidv commented #276
  • 02:58
    aidv commented #276
  • 02:00
    junh1024 commented #276
  • 01:28
    junh1024 commented #276
  • 01:25
    junh1024 commented #276
  • Apr 08 22:42
    aidv commented #276
  • Apr 08 22:37
    duhaime commented #279
  • Apr 08 22:29
    spereree commented #276
  • Apr 08 22:14
    aidv commented #276
  • Apr 08 22:09
    spereree commented #276
  • Apr 08 14:52
    Cathon commented #306
  • Apr 08 01:58
    DB2799 commented #279
  • Apr 07 17:27
    shikharsingla commented #279
  • Apr 07 16:42
    patcheszone commented #306
  • Apr 07 12:47
    sarthak405 commented #279
  • Apr 07 06:59
    VivianCheng228 commented #279
  • Apr 06 16:04
    aidv commented #316
  • Apr 06 15:26
    aidv commented #276
  • Apr 06 13:55
    githublyff labeled #317
  • Apr 06 13:55
    githublyff opened #317
Geraldo Ramos
that's cool @mmoussallam you're leading the development of spleeter?
Not really, it's been developed by the research team over the last couple of years
and packaged recently for the open-sourcing
i'm still having the issue with ffprobe
Hello, great tools but have also the ffprobe error...someone that make it work can make a tutorial on how he installed everything including python? should help people with the fucking ffprobe error, or better if a nice guy could make it "gui"
Hi everyone. Just letting you know I built this free service https://acapella-extractor.com/ (to extract acapellas, obviously) using spleeter. Thank you to the developers for your work and making it open source :)
for thoses having ffprobe error i think it's a folder permission things cause if you move the files into user folder C:\Users\xxxxx for me only by moving it and use the command spleeter separate -i test.flac -p spleeter:2stems -o test worked
if you set a path it often don't work
maybe it's something related to security etc
if i correctly understood it can again improve by the ai training?
@mmoussallam is there a way to remove the 10mn limitation? it seems not doing anything on audios longer than 10mn
new issue, i have countless of thoses lines, most of the time even with thoses it work, sometimes i only have vocals sometimes only instrumental, i tried a 20mn audio after settlings to 9000 but it overload cpu and there's no output after loading audios
Tom Watts
A bit late to the party: Thanks to everyone at Deezer who made this! Really excited for the future, in this research area
Forgot to say: i'm the maker of melody.ml the first website (there are around 5 websites serving the model now!)
That bug, with ffmpeg caused me a few sleepless nights with servers hanging, thanks for fixing it
Tom Watts
How big was the training dataset? In the paper you say around 22,000 songs, so if they're wavs roughly: 50mb 4 22,000, roughly 4 terrabytes of data? That's incredible, must have been quite an engineering challenge
Hello sorry to bother again but i have a question about the stems provided: are they final versions or they can be improved and uploaded later? cause even if it's the better tools i ever used for that purpose, sometimes it don't perfectly work like for this youtube video https://www.youtube.com/watch?v=KflJJdfFC8g you have instrumental with guitars behind and it was voiced other for the dub, but the tool don't perfectly keep everything except vocals (i want everything except vocals)
Hi there, can anybody help?
Problem I'm having using the audio_example.mp3 file, it seems to process successfully, but when i go to the out out folder and play the 2 stems there, they both are not seperated. The one named voice plays the whole song, and so does the other!
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!
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 : https://github.com/deezer/spleeter/wiki/5.-FAQ that we have completed these last days
Tom Watts
melody.ml 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?
Rayan fernandes
im trying to deisgn a custom melody.ml website
where u can seperate youtube songs by searching its name


Hi everyone, I propose to use replies to this thread to reference external services that are using spleeter
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'
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?
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: https://github.com/gvne/spleeterpp). 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 (https://github.com/deezer/spleeter/wiki/3.-Models#audio-parameters). 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 !

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?

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... https://spleetercore.bandcamp.com/releases
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.
Hey guys! What is best place, in your opinion, to host/wrap spleeter lib. GCloudFunctions? GAppEngine? VPS?