by

Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
  • 06:19
    amo13 commented #437
  • 01:17
    geraldoramos commented #437
  • Jul 01 16:24
    amo13 commented #437
  • Jul 01 16:22
    amo13 commented #437
  • Jul 01 16:19
    romi1502 commented #437
  • Jul 01 14:04
    geraldoramos edited #437
  • Jul 01 13:55
    geraldoramos edited #437
  • Jul 01 13:54
    geraldoramos edited #437
  • Jul 01 13:53
    geraldoramos edited #437
  • Jul 01 13:47
    geraldoramos commented #391
  • Jul 01 13:46
    geraldoramos labeled #437
  • Jul 01 13:46
    geraldoramos labeled #437
  • Jul 01 13:46
    geraldoramos opened #437
  • Jul 01 12:52
    geraldoramos commented #391
  • Jul 01 10:43
    amo13 commented #391
  • Jul 01 10:41
    amo13 commented #391
  • Jul 01 05:37
    geraldoramos commented #391
  • Jul 01 05:21
    geraldoramos commented #391
  • Jul 01 05:01
    geraldoramos commented #391
  • Jul 01 04:08
    LaoTsing commented #279
Johnny
@Johnny25232_twitter
Hi all
Geraldo Ramos
@geraldoramos
Hi there!
I've received this ticket, anyone has any thoughts? I believe this is a Spleeter thing since we at moises.ai don't do any post-processing.
Unfortunally, the spectral frequences been cuted at 11kHz after processing Is that problem with Spleeter or maybe there is bug in your servise?
image.png
Moussallam
@mmoussallam
Hi Geraldo, yes cutting the high frequencies happens during the processing in spleeter
There's a parameter to change this output behavior: https://github.com/deezer/spleeter/wiki/3.-Models#separation-parameters
Geraldo Ramos
@geraldoramos
very nice, will incorporate that as an option. Thank you so much for the fast response.
Moussallam
@mmoussallam
but the models have not been trained to predict these frequencies (it's actually quite hard to do) so the results may be not so good
Geraldo Ramos
@geraldoramos
I see, will give some tests and if results are reasonable may be worth it making it configurable.
Moussallam
@mmoussallam
suit yourself :)
Geraldo Ramos
@geraldoramos
o/
eddiegueiros
@eddiegueiros
that's cool @mmoussallam you're leading the development of spleeter?
Moussallam
@mmoussallam
Not really, it's been developed by the research team over the last couple of years
and packaged recently for the open-sourcing
Johnny
@Johnny25232_twitter
i'm still having the issue with ffprobe
1 reply
Mickiki
@Mickiki4_twitter
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"
test_account
@test_account_ac_twitter
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 :)
Mickiki
@Mickiki4_twitter
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
Mickiki
@Mickiki4_twitter
if i correctly understood it can again improve by the ai training?
Mickiki
@Mickiki4_twitter
@mmoussallam is there a way to remove the 10mn limitation? it seems not doing anything on audios longer than 10mn
3 replies
Mickiki
@Mickiki4_twitter
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
@Tom-Watts_gitlab
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
@Tom-Watts_gitlab
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
2 replies
Mickiki
@Mickiki4_twitter
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)
2 replies
bespokecomponents
@bespokecomponents
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!