Separate the voice and music

Started by karthikbalaguru December 6, 2009
Hi,
Is there any opensource software that
could separate the music and voice
into two separate files ?
Required for one of the experiments.

Thx in advans,
Karthik Balaguru
On Dec 7, 6:33&#2013266080;am, karthikbalaguru <karthikbalagur...@gmail.com>
wrote:
> Hi, > Is there any opensource software that > could separate the music and voice > into two separate files ? > Required for one of the experiments. > > Thx in advans, > Karthik Balaguru
Phew - if there was I would be the first to sing its virtues! I assume you mean something like a CD recording of a singer and you wish to separate the music from the singer? There a a few approaches (I dount anything works anything near perfectly). Blins source separation or so-called unsupervised adaptive filtering. The method to Google is Independent component Analysis. (ICA). If you get lucky you might even try ordinary adaptive filtering using the 2 channels from the stereo recording. try subtracting the channels! I have no idea if there is open-source code on this. To understand it it is best to write your own. I would doubt if you could at this time have success but you never know, there just might be an ICA or adaptive filtering method out there that nearly does the job. Hardy
On Dec 7, 1:12&#2013266080;pm, HardySpicer <gyansor...@gmail.com> wrote:
> On Dec 7, 6:33&#2013266080;am, karthikbalaguru <karthikbalagur...@gmail.com> > wrote: > > > Hi, > > Is there any opensource software that > > could separate the music and voice > > into two separate files ? > > Required for one of the experiments. > > > Thx in advans, > > Karthik Balaguru > > Phew - if there was I would be the first to sing its virtues! I assume > you mean something like a CD recording of a singer and you wish to > separate the music from the singer? > There a a few approaches (I dount anything works anything near > perfectly). Blins source separation or so-called unsupervised adaptive > filtering. The method to Google is Independent component Analysis. > (ICA). If you get lucky you might even try ordinary adaptive filtering > using the 2 channels from the stereo recording. try subtracting the > channels! &#2013266080;I have no idea if there is open-source code on this. To > understand it it is best to write your own. I would doubt if you could > at this time have success but you never know, there just might be an > ICA or adaptive filtering method out there that nearly does the job. > > Hardy
Sorry I meant Blind Source Separation.
>On Dec 7, 1:12=A0pm, HardySpicer <gyansor...@gmail.com> wrote: >> On Dec 7, 6:33=A0am, karthikbalaguru <karthikbalagur...@gmail.com> >> wrote: >> >> > Hi, >> > Is there any opensource software that >> > could separate the music and voice >> > into two separate files ? >> > Required for one of the experiments. >> >> > Thx in advans, >> > Karthik Balaguru >> >> Phew - if there was I would be the first to sing its virtues! I assume >> you mean something like a CD recording of a singer and you wish to >> separate the music from the singer? >> There a a few approaches (I dount anything works anything near >> perfectly). Blins source separation or so-called unsupervised adaptive >> filtering. The method to Google is Independent component Analysis. >> (ICA). If you get lucky you might even try ordinary adaptive filtering >> using the 2 channels from the stereo recording. try subtracting the >> channels! =A0I have no idea if there is open-source code on this. To >> understand it it is best to write your own. I would doubt if you could >> at this time have success but you never know, there just might be an >> ICA or adaptive filtering method out there that nearly does the job. >> >> Hardy > >Sorry I meant Blind Source Separation. >
Just wondering Hardy, can it be done by simple high/low pass filtering into two channels? Hip hop may be difficult, but something like country can be separated?

HardySpicer wrote:

> On Dec 7, 1:12 pm, HardySpicer <gyansor...@gmail.com> wrote: > >>On Dec 7, 6:33 am, karthikbalaguru <karthikbalagur...@gmail.com> >>wrote: >> >> >>>Hi, >>>Is there any opensource software that >>>could separate the music and voice >>>into two separate files ? >>>Required for one of the experiments.
1. Convert voice + music into text and midi. 2. Then text to speech and midi to music.
>>Phew - if there was I would be the first to sing its virtues! I assume >>you mean something like a CD recording of a singer and you wish to >>separate the music from the singer? >>There a a few approaches (I dount anything works anything near >>perfectly).
IIRC Compandent offered reasonably working software for cleaning speech from the background. However it worked well only if the initial SNR was sufficiently high to begin with. VLV
On Dec 8, 1:07&#2013266080;am, "commengr" <communications_engin...@yahoo.com>
wrote:
> >On Dec 7, 1:12=A0pm, HardySpicer <gyansor...@gmail.com> wrote: > >> On Dec 7, 6:33=A0am, karthikbalaguru <karthikbalagur...@gmail.com> > >> wrote: > > >> > Hi, > >> > Is there any opensource software that > >> > could separate the music and voice > >> > into two separate files ? > >> > Required for one of the experiments. > > >> > Thx in advans, > >> > Karthik Balaguru > > >> Phew - if there was I would be the first to sing its virtues! I assume > >> you mean something like a CD recording of a singer and you wish to > >> separate the music from the singer? > >> There a a few approaches (I dount anything works anything near > >> perfectly). Blins source separation or so-called unsupervised adaptive > >> filtering. The method to Google is Independent component Analysis. > >> (ICA). If you get lucky you might even try ordinary adaptive filtering > >> using the 2 channels from the stereo recording. try subtracting the > >> channels! =A0I have no idea if there is open-source code on this. To > >> understand it it is best to write your own. I would doubt if you could > >> at this time have success but you never know, there just might be an > >> ICA or adaptive filtering method out there that nearly does the job. > > >> Hardy > > >Sorry I meant Blind Source Separation. > > Just wondering Hardy, can it be done by simple high/low pass filtering > into two channels? Hip hop may be difficult, but something like country can > be separated?
Good luck with that! The trouble is that this is a general problem of filtering of non-stationary signals from each other whose spectra overlap. It ain't easy but I expect in some cases you can get lucky.