Im trying to find software that allows me to import a song from a band and have that sound broken down into all its different components (like in CSI when they have all the different sound waves and can listen to each one individually).. not sure if this is possible so any suggestions would be helpful
sound isolation
Started by ●January 17, 2008
Reply by ●January 17, 20082008-01-17
jrsmith5601 wrote:> Im trying to find software that allows me to import a song from a band and > have that sound broken down into all its different components (like in CSI > when they have all the different sound waves and can listen to each one > individually).. not sure if this is possible so any suggestions would be > helpfulPretty much, that happens only in the movies. You can script it, but you can't do it. It makes for good stories, though. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●January 17, 20082008-01-17
On 17 Jan., 23:49, "jrsmith5601" <jrsmith5...@gmail.com> wrote:> Im trying to find software that allows me to import a song from a band and > have that sound broken down into all its different components (like in CSI > when they have all the different sound waves and can listen to each one > individually).. not sure if this is possible so any suggestions would be > helpfulSorry thats not possible, but I can tell you how to design a laser sword ... no, wait ! I can think of at least 2 ways of achieving this. The trivial way is to extract different sources by frequency, although of limited use because of the overlapping spectra. Another intersting way would be to put some additional knowledge in the extraction of single sources in the style of a sound generator model. Let's explain this for a recording which has the sound of a refrigerator interfering. You already know that it is a refrigerator, so it should be possible to build a model generating the same sound as the refrigerator in the recording (including the impulse response for the way this sound travels). This model should be guided by the recording to adapt and track variations to generate a very close match of the sound of the refrigerator. This is the hard part of the work (to do this model until it closely matches and tracks). Maybe you separate the impulse response the sound experiences on it's travel or include this in your refrigerator sound model. And then you simply subtract the sound of your refrigerator model in time domain to have it perfectly removed. And you can repeat this for other interferers. At some point additional Information given by the human operator of sound extraction gets essential. You could also think of doing this only in fequency domain. Or you could combine this with spectral separation. Good luck, and share your success story here.
Reply by ●January 17, 20082008-01-17
Jerry Avins wrote:> jrsmith5601 wrote: > >> Im trying to find software that allows me to import a >> song from a band and have that sound broken >> down into all its different components (like in >> CSI when they have all the different sound waves and can>> listen to each one individually).. not sure if this is>> possible so any suggestions would be helpful> Pretty much, that happens only in the movies. You can script > it, but you can't do it. It makes for good stories, though.I agree, but computers are getting faster all the time. It should be possible to do some things that couldn't be done not so long ago. One that I have wondered about is using phase and amplitude differences in stereo signals to detect and separate out parts. Or maybe correlation of harmonics in signals (especially musical ones) that have significant amounts of them. Though I suppose the easiest, and as far as I know still hard, is removing 60Hz hum, including all the harmonics. -- glen
Reply by ●January 17, 20082008-01-17
filter001@desinformation.de wrote:> On 17 Jan., 23:49, "jrsmith5601" <jrsmith5...@gmail.com> wrote: >> Im trying to find software that allows me to import a song from a band and >> have that sound broken down into all its different components (like in CSI >> when they have all the different sound waves and can listen to each one >> individually).. not sure if this is possible so any suggestions would be >> helpful > > Sorry thats not possible, but I can tell you how to design a laser > sword ... no, wait ! > > I can think of at least 2 ways of achieving this. > > The trivial way is to extract different sources by frequency, although > of limited use because of the overlapping spectra. > > Another intersting way would be to put some additional knowledge in > the extraction of single sources in the style of a sound generator > model. > Let's explain this for a recording which has the sound of a > refrigerator interfering. > You already know that it is a refrigerator, so it should be possible > to build a model generating the same sound as the refrigerator in the > recording (including the impulse response for the way this sound > travels). > This model should be guided by the recording to adapt and track > variations to generate a very close match of the sound of the > refrigerator. This is the hard part of the work (to do this model > until it closely matches and tracks). Maybe you separate the impulse > response the sound experiences on it's travel or include this in your > refrigerator sound model. > And then you simply subtract the sound of your refrigerator model in > time domain to have it perfectly removed. > And you can repeat this for other interferers. > At some point additional Information given by the human operator of > sound extraction gets essential. > > You could also think of doing this only in fequency domain. > Or you could combine this with spectral separation. > > Good luck, and share your success story here.If you've heard one refrigerator, you've heard them all? I don't think so. It might be done by successive approximation, but you need criteria for generating the next step, and a procedure that will converge. Maybe with a massively parallel quantum computer? Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●January 17, 20082008-01-17
glen herrmannsfeldt wrote:> One that I have wondered about is using phase and amplitude > differences in stereo signals to detect and separate out parts.Despite of the popular belief, there is generally not too much of the correlation between the left and the right channels.> Or maybe correlation of harmonics in signals (especially musical > ones) that have significant amounts of them.Every time you take the same note on a musical instrument, the waveform has some random variations. Modern synthesizers do that, too. It improves the naturalness.> Though I suppose the easiest, and as far as I know still hard, > is removing 60Hz hum, including all the harmonics.The problem with the power hum is that it is not just 60Hz and its harmonics. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Reply by ●January 18, 20082008-01-18
On 17 Jan, 23:49, "jrsmith5601" <jrsmith5...@gmail.com> wrote:> Im trying to find software that allows me to import a song from a band and > have that sound broken down into all its different components (like in CSI > when they have all the different sound waves and can listen to each one > individually).. not sure if this is possible so any suggestions would be > helpfulWhat do you mean by "component"? Somethind similar to isolating individual instruments in the band? Can't be done by means of DSP. It's as simple as that. Why? Well, it is said that parent penguins can locate their offspring by the sound alone, in a colony comprising tens of thousands of penguins where all the penguins call for parents or offspring at the same time. What does the story tell us about the penguins? It tells that they have an auditory system which is perfectly tuned to the task of finding their relatives amongst tens of thousands of their kind. Which, BTW, makes perfect sense, from a survival point of view. What does that tell us about the noises the penguins make? Nothing at all. The noise they make is so completely into grips with ther auditory system that one makes no sense without the other. Asking a person to automize a method to isolate one particular voice in a crowd or instrument in a band is similar to as somebody to find one particular penguin squeak in the colony. We, as humans, think the former is easy because we can do that in our minds. When faced with the penguin problem we might adopt a different view, for the sole reason that we have no personal experience with the problem and can evaluate the problem from a more balanced point of view. Rune
Reply by ●January 18, 20082008-01-18
> Im trying to find software that allows me to import a song from a band and > have that sound broken down into all its different components (like in CSI > when they have all the different sound waves and can listen to each one > individually).. not sure if this is possible so any suggestions would be > helpfulThe bass guitar and the hi-hats/cymbals will be fairly isolatable because of the frequency locations. But don't expect any miracles. If you have some software with a multi-band equaliser, try cutting everything but a narrow band and sweep this up and down the spectrum - you'll soon get a good idea of what's possible. In the early days of stereo, instruments were often panned to the extreme left and right, making them easy to sample out these days! : )
Reply by ●January 18, 20082008-01-18
"Rune Allnor" schrieb> > Im trying to find software that allows me to import a song > > from a band and have that sound broken down into all its > > different components (like in CSI ... > > Can't be done by means of DSP. It's as simple as that. Why? > Well, it is said that parent penguins can locate their > offspring by the sound alone, in a colony comprising tens > of thousands of penguins where all the penguins call for > parents or offspring at the same time. >Even when there were dozens of kids on the playground, I could tell when mine was whining. Does that qualify me as a penguin? I think that a parent's neural networks are trained to recognize their offspring even under the harsh environmental conditions of a playground :-) But then, recognizing is different from and IMHO much easier than separating. Separating might work to some extent if you do a parametric estimation. Just my 2 cents. Maybe even less. Martin
Reply by ●January 18, 20082008-01-18
