How to Use Fourier Transforms to Manipulate Sound Waves?

[dreamsfly]

It's a tech.,and it is said that that had been done by MIT,then anybody know the details?

 

[FredGarvin]

You'll have to be more specific when you say "separate". The first thing I thought of was simply getting a frequency spectrum which is hardly a rare thing to do. Can you be a bit more detailed in your description? How about an article or some kind of link to what you are referring?

 

[dreamsfly]

ok,I'll make it clear here:for example,we can combine two sound waves into one compositive wave,then if we know the compositive wave,how to separate it into original wave?
I believe it can be true,just as our ears can do.

 

[Danger]

I'm a bit confused by your use of the singular 'wave'. Anything other than an absolutely pure note consists of many different waves. Separating them is usually a matter of applying frequency filters to isolate one particular wavelength from the others. That's how an audio lab will approach a problem of, for instance, isolating voices from the background on a surveillance tape.

 

[dreamsfly]

Oh,I think I　missed -s,sorry!
I'm interested in frequency filters ,would you please speak more about that?

 

[Danger]

I'm afraid that I don't know much about them. Essentially, though, they're circuits that will only pass particular frequencies while blocking the rest. Say you've got an oscilliscope in the circuit. You'll see a bunch of different traces, corresponding to the different sound frequencies. By applying various filters, you can eliminate them selectively until there's only one sine on the screen. It's pretty much the same as the graphic equilizer on a stereo.

 

[rbj]

is your question about how do filters work? or is it about how humans perceive different waveforms (of different fundamental frequency) that are mixed together?

 

[pallidin]

Ok, I don't know how MIT did it, but the concept is fairly simple:

Take a metal rod of some arbitrary dimension, and weld "tunning forks" of different reasonant responses to the rod(the tunning rods are separated from each other, of course)
Applying an "accoustical noise" to one end of the rod will cause the varied frequencies to be "expressed" through the tunning forks, which will effect a non-perfect "filtering"

Ok, that's a gross example, but illustrates frequency separation through the application of reasonant-response materials

 

[FredGarvin]

It would definitely helpif the OP would explain a bit more about what they have seen from MIT. Was it a demonstration of some kind? Honestly, it simply sounds like a spectrum analyzer technique. If that is the case, you may start by looking into the fast Fourier transform and it's applications in sound and vibrations.

 

[-Job-]

I don't know how "simple" the task is. If we are adding the two waves together, then some ambiguity is generated. For example, if we add 2 and 8 to obtain 10, from 10 we can't determine if the original addends were 1 and 9, 2 and 8, 3 and 7, 4 and 6, or 5 and 5. With waves, we are adding a sequence of numbers, each following a pattern, into a single sequence. By analyzing the composed sequence we can approximate the original waves, but it might well still be ambiguous. This task would be much easier if the waves had prime values and we were multiplying rather than adding.
In any case it seems that there would be scenarios where the composed wave might not offer enough information to determine the parent waves, making the task impossible, not to mention time-intensive, from a computer perspective. Too much noise.

 

[pallidin]

I found some info here. It references MIT, so maybe it can be of help:

http://citeseer.ist.psu.edu/context/104563/0

 

[dreamsfly]

Thank you!

 

[dephys]

Tell exactly what u want to know

 

[dimensionless]

I think what your looking for is "blind source separation." You can search for it, or check the paragraph Wikipedia has on it. Basically there are two sources of sound and they are recorded with two microphones. Everything else, such as location of the source, is unknown. In blind source separation one source is isolated from the other.

 

[dreamsfly]

I think what your looking for is "blind source separation." You can search for it, or check the paragraph Wikipedia has on it. Basically there are two sources of sound and they are recorded with two microphones. Everything else, such as location of the source, is unknown. In blind source separation one source is isolated from the other.

Assume they are not recorded separately,there's only a sound-background,just kick out a certain sound,which can be used in mobilephone:Although you speak in publical place with loud noise,but the listener can only hear your voive,that's the aim.

 

[dreamsfly]

Ok, I don't know how MIT did it, but the concept is fairly simple:

Take a metal rod of some arbitrary dimension, and weld "tunning forks" of different reasonant responses to the rod(the tunning rods are separated from each other, of course)
Applying an "accoustical noise" to one end of the rod will cause the varied frequencies to be "expressed" through the tunning forks, which will effect a non-perfect "filtering"

Ok, that's a gross example, but illustrates frequency separation through the application of reasonant-response materials

I have ever thought about this method,but can it be separated two mixed sound by resonance?

 

[dreamsfly]

Ok, I don't know how MIT did it, but the concept is fairly simple:

Take a metal rod of some arbitrary dimension, and weld "tunning forks" of different reasonant responses to the rod(the tunning rods are separated from each other, of course)
Applying an "accoustical noise" to one end of the rod will cause the varied frequencies to be "expressed" through the tunning forks, which will effect a non-perfect "filtering"

Ok, that's a gross example, but illustrates frequency separation through the application of reasonant-response materials

I have ever thought about this method,but can it be separated two mixed sounds by resonance?

 

[dreamsfly]

I don't know how "simple" the task is. If we are adding the two waves together, then some ambiguity is generated. For example, if we add 2 and 8 to obtain 10, from 10 we can't determine if the original addends were 1 and 9, 2 and 8, 3 and 7, 4 and 6, or 5 and 5. With waves, we are adding a sequence of numbers, each following a pattern, into a single sequence. By analyzing the composed sequence we can approximate the original waves, but it might well still be ambiguous. This task would be much easier if the waves had prime values and we were multiplying rather than adding.
In any case it seems that there would be scenarios where the composed wave might not offer enough information to determine the parent waves, making the task impossible, not to mention time-intensive, from a computer perspective. Too much noise.

That's just what I care about.

 

[dimensionless]

Assume they are not recorded separately,there's only a sound-background,just kick out a certain sound,which can be used in mobilephone:Although you speak in publical place with loud noise,but the listener can only hear your voive,that's the aim.

I'm not sure if I was clear or not.

Basically you can have two people speaking in the same room at the same time. Blind source separation processes a stereo recording of the two talkers with computer algorithms. In doing so, one of the talkers voices is removed from the recording while the other is left intact.

 

[ObsessiveMathsFreak]

Get the Fourier Transform of the sound signal. This tells you how strong each paticular frequency is in the signal. Or I suspect what you really want is the Fourier Series of the signal. This expresses the signal as a sum, or superposition if you like, of basic signals like sine and cosine.

The series is what you want, but the transform is useful too. Once you have it, you can tweak it a little bit, say reduce the strength of higher frequenices, then perform an inverse Fourier transform to get back a modified signal. If you want to see how this works in practice, just open up any modern music player and fiddle with those hertz settings. My understanding is that this is using Fourier transforms in some way.