Separation of speech from multiple speakers speaking in different microphones connected to a mixer
in [DSP]
|
Prev: Run Time Stack Usage
Next: Separation of speech from multiple speakers speaking in different microphones connected to a mixer
From: markt on 6 May 2008 15:26 >I do know that people call both the one channel multiple speaker >problem (AKA cochannel speakers problem), and the two channel >multiple speaker problem, and higher numbers of channels multiple >speaker problem, =93the cocktail party problem.=94 Which of these do you >suppose has any relevance to what happens at a cocktail party? Wait, >don=92t answer! It really doesn=92t matter in my response as you will >see. >The 3 audio sources >are indeed linearly summed but not with the same weights, so they >produce 3 different signals, not all just mixed into one signal. > Big deal, whether they are scaled or not. LINEAR means sum plus scale. This is a basic concept in linear algebra. That has no bearing on the cocktail party problem. The goal of ICA, or any linear separation problem, is to estimate the mixing matrix, scales included. Most ICA solutions have a scale ambiguity anyway (though there are ways to work around any phase ambiguity). >Because of these complications, audio separation is a >largely unsolved problem.=94 Wow! Just as I'd remembered it. So page >446 contradicts page 147 regarding this being a "practical >application" of ICA. > Um, gee, the book was written in 2001, and it is now 2008, hence my comment "dig deeper." I never said the book solved the problem. >Now you have been quite generous with your characterization of me as >=93ignorant=94. A common definition of "ignorant" is =93lacking information >or knowledge=94. Yes, clearly. You ignorantly assumed that I said the book solved the problem, and you STILL don't understand the difference between a problem statement and a potential solution, as evidenced by this silly post. You also put the book down and assumed that since it had not been solved, it could not be solved and thus "ICA is no good." Good detective work there. >Without a doubt you are =93ignorant=94 of the contents of >your own recommended reference regarding the =93cocktail party problem=94. >Apparently my memory at my almost AARP age still functions better than >your own. I fully know the contents of the book, and not once did I ever say that the book solved the problem. You did, not me. I simply provided that as a reference. > rant You're still ignorant. Mark
From: markt on 6 May 2008 15:45 >Pick any research topic and you will get a flood of hits. You'll even >find plenty that claim wonderful things. Do you know of a hit backed up >with a wave file or two that sounds impressive? No, I merely provided a starting point for investigation. The investigation itself is not my job. I did run across a dissertation yesterday while I was cleaning up my references for another work and they seem to have a solution. I did not read it since I don't really care as I typically do radar/comm. ICA really hasn't gotten much attention till the past few years. Mark
From: markt on 6 May 2008 15:58 >It takes a man to >say he's wrong and move on. Then it's no big deal. Everyone is wrong >sometimes. So, what exactly am I wrong about? Oh, that you didn't say I claimed ICA had solved the problem in the book... so what's your point about all this ranting then? Curious, since I never made any claims about it's validity. So, we're waiting, go ahead. While you're at it, look up the term "linear" with respect to mixtures. I'm still laughing at that "point" you made. Mark
From: markt on 5 May 2008 14:57 >Dig deeper into what? RADC (not the current name) was trying to solve >this problem in the 1980's-early 90's. I followed their efforts at >that time. Their efforts up through that period were ultimately >admitted to not perform well. The last solution that I am aware of >threw out most of the processing of previous solutions, because a much >simpler method produced more useful (not good) results. I gave you two references, both of which are more current than the 80s/90s. Do a simple Google search on "cocktail party problem" and ICA and you'll get a bevy of hits. >I would expect that slow convergence of ICA would be a problem because >the acoustic environment (head orientation, etc) often changes pretty >quickly. I've implemented ICA on Rayleigh fading channels for a DS-CDMA system not unlike that used in US PCS systems. It is not the end-all, yet, but there's a lot of research in this area. Do the Google search I mention above and the folks at Helsinki U of T are the first that pop up (not coincidentally, Hyvarinen is the author of the book). >BTW what is being described here is NOT the well known "Cocktail >Party" problem. That problem has to do with two microphones (or ears) >that are not mixed together. Also, most people don't realize how the >actual "Cocktail Party" problem differs from what actually happens at >a cocktail party. If you are interested in the problem you can figure >out the huge difference. I disagree. The cocktail party problem as typically devised is simply the mixing of multiple speakers in a room. Using multiple receivers, e.g. two ears, is merely one method to separate the signals (by exploiting phase differences which provides position information). How they get mixed is really immaterial (well, assuming some linear summation), what matters is their independence. Try not to be so ignorant in your replies and maybe others will be helpful, too. Remember, YOU are asking for help, not me. Mark
From: markt on 5 May 2008 17:47
>You are a jackass. The first post wasn't mine. Pay attention to >details. You are correct that the first post was not yours, and for that I apologize. However, you did not attach any date to what you researched, other than "i read a book a couple of years ago" and left it at that, not even a title. At least I mentioned a title for the book I refer to. It is rather recent and the author is heavily active in the area. You further made statements clearly indicating that you don't understand the problem statement to begin with, while getting ignorant with me for no apparent reason other than some author said ICA doesn't work. Well, a lot of others certainly disagree, and I'm more willing to believe them than some yoyo that doesn't even know what the cocktail party problem is. I'm not the jackass here. I was merely trying to be helpful on a subject that I have been studying rather intensely of late. Randy, plenty of ICA apps abound. The Hyvarinen book actually does a few DS-CDMA implementations (as well as other separation problems). One is a channel estimation solution, and another, using FastICA (by Bingham and Hyvarinen) is an actual symbol detection solution. Some good authors for research are Cichocki, Tsatsanis, Wang and Poor and one of the original blind detection authors, Verdu. A starting point for some recent work I've done on this subject was this: K.-H. Yap, L. Guan and J. Evans, “Blind adaptive detection for CDMA systems based on regularized independent component analysis,” Proc. of IEEE Global Telecommunications Conference, San Antonio, USA, vol. 1, pp. 249-253, Nov. 2001. This paper uses the actual spreading sequences, which differs from the fully-blind methods employing Lagrange multipliers (Tsatsanis has done several detectors in this area). There are many new authors taking on the task very recently (not just the big ones over at Helsinki) including Tulay Adali more recently. She is the technical chair at the IEEE International Workshop on Machine Learning for Signal Processing. There should be some good papers coming out of this workshop. If you'd like more info, send me an email. I'll be more than happy to point you towards some of the research I've done. Mark |