I'm probably interpreting what he said in too simplistic a manner: That is, a number of different sound sources, in blue, could have been mixed to produce the same result, in red.
The point is made when one looks at our best efforts in signal processing to analyse even something as specific as speech in noise - it's not 100% successful.
No programs are able to extract & follow a conversation in a noisy room full of different conversations which we can do. This is the so-called "Cocktail Party Effect"
No programs are able to extract & follow a conversation in a noisy room full of different conversations which we can do. This is the so-called "Cocktail Party Effect"
Last edited:
The blue sources are two of the various conversations happening concurrently - what's arriving at the eardrum is the combination of these two sound sources simply mixed (shown in red) as vibrations that move the eardrum, generating nerve impulses.
Those nerve impulses are just a string of electrical signals which correlate to the eardrum's vibration.
Auditory processing's job is to separate out source1 from the rest of this mix of electrical nerve signals so that we can follow the conversation from source1
You're so sure of yourself that you see no possibility that you could be wrong, do you? And with the feverish surety of a zealot you castigate any & all that I say.
At one level it's simply stating the obvious, take any 10 second sound clip at 44.1KHz and you can decompose it into a sum of two streams of 441,000 numbers (essentially infinite choices) most of them taken individually are just noise/garbage.
I'm afraid that this is a completely incorrect analogy to what is being said and what's in the video - it has nothing to do with digital audio - it's to do with the makeup of auditory scenes in the real world & the issue auditory perception faces in making a realistic & reasonably accurate internal representation of this auditory scene.
I see from the rest of his talk (I wish he didn't sound so bored!) that he's setting the scene 😉 for the explanation of how we separate direct sound from reflections, no problems there 🙂
Sorry, given any phenomena that is a linear sum of two others you have no a priori knowledge of anything but the result. It has nothing to do with digital. People hear voices in totally random noise (don't tell me you haven't listened for them).
I'm not missing his point - he is just wrong, I'm afraid. A well-posed problem have the properties that:
- a solution exists,
- the solution is unique,
- the solution's behavior changes continuously with the initial conditions.
An ill-posed problem does NOT have an infinite number of solutions.
When we are talking about perception, there are not infinite possibilities - how fast do you think the brain can process data?
Sorry but I find his post is just confusing matters
Last edited:
But you do have a priori knowledge - knowledge of how sound behaves in the real world & this is used in auditory processing analysis & greatly limits the possible solutions.
Keep in mind that we are talking about perception - a biological process, not a mathematical abstraction
In fact it does have virtually infinite local minima in the mathematical sense. Come on this guy is MIT I know what he means. He starts out with the typical abstraction of this problem and proceeds to show how it's different, as simply a waveform on a scope you can come up with (virtually) infinite choices of two waveforms that add up to the one you have. So move on.
Last edited:
Cleared them up for me, isn't language great? 😉
Anyway, the rest of the talk is hunky dory
This is a joke right? "This guy is from MIT I know what he means"
You are simply confusing everything hes says & proving by your posts that you don't understand what is being said
I gave you an example of how we have a priori knowledge about the problem space which you stated we didn't. Josh Mcdermott is a well known researcher in the area & I know what he says - never does he say that there are an infinite number of possible solutions - he goes on to say at 3:00 that assumptions have to be used to solve these problems & we know that "real world sounds have some degree of regularity"
So, again, I'm afraid that you are wrong here so please move on, you're only confusing matters
Last edited:
- Status
- Not open for further replies.