I checked out a method for spectral estimation that estimates the spectrum of speech from degraded speech (speech+noise) input which involves autocorrelation. But i fail to understand how autocorrelation would achieve this. Wasn't autocorrelation meant for checking out the similarities between two points in the same signal right?
Yes, similarities between two points in the same signal. The case is that any two points of noise do not have similarities at all (in spite of how close they are to each other), but two points of speech do. This is the grounds for implementing this method. Unfortunately I dont know details...
Sure. Do your words sound like noise? signal and noise are uncorrelated with each other. Every sound of your alphabet takes some time. Along its duration all possible sound samples are correlated with each other, but the samples of noise, even adjacent, are not.
hey thank you for your answer i am now able to relate autocorrelation with spectral estimation. I am kind of new to speech processing techniques so sorry in advance for my naive question in advance but by any chance do we have anything as correlated noise? I ask this because some of the technical papers i referred for spectral estimation specify that this method is only for uncorrelated noise. this specification means that we have correlated noise as well. can you please specify anything on correlated noise i fail to find any good material for beginners on this topic. thanks!