Forum: help

RE: questions of Gabor Dic. & ICASSP'08 paper [ Reply ]
By: Boris Mailhe on 2009-02-02 08:23

Dear Yi Zhan,

From what I get, the u parameter in this paper refers to the shift of the atom in the signal (the signal sample where the atom starts). In MPTK, it corresponds to the "filterShift" attribute. This is not really related to the atom length, rather to the signal length (which is usually much greater).

If N is the atom's length (the "filterLen" attribute in MPTK), then t should span the interval [0, N-1].

You should get better insight by directly asking the paper's authors.

Boris

questions of Gabor Dic. & ICASSP'08 paper [ Reply ]
By: Yi Zhan on 2009-02-02 01:44

[forum:101505]

Paper 'Enviromental Sound Recognition Using MP-Based features' in ICASSP'08, I have some questions.
If someone read this paper, please help me.

1: About the Gabor atoms, in the paper, as my understanding right now.
There are 3 parameters decides how many atoms in the Gabor dictionary, there are s(scale), w(Omiga, frequency), u(time) , suppose Sita(phase)=0.
s [2, 4, 8, 16, 32, 64, 128, 256]
w [1<=i<=35]
u [0, 64, 128, 192]
therefore, the dictionary size: 8*35*4=1120.
My question is here or what I am wondering is parameter 'u'. N is atom's size(256), is the N means Gabor fomula 't'? The size of 't' and 'u' can not be same?
I think, if 'u' is 256, the Gabor dictionary will be huge, am I correct?

2: take one frame(256 points) for example, 5 atoms are extracted, averaged and standard deviation of these 5 atoms' [s w u] as the feature vectors (Am I correct?)
Someone can describle the feature vectors in more detail?

Thanks!