Forum: help

RE: Building Dictionaries [ Reply ] By: Rémi Gribonval on 2008-11-07 19:51	[forum:98889]
What I was referring to is the file MPTK-Source-0.5.6/src/matlab/experimental/doc/readme-Development.txt from the archive you downloaded and uncompressed from our site. Remi.

RE: Building Dictionaries [ Reply ] By: chris layman on 2008-11-07 18:51	[forum:98888]
Hey Remi I don't see that I have an 'experimental' folder in /scr/matlab Does this have to do with me turning off the BUILD_MATLAB_MEX_FILES_EXPERIMENTAL when I did the ccmake? -chris

RE: Building Dictionaries [ Reply ]
By: Rémi Gribonval on 2008-11-07 18:28

[forum:98887]

Hi Chris,

The structure of the index and atom fields is described in the (quite hidden, I must say) file

src/matlab/experimental/doc/readme-Development.txt

We are not advertising it too much since the structure is probably going to change in future versions to be more user friendly, but this is the best we could figure out so far.

You can have a look at the script bookedit to see how this is used for, e.g., attempting some pitch-shifting.

Regards,

Remi.

RE: Building Dictionaries [ Reply ] By: chris layman on 2008-10-30 19:54	[forum:98844]
If anybody has had any luck with plotting the individual atoms in Matlab, in the order in which they are selected by MPTK, could you help me? For the life of me I can't figure out how to do this seemingly simple task. thanks -chris

RE: Building Dictionaries [ Reply ] By: chris layman on 2008-10-28 19:24	[forum:98829]
--bump--

RE: Building Dictionaries [ Reply ] By: chris layman on 2008-10-11 00:50	[forum:98657]
i have tried something like this book1 = book; book1.atom = book1.atom(1) or [book1.atom(1)] or {book1.atom(1)} all these syntaxes gives errors like this when: >> bookplot(book1) ??? Index exceeds matrix dimensions. Error in ==> bookplot>extract_atom_from_book at 178 atom.type = book.atom(typeindex).type; Error in ==> bookplot at 85 atom = extract_atom_from_book(book,i);

RE: Building Dictionaries [ Reply ]
By: chris layman on 2008-10-10 01:42

[forum:98645]

Thanks,

question: when i run mpdecomp in matlab it returns a data structure like this one:

format: '0.1'
numAtoms: 10
numChans: 1
numSamples: 152756
sampleRate: 610
index: [5x10 double]
atom: [1x3 struct]

i am wondering what the values mean under the 'index'.

I can't seem to find out how to plot a group of atoms. Say if i decompose a signal into 100 atoms. I can see them grouped in their blocks under book.atom.params , but i want to plot say atoms 1-10 in the oder that they are picked out by MPTK.

-chris

RE: Building Dictionaries [ Reply ] By: Rémi Gribonval on 2008-10-09 07:47	[forum:98631]
No, you can choose any integer size. Remi.

RE: Building Dictionaries [ Reply ] By: chris layman on 2008-10-09 07:24	[forum:98630]
hi and thank you is it required that the scale be a power of 2? -chris

RE: Building Dictionaries [ Reply ]
By: Rémi Gribonval on 2008-10-09 07:08

[forum:98629]

Yes, in a sense: if you specify a very large dictionary containing atoms at many different scales, then MP will automatically pick up the one which are most adapted to the analyzed signal. For example, you can use a dictionary which contains scales 2,4,8,... This will make the computation time significantly longer, but may help you determine which scales are relevant for your signals.

Best regards,

Remi.

RE: Building Dictionaries [ Reply ] By: chris layman on 2008-10-08 23:53	[forum:98628]
Thanks, I will try that. Question though, is it possible to include the scale of the atom as a parameter to be optimized like the other parameters such as amp, freq, phase etc ? thanks -chris

RE: Building Dictionaries [ Reply ]
By: Boris Mailhe on 2008-10-08 09:48

[forum:98613]

Hi,

Dictionary design is still an open research topic. To our knowledge there is currently nothing else available than rules of thumb.

You can try several combinations and measure the obtained SNR for each, or use some a priori knowledge about the typical stationary durations in your signals. You can also perform a test decomposition with a very large dictionary, then check which scales captured the most energy.

Boris

RE: Building Dictionaries [ Reply ] By: chris layman on 2008-10-07 01:39	[forum:98582]
Anybody?

Building Dictionaries [ Reply ] By: chris layman on 2008-09-26 23:25	[forum:98542]
Hi, I was wondering if someone could help me with building appropriate dictionaries in MPTK. Given a specific signal I want to decompose, how does one go about deciding on an appropriate dictionary? I have just been playing around with the default dictionaries, but this seems a bit arbitrary. For example, how does one decide on how many scales to include in the dictionary? thanks -chris