Hi Olivier and list, at first thanks for the modification of uncell.m!At the moment I have another problem with the new version: mirinharmonicity function estimates too much NaN values for several test songs (the old version estimated a couple of NaNs only at the beginning of the songs).
Here is my code: [ih,s,ff] = mirinharmonicity(a,'Frame',4096,'sp',4096,'sp'); res=mirgetdata(ih); Do you have any idea? Kind regards, Igor Am 14.08.2012 00:47, schrieb Olivier Lartillot:
Hi Igor, I finally found time to look at that problem. In fact, mirgetdata was not preserving the array direction (column vs. row) of unidimensional cell arrays. I modified the file uncell.m (cf. enclosed) and it seems to work on your examples. I will include this fix in update 1.4.1. All the best, Olivier On 24.7.2012, at 11.12, Igor Vatolkin wrote:The same situation occurs e.g. for [ih,s,ff] = mirinharmonicity(a,'Frame',512,'sp',512,'sp'); res1=mirgetdata(ff); res2=mirgetdata(ih); (comparing res1 and res2) - so this behaviour seems to be systematically if several outputs are generated? -------- Original Message -------- Subject: [mirtoolbox] Vector dimensions in later MIR Toolbox versions Date: Tue, 24 Jul 2012 10:27:59 +0200 From: Igor Vatolkin <igor.vatolkin@xxxxxxx> Reply-To: mirtoolbox@xxxxxxxxxxxxx To: mirtoolbox@xxxxxxxxxxxxx Hi Olivier and MIR Toolbox list, I have a question about vector dimensions in the later MIR Toolboxes. It seems to be that some dimensions have been switched, but not always consistently. Consider the following code: a=miraudio(input_file); f=mirframe(a,'Length',wlength,'sp','Hop',woffset,'sp'); chroma = mirchromagram(f); [k c]=mirkey(chroma); key=mirgetdata(k); clarity=mirgetdata(c); The last version (1.4) returns key as 6878x1 double and clarity as 1x6878 double. Does any reason exist for such behaviour? Best regards, Igor