I have to install Maryas again, since the testing (prediction) step can not done properly, I guess maybe I should reinstall the whole package.
So I follow the instructions on the main webpage, everything is OK.
(Here are the instructions:
http://marsyas.info/docs/manual/marsyas-user/Compiling-on-Ubuntu.html#Compiling-on-Ubuntu
)
However, when I reproduce the arff file, there are some problems:
(bextract -sv genres10.mf -w genres10.arff -p genres10.mpl)
RtApi3: no devices found for given stream parameters:
RtApi3Alsa: pcm device (hw:Intel,0) won't open: Device or resource busy.
RtApi3Alsa: error setting sample rate (22050) on device (hw:Intel,1): Invalid argument.
When I using the command without produce the mpl file, this problem did not appear.
When I try to sfplay to play the wav file directly, the same error message appear again, so I guess this is some configuration problem....
RtApi3: no devices found for given stream parameters:
RtApi3Alsa: pcm device (hw:Intel,0) won't open: Device or resource busy.
RtApi3Alsa: error setting sample rate (16000) on device (hw:Intel,1): Invalid argument.
Segmentation fault
Now I want to figure out what is the solution to this problem
=================================
arapi: real time application programming interface
here are the page introduce the sound in linux: http://www-uxsup.csx.cam.ac.uk/pub/doc/suse/suse9.0/userguide-9.0/ch18.html
====================================================================================
For the ERROR message above, I still do not know how to solve, but I found that if I do not produce the mpl file in this step, everything is OK. And also the bextract can be used to do prediction. So I may just put the problem here.....
Actually after reinstall the whole package for several times, and re do all the steps to train and classify for several times, still bextract can not predict.
Finally, I look through several papers written by George Tzanetakis, which is the submission paper to the MIREX 2007, 2009, 2010, and some insignts have been gotten.
Since when we extract the arff files, we use -sv to treat the complete wav file as a whole, so should the prediction part. (I always forgot to append the -sv in the tesiting command). So the final command line should be: bextrac -sv train.mf -tc test.mf
This will give the prediction in the test file list.
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++