A new approach to audio-mosaicing using probabilistic modelling.
Once downloaded, open Simulsig.xcodeproj found in Markov Chains/FFT Chain/
-
In SetupMacros.h, found in the Main / Setup.h subfolder in Navigator window, change the Macro FOLDERPATH to the included Products directory.
-
Populate the Analysis Corpus folder, found in Products/, with a selection of mono .wav files.
-
Populate the Training Corpus folder, found in Products/, with a selection of mono .wav files on which the Markov chain shall be trained.
-
In main.cpp, change the variable Phase to AllPhase to run through the entire program. dependencies. If you do not wish to see the multitude of console printouts, alter the macro in PrintfDebug.h
-
Tracks are written to /Products/Final Tracks
Several macros determine the type out output produced, all of which are found in SetupMacros.h
NSAMP - Audio segment size
FNMELBANDS - Number of mel-filterbanks used during extraction. this increases the number of MFFCs
MFCCGROUPING - Number of MFC frames grouped together
MAXCLUSTITERATIONS - Maximum number of iterations of K-means algorithm. The larger the better.
NUMCLUSTER - Maximum number of clusters. The majority of clusters do not have a unique centroid, so the number of active clusters will be significantly less than this.
Texture-synthesis like results: NSAMP - 4096 FNMELBANDS - 14 MFCCGROUPING - 1
Automatic track remixer example: NSAMP - 22050 FNMELBANDS - 14 MFCCGROUPING - 6
FFTW3, libsndfile, TNT, LAPACK, NanoFlann (C++ only subset of FLANN), Accelerate framework (subset of within OSX veclib), Dirent.h (included by default OSX 10.7 > )