cannam@26: /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*-  vi:set ts=8 sts=4 sw=4: */
cannam@25: 
cannam@26: /*
cannam@26:     QM DSP Library
cannam@25: 
cannam@26:     Centre for Digital Music, Queen Mary, University of London.
cannam@26:     This file copyright 2005 Nicolas Chetry, copyright 2008 QMUL.
Chris@84: 
Chris@84:     This program is free software; you can redistribute it and/or
Chris@84:     modify it under the terms of the GNU General Public License as
Chris@84:     published by the Free Software Foundation; either version 2 of the
Chris@84:     License, or (at your option) any later version.  See the file
Chris@84:     COPYING included with this distribution for more information.
cannam@26: */
cannam@25: 
cannam@26: #ifndef MFCC_H
cannam@26: #define MFCC_H
cannam@26: 
cannam@26: #include "base/Window.h"
cannam@26: 
cannam@64: class FFTReal;
cannam@64: 
cannam@26: struct MFCCConfig {
cannam@26:     int FS;
cannam@26:     int fftsize;
cannam@26:     int nceps;
cannam@30:     double logpower;
cannam@26:     bool want_c0;
cannam@32:     WindowType window;
cannam@30:     MFCCConfig(int _FS) :
cannam@32:         FS(_FS), fftsize(2048), nceps(19),
cannam@32:         logpower(1.0), want_c0(true), window(HammingWindow) { }
cannam@26: };
cannam@26: 
cannam@26: class MFCC
cannam@26: {
cannam@26: public:
cannam@26:     MFCC(MFCCConfig config);
cannam@26:     virtual ~MFCC();
cannam@26: 
cannam@30:     /**
cannam@30:      * Process time-domain input data.  inframe must contain
cannam@30:      * getfftlength() samples.  outceps must contain space for nceps
cannam@30:      * values, plus one if want_c0 is specified.
cannam@30:      */
cannam@30:     int process(const double *inframe, double *outceps);
cannam@30: 
cannam@30:     /**
cannam@30:      * Process time-domain input data.  real and imag must contain
cannam@30:      * getfftlength()/2+1 elements (i.e. the conjugate half of the FFT
cannam@30:      * is not expected).  outceps must contain space for nceps values,
cannam@30:      * plus one if want_c0 is specified.
cannam@30:      */
cannam@30:     int process(const double *real, const double *imag, double *outceps);
cannam@26: 
cannam@26:     int getfftlength() const { return fftSize; }
cannam@26: 
cannam@26: private:
cannam@26:     /* Filter bank parameters */
cannam@26:     double  lowestFrequency; 
cannam@26:     int     linearFilters; 
cannam@26:     double  linearSpacing;
cannam@26:     int     logFilters;
cannam@26:     double  logSpacing;
cannam@26:     
cannam@26:     /* FFT length */
cannam@26:     int     fftSize;
cannam@26:     
cannam@26:     int     totalFilters;
cannam@30:     double  logPower;
cannam@26:     
cannam@26:     /* Misc. */
cannam@26:     int     samplingRate;
cannam@26:     int     nceps;
cannam@26:     
cannam@26:     /* MFCC vector */
cannam@26:     double  *ceps;
cannam@26:     
cannam@26:     double  **mfccDCTMatrix;
cannam@26:     double  **mfccFilterWeights;
cannam@26:     
cannam@26:     /* The analysis window */
cannam@26:     Window<double> *window;
cannam@26:     
cannam@26:     /* For the FFT */
cannam@30:     double *realOut;
cannam@30:     double *imagOut;
cannam@30:     double *fftMag;
cannam@30:     double *earMag;
cannam@64:     FFTReal *fft;
cannam@30: 
cannam@26:     /* Set if user want C0 */
cannam@26:     int WANT_C0;
cannam@26: };
cannam@26: 
cannam@25: 
cannam@25: #endif
cannam@25: