Mercurial > hg > qm-dsp
changeset 104:1e433aaa44ad
Merge mepd_new_params branch
| author | Chris Cannam |
|---|---|
| date | Mon, 02 Sep 2013 09:26:46 +0100 |
| parents | 37449f085a4c (current diff) 941b28572da2 (diff) |
| children | c32a2446f7fe |
| files | dsp/tempotracking/TempoTrackV2.cpp |
| diffstat | 3 files changed, 92 insertions(+), 48 deletions(-) [+] |
line wrap: on
line diff
--- a/build/osx/Makefile.osx Thu Jun 13 10:23:09 2013 +0100 +++ b/build/osx/Makefile.osx Mon Sep 02 09:26:46 2013 +0100 @@ -1,5 +1,5 @@ -CFLAGS := -arch i386 -arch ppc -arch x86_64 -isysroot /Developer/SDKs/MacOSX10.4u.sdk -I/Developer/SDKs/MacOSX10.4u.sdk/System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/Headers/ -DUSE_PTHREADS +CFLAGS := -arch i386 -arch x86_64 -isysroot /Developer/SDKs/MacOSX10.6.sdk -I/Developer/SDKs/MacOSX10.6.sdk/System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/Headers/ -DUSE_PTHREADS CXXFLAGS := $(CFLAGS)
--- a/dsp/tempotracking/TempoTrackV2.cpp Thu Jun 13 10:23:09 2013 +0100 +++ b/dsp/tempotracking/TempoTrackV2.cpp Mon Sep 02 09:26:46 2013 +0100 @@ -41,7 +41,7 @@ b[0] = 0.2066; b[1] = 0.4131; b[2] = 0.2066; - + double inp1 = 0.; double inp2 = 0.; double out1 = 0.; @@ -67,7 +67,7 @@ for (unsigned int i = 0;i < df.size();i++) { - lp_df[i] = 0.; + lp_df[i] = 0.; } inp1 = 0.; inp2 = 0.; @@ -91,10 +91,17 @@ } +// MEPD 28/11/12 +// This function now allows for a user to specify an inputtempo (in BPM) +// and a flag "constraintempo" which replaces the general rayleigh weighting for periodicities +// with a gaussian which is centered around the input tempo +// Note, if inputtempo = 120 and constraintempo = false, then functionality is +// as it was before void TempoTrackV2::calculateBeatPeriod(const vector<double> &df, vector<double> &beat_period, - vector<double> &tempi) + vector<double> &tempi, + double inputtempo, bool constraintempo) { // to follow matlab.. split into 512 sample frames with a 128 hop size // calculate the acf, @@ -103,13 +110,42 @@ // and get best path unsigned int wv_len = 128; - double rayparam = 43.; + + // MEPD 28/11/12 + // the default value of inputtempo in the beat tracking plugin is 120 + // so if the user specifies a different inputtempo, the rayparam will be updated + // accordingly. + // note: 60*44100/512 is a magic number + // this might (will?) break if a user specifies a different frame rate for the onset detection function + double rayparam = (60*44100/512)/inputtempo; + + // these debug statements can be removed. + std::cout << "inputtempo" << inputtempo << std::endl; + std::cout << "rayparam" << rayparam << std::endl; + std::cout << "constraintempo" << constraintempo << std::endl; // make rayleigh weighting curve d_vec_t wv(wv_len); - for (unsigned int i=0; i<wv.size(); i++) + + // check whether or not to use rayleigh weighting (if constraintempo is false) + // or use gaussian weighting it (constraintempo is true) + if (constraintempo) { - wv[i] = (static_cast<double> (i) / pow(rayparam,2.)) * exp((-1.*pow(-static_cast<double> (i),2.)) / (2.*pow(rayparam,2.))); + for (unsigned int i=0; i<wv.size(); i++) + { + // MEPD 28/11/12 + // do a gaussian weighting instead of rayleigh + wv[i] = exp( (-1.*pow((static_cast<double> (i)-rayparam),2.)) / (2.*pow(rayparam/4.,2.)) ); + } + } + else + { + for (unsigned int i=0; i<wv.size(); i++) + { + // MEPD 28/11/12 + // standard rayleigh weighting over periodicities + wv[i] = (static_cast<double> (i) / pow(rayparam,2.)) * exp((-1.*pow(-static_cast<double> (i),2.)) / (2.*pow(rayparam,2.))); + } } // beat tracking frame size (roughly 6 seconds) and hop (1.5 seconds) @@ -130,9 +166,9 @@ dfframe[k] = df[i+k]; } // get rcf vector for current frame - d_vec_t rcf(wv_len); + d_vec_t rcf(wv_len); get_rcf(dfframe,wv,rcf); - + rcfmat.push_back( d_vec_t() ); // adds a new column col_counter++; for (unsigned int j=0; j<rcf.size(); j++) @@ -140,7 +176,7 @@ rcfmat[col_counter].push_back( rcf[j] ); } } - + // now call viterbi decoding function viterbi_decode(rcfmat,wv,beat_period,tempi); } @@ -161,7 +197,7 @@ d_vec_t acf(dfframe.size()); - + for (unsigned int lag=0; lag<dfframe.size(); lag++) { double sum = 0.; @@ -169,7 +205,7 @@ for (unsigned int n=0; n<(dfframe.size()-lag); n++) { - tmp = dfframe[n] * dfframe[n+lag]; + tmp = dfframe[n] * dfframe[n+lag]; sum += tmp; } acf[lag] = static_cast<double> (sum/ (dfframe.size()-lag)); @@ -177,7 +213,7 @@ // now apply comb filtering int numelem = 4; - + for (unsigned int i = 2;i < rcf.size();i++) // max beat period { for (int a = 1;a <= numelem;a++) // number of comb elements @@ -188,10 +224,10 @@ } } } - + // apply adaptive threshold to rcf MathUtilities::adaptiveThreshold(rcf); - + double rcfsum =0.; for (unsigned int i=0; i<rcf.size(); i++) { @@ -218,11 +254,11 @@ { tmat.push_back ( d_vec_t() ); // adds a new column for (unsigned int j=0; j<wv.size(); j++) - { + { tmat[i].push_back(0.); // fill with zeros initially } } - + // variance of Gaussians in transition matrix // formed of Gaussians on diagonal - implies slow tempo change double sigma = 8.; @@ -230,7 +266,7 @@ for (unsigned int i=20;i <wv.size()-20; i++) { for (unsigned int j=20; j<wv.size()-20; j++) - { + { double mu = static_cast<double>(i); tmat[i][j] = exp( (-1.*pow((j-mu),2.)) / (2.*pow(sigma,2.)) ); } @@ -246,7 +282,7 @@ delta.push_back( d_vec_t()); psi.push_back( i_vec_t()); for (unsigned int j=0; j<rcfmat[i].size(); j++) - { + { delta[i].push_back(0.); // fill with zeros initially psi[i].push_back(0); // fill with zeros initially } @@ -265,16 +301,16 @@ delta[0][j] = wv[j] * rcfmat[0][j]; psi[0][j] = 0; } - + double deltasum = 0.; for (unsigned int i=0; i<Q; i++) { deltasum += delta[0][i]; - } + } for (unsigned int i=0; i<Q; i++) { delta[0][i] /= (deltasum + EPS); - } + } for (unsigned int t=1; t<T; t++) @@ -286,12 +322,12 @@ for (unsigned int i=0; i<Q; i++) { tmp_vec[i] = delta[t-1][i] * tmat[j][i]; - } - - delta[t][j] = get_max_val(tmp_vec); + } + + delta[t][j] = get_max_val(tmp_vec); psi[t][j] = get_max_ind(tmp_vec); - + delta[t][j] *= rcfmat[t][j]; } @@ -300,23 +336,23 @@ for (unsigned int i=0; i<Q; i++) { deltasum += delta[t][i]; - } + } for (unsigned int i=0; i<Q; i++) { delta[t][i] /= (deltasum + EPS); - } + } } i_vec_t bestpath(T); d_vec_t tmp_vec(Q); for (unsigned int i=0; i<Q; i++) - { + { tmp_vec[i] = delta[T-1][i]; } // find starting point - best beat period for "last" frame bestpath[T-1] = get_max_ind(tmp_vec); - + // backtrace through index of maximum values in psi for (unsigned int t=T-2; t>0 ;t--) { @@ -328,7 +364,7 @@ unsigned int lastind = 0; for (unsigned int i=0; i<T; i++) - { + { unsigned int step = 128; for (unsigned int j=0; j<step; j++) { @@ -361,7 +397,7 @@ maxval = df[i]; } } - + return maxval; } @@ -378,7 +414,7 @@ ind = i; } } - + return ind; } @@ -390,17 +426,21 @@ { sum += df[i]; } - + for (unsigned int i=0; i<df.size(); i++) { df[i]/= (sum + EPS); } } +// MEPD 28/11/12 +// this function has been updated to allow the "alpha" and "tightness" parameters +// of the dynamic program to be set by the user +// the default value of alpha = 0.9 and tightness = 4 void TempoTrackV2::calculateBeats(const vector<double> &df, const vector<double> &beat_period, - vector<double> &beats) + vector<double> &beats, double alpha, double tightness) { if (df.empty() || beat_period.empty()) return; @@ -414,8 +454,12 @@ backlink[i] = -1; } - double tightness = 4.; - double alpha = 0.9; + //double tightness = 4.; + //double alpha = 0.9; + // MEPD 28/11/12 + // debug statements that can be removed. + std::cout << "alpha" << alpha << std::endl; + std::cout << "tightness" << tightness << std::endl; // main loop for (unsigned int i=0; i<localscore.size(); i++) @@ -436,7 +480,7 @@ // ELSE LEAVE AT DEFAULT VALUE FROM INITIALISATION: D_VEC_T SCORECANDS (TXWT.SIZE()); int cscore_ind = i+prange_min+j; - if (cscore_ind >= 0) + if (cscore_ind >= 0) { scorecands[j] = txwt[j] * cumscore[cscore_ind]; } @@ -457,7 +501,7 @@ for (unsigned int i=cumscore.size() - beat_period[beat_period.size()-1] ; i<cumscore.size(); i++) { tmp_vec.push_back(cumscore[i]); - } + } int startpoint = get_max_ind(tmp_vec) + cumscore.size() - beat_period[beat_period.size()-1] ; @@ -476,10 +520,10 @@ if (backlink[b] == b) break; // shouldn't happen... haha ibeats.push_back(backlink[b]); } - + // REVERSE SEQUENCE OF IBEATS AND STORE AS BEATS for (unsigned int i=0; i<ibeats.size(); i++) - { + { beats.push_back( static_cast<double>(ibeats[ibeats.size()-i-1]) ); } }
--- a/dsp/tempotracking/TempoTrackV2.h Thu Jun 13 10:23:09 2013 +0100 +++ b/dsp/tempotracking/TempoTrackV2.h Mon Sep 02 09:26:46 2013 +0100 @@ -18,16 +18,14 @@ #define TEMPOTRACKV2_H #include <vector> -#include <cstddef> - -using std::vector; +using namespace std; //!!! Question: how far is this actually sample rate dependent? I // think it does produce plausible results for e.g. 48000 as well as // 44100, but surely the fixed window sizes and comb filtering will // make it prefer double or half time when run at e.g. 96000? -class TempoTrackV2 +class TempoTrackV2 { public: /** @@ -41,15 +39,17 @@ TempoTrackV2(float sampleRate, size_t dfIncrement); ~TempoTrackV2(); - // Returned beat periods are given in df increment units; tempi in bpm + // Returned beat periods are given in df increment units; inputtempo and tempi in bpm + // MEPD 28/11/12 Expose inputtempo and constraintempo parameters void calculateBeatPeriod(const vector<double> &df, vector<double> &beatPeriod, - vector<double> &tempi); + vector<double> &tempi, double inputtempo, bool constraintempo); // Returned beat positions are given in df increment units + // MEPD 28/11/12 Expose alpha and tightness parameters void calculateBeats(const vector<double> &df, const vector<double> &beatPeriod, - vector<double> &beats); + vector<double> &beats, double alpha, double tightness); private: typedef vector<int> i_vec_t;