Python3 support.

PyConcat/Classifier.py

@@ -0,0 +1,15 @@
+from sklearn.neighbors import KNeighborsClassifier
+
+class Classifier:
+
+    def __init__(self):
+        pass
+
+    def trainClassifier(self):
+        neigh = KNeighborsClassifier(n_neighbors=3)
+        neigh.fit(X, y)
+
+    def classifyInstance(self):
+        pass
+
+

PyConcat/Extractor.py

@@ -8,6 +8,8 @@
 import fnmatch
 import os
 import madmom
+import librosa
+from madmom.audio.filters import LogarithmicFilterbank
 from multiprocessing import Process, Queue
 from multiprocessing import Pool
 
@@ -167,8 +169,11 @@ def loadAudio(self, filename):
 
             #Essentia's loader (above)  has a bug that doesn't close files
             #It causes problems processing large number of files, use madmom instead
-            audio, sample_rate = madmom.audio.signal.load_wave_file(filename, num_channels=1)
-            audio = essentia.array(audio)
+            # audio, sample_rate = madmom.audio.signal.load_wave_file(filename, num_channels=1)
+
+            y, sr = librosa.load(filename, sr=None)
+
+            audio = essentia.array(y)
 
         return audio
 
@@ -182,6 +187,7 @@ def getListOfFiles(self, path, filterPattern=""):
         """
         files = []
 
+        #For the recursive functionality we need a list of paths
         if type(path) is not list:
             path = [path]
 
@@ -235,7 +241,7 @@ def extractBeats(self, fileName):
         return ticks, slices, fileName
 
     def extractOnsetsMadmom(self, filename, method="CNN"):
-        """Use an onset detector to return beat locations
+        """Use Madmom's Superflux or CNN detector to return onset locations
 
         :param fileName: the file to load and extract beats from
 
@@ -247,25 +253,49 @@ def extractOnsetsMadmom(self, filename, method="CNN"):
             fileName: pass out the filename again
 
         """
-        act = None
+        onsetFunctionExtractor = None
+        peakPicker = None
 
         #Choose an onset detection function
         if "CNN" in method:
-            act = madmom.features.onsets.CNNOnsetProcessor()
+            onsetFunctionExtractor = madmom.features.onsets.CNNOnsetProcessor(fps=100)
+            peakPicker = madmom.features.onsets.OnsetPeakPickingProcessor(threshold=0.54, smooth=0.05)
         else:
-            act = madmom.features.onsets.SpectralOnsetProcessor(onset_method='superflux', fps=200,
-                                                                 filterbank = madmom.audio.filters.LogarithmicFilterbank,
-                                                                 num_bands = 24, log = np.log10)
-        #get the onsets
-        onsetFunction = act(filename)
+            # onsetFunctionExtractor = madmom.features.onsets.SpectralOnsetProcessor(onset_method='superflux',
+            #                                                     filterbank = LogarithmicFilterbank,
+            #                                                     num_bands = 24, log = np.log10,
+            #                                                     fmin=30, fmax=17000,
+            #                                                     mul=1, add=1,
+            #                                                     diff_ratio=0.5,
+            #                                                     diff_max_bins=3,
+            #                                                     positive_diffs=True
+            #                                                     )
+            # peakPicker = madmom.features.onsets.OnsetPeakPickingProcessor(threshold=1.1, pre_max=0.01,
+            #                                                               post_max=0.05, pre_avg=0.15,
+            #                                                               post_avg=0, combine=0.03, delay=0)
+
+            onsetFunctionExtractor = madmom.features.onsets.SpectralOnsetProcessor(onset_method='superflux',
+                                                                                   filterbank=LogarithmicFilterbank,
+                                                                                   num_bands=24, log=np.log10,
+                                                                                   fmin=27.5, fmax=16000,
+                                                                                   mul=1, add=1,
+                                                                                   diff_ratio=0.5,
+                                                                                   diff_max_bins=3,
+                                                                                   positive_diffs=True
+                                                                                   )
+            #Old SupeFlux Params
+            peakPicker = madmom.features.onsets.OnsetPeakPickingProcessor(threshold=1.25, pre_max=0.03,
+                                                                          post_max=0.03, pre_avg=0.10,
+                                                                          post_avg=0.07, combine=0.03, delay=0)
 
-        proc = madmom.features.onsets.OnsetPeakPickingProcessor(fps=100)
-        onsetTimes = proc(onsetFunction)
+        #get the onsets
+        onsetFunction = onsetFunctionExtractor(filename)
+        onsetTimes = peakPicker(onsetFunction)
 
         return onsetTimes
 
     def extractOnsetsEssentia(self,filename):
-        """Use an onset detector to return beat locations
+        """Use Essentia's Onset Rate onset detector to return beat locations
 
         :param fileName: the file to load and extract beats from
 
@@ -281,12 +311,9 @@ def extractOnsetsEssentia(self,filename):
         onsetTimes = None
 
         onsetRate = essentia.standard.OnsetRate()
-        duration = essentia.standard.Duration()
 
         audio = self.loadAudio(filename)
-        onsetRateResult = onsetRate(audio)
-
-        onsetTimes, onsetRate  = onsetRateResult
+        onsetTimes, onsetRate = onsetRate(audio)
 
         return onsetTimes
 

PyConcat/UnitSelection.py

@@ -1,7 +1,7 @@
 import numpy as np
-import HMM
-import kBestViterbi.kBestViterbi as kb
-import kBestViterbi.networkx_viterbi as kbg
+from .HMM import HMM
+from .kBestViterbi import kBestViterbi as kb
+from .kBestViterbi import networkx_viterbi as kbg
 from scipy.spatial import distance
 from sklearn import preprocessing
 from time import time
@@ -189,7 +189,7 @@ def computeDistanceWithWeights(targetFeatures, corpusFeatures):
 
     return a, b
 
-def unitSelection(targetFeatures, corpusFeatures, method="kdtree", normalise="MinMax", topK=30):
+def unitSelection(targetFeatures, corpusFeatures, method="kdtree", normalise="MinMax", topK=10):
     """Optionally normalise and use one of the methods to return a sequence of indices
     
     :param targetFeatures:
@@ -204,7 +204,7 @@ def unitSelection(targetFeatures, corpusFeatures, method="kdtree", normalise="Mi
     
     :return: the sequence path(s)
     """
-    print "    Scaling and weighting feature vectors..."
+    print("    Scaling and weighting feature vectors...")
 
     scalar = None
 
@@ -229,12 +229,12 @@ def unitSelection(targetFeatures, corpusFeatures, method="kdtree", normalise="Mi
         corpusFeatures = combinedFeatures[len(targetFeatures):, :]
 
     #Call this method to compute the weighted a/b matrices for HMM
-    print "    Computing distance matrices..."
+    print("    Computing distance matrices...")
     a, b = computeDistanceWithWeights(targetFeatures, corpusFeatures)
 
-    print "    Performing unit selection..."
+    print("    Performing unit selection...")
     targetCostWeight = 1.0
-    concatCostWeight = 1.0
+    concatCostWeight = 0.0
 
     #Return the results of the chosen unit selection
     if method == "kdTree":

PyConcat/kBestViterbi/kBestViterbi.py

@@ -1,7 +1,7 @@
 import numpy as np
 import pandas as pd
 import networkx as nx
-import networkx_viterbi as nxv
+from . import networkx_viterbi as nxv
 import heapq
 import itertools
 
@@ -34,7 +34,7 @@ def exhaustiveWithCosts(a, b):
 
         scores.append((score, ss))
 
-        print ss
+        print(ss)
 
     return sorted(scores)
 

PyConcat/kBestViterbi/networkx_viterbi.py

@@ -1,6 +1,6 @@
 import networkx as nx
 import numpy as np
-from simple_paths_with_costs import * #Our custom simple_paths
+from . simple_paths_with_costs import * #Our custom simple_paths
 import heapq
 import itertools
 
@@ -295,6 +295,6 @@ def kViterbiGraphWithCosts(a, b, topK, weights=(1.0, 1.0)):
 
     t1 = time()
 
-    print 'function ITSELF takes %f' % (t1 - t0)
+    print('function ITSELF takes %f' % (t1 - t0))
 
     return paths

config.yaml

@@ -1,5 +1,5 @@
 timeScale : 'onsets'
 writeOnsets : False
-unitSelection : 'linearSearch'
+unitSelection : 'kdTree'
 normalisation : 'MinMax'
 stretchUnits : True

main.py

@@ -92,7 +92,7 @@ def parser(lgd=False, threshold=1.1):
     args = p.parse_args()
     # print arguments
     if args.verbose:
-        print args
+        print(args)
     # return args
 
     #Override the command line arguments with YAML
@@ -126,14 +126,17 @@ def main(args):
     targetFilename, corpusPath = getCorpus(args.input)
 
     #Get list of corpus files
-    corpusFilenames = extractor.getListOfWavFiles(corpusPath)
+    corpusFilenames = extractor.getListOfFiles(corpusPath, "*.mp3")
+
+    print(corpusPath)
 
     #Segment and extract features
     print("Extracting Target")
     targetFeatures, targetUnits, targetUnitTimes = extractor.analyseFile(targetFilename, writeOnsets, scale=timeScale)
     print("Extracting Corpus")
     corpusFeatures, corpusUnits, corpusUnitTimes = extractor.analyseFiles(corpusFilenames, writeOnsets, scale=timeScale, yamlOutputFolder=outputPath)
 
+
     # # For graphing
     # costMatrix = computeDistanceMatrix(targetFeatures, targetFeatures)
     #
@@ -171,7 +174,7 @@ def main(args):
     #Optionally plot data
     #plotData(sequence, targetFeatures, corpusFeatures)
 
-    print "done"
+    print("done")
 
 def readYAMLConfig(filename, args):
     """