def feature_allframes(input_features, frame_indexer = None): audio = input_features['audio'] beats = input_features['beats'] # Initialise the algorithms w = Windowing(type = 'hann') loudness = Loudness() if frame_indexer is None: frame_indexer = range(1,len(beats) - 1) # Exclude first frame, because it has no predecessor to calculate difference with # 1 loudness value by default loudness_values = np.zeros((len(beats), 1)) # 1 difference value between loudness value cur and cur-1 # 1 difference value between loudness value cur and cur-4 # 1 difference value between differences above loudness_differences = np.zeros((len(beats), 9)) # Step 1: Calculate framewise for all output frames # Calculate this for all frames where this frame, or its successor, is in the frame_indexer for i in [i for i in range(len(beats)) if (i in frame_indexer) or (i+1 in frame_indexer) or (i-1 in frame_indexer) or (i-2 in frame_indexer) or (i-3 in frame_indexer) or (i-4 in frame_indexer) or (i-5 in frame_indexer) or (i-6 in frame_indexer) or (i-7 in frame_indexer) or (i-8 in frame_indexer)]: SAMPLE_RATE = 44100 start_sample = int(beats[i] * SAMPLE_RATE) end_sample = int(beats[i+1] * SAMPLE_RATE) #print start_sample, end_sample frame = audio[start_sample : end_sample if (start_sample - end_sample) % 2 == 0 else end_sample - 1] loudness_values[i] = loudness(w(frame)) # Step 2: Calculate the cosine distance between the MFCC values for i in frame_indexer: loudness_differences[i][0] = (loudness_values[i] - loudness_values[i-1]) loudness_differences[i][1] = (loudness_values[i+1] - loudness_values[i]) loudness_differences[i][2] = (loudness_values[i+2] - loudness_values[i]) loudness_differences[i][3] = (loudness_values[i+3] - loudness_values[i]) loudness_differences[i][4] = (loudness_values[i+4] - loudness_values[i]) loudness_differences[i][5] = (loudness_values[i+5] - loudness_values[i]) loudness_differences[i][6] = (loudness_values[i+6] - loudness_values[i]) loudness_differences[i][7] = (loudness_values[i+7] - loudness_values[i]) loudness_differences[i][8] = (loudness_values[i-1] - loudness_values[i+1]) # Include the raw values as absolute features result = loudness_differences[frame_indexer] #~ print np.shape(result), np.shape(loudness_values), np.shape(loudness_differences) return preprocessing.scale(result)
def lowLevel(songName):
global dataset
global lock
print songName
#REMOVE ; AND , FROM SONGNAMES
key = re.sub(r',', "", songName.split('/')[-1])
key = re.sub(r';', "", key)
#DONT HAVE TO EXTRACT IF IT IS ALREADY EXTRACTED
if key in dataset.keys():
feature = dataset[key]
return feature
else:
loader = MonoLoader(filename=songName)
audio = loader()
extractor = LowLevelSpectralEqloudExtractor()
feature = list(extractor(audio))
del feature[1]
del feature[1]
extractor = LowLevelSpectralExtractor()
featureTwo = list(extractor(audio))
del featureTwo[0]
del featureTwo[-2]
featureTwo[4] = feature[4][1]
feature.extend(featureTwo)
extractor = Loudness()
feature.append(extractor(audio))
extractor = LogAttackTime()
feature.append(extractor(audio)[0])
extractor = KeyExtractor()
feature.append(extractor(audio)[2])
extractor = RhythmExtractor2013()
data = extractor(audio)
feature.append(data[0])
feature.append(data[2])
for x in range(len(feature)):
if type(feature[x]) is np.ndarray:
#feature[x] = avg(feature[x])
mean, std = stdDev(feature[x])
feature[x] = mean
feature.append(std)
arr = key + "," + str(feature)[1:-1] + "\n"
f = open('data.csv', 'a')
lock.acquire()
f.write(arr)
lock.release()
f.close()
return feature
def lowLevel(songName):
global dataset
global lock
print songName
key = re.sub(r',', "", songName.split('/')[-1])
#IF already present in dataset dont extract
if dataset.has_key(key):
feature = dataset[key]
return feature
else:
#Loading song and using Extractors
loader = MonoLoader(filename=songName)
audio = loader()
extractor = LowLevelSpectralEqloudExtractor()
feature = list(extractor(audio))
del feature[1]
del feature[1]
extractor = LowLevelSpectralExtractor()
featureTwo = list(extractor(audio))
del featureTwo[0]
del featureTwo[-2]
featureTwo[4] = feature[4][1]
feature.extend(featureTwo)
extractor = Loudness()
feature.append(extractor(audio))
extractor = LogAttackTime()
feature.append(extractor(audio)[0])
extractor = KeyExtractor()
feature.append(extractor(audio)[2])
extractor = RhythmExtractor2013()
data = extractor(audio)
feature.append(data[0])
feature.append(data[2])
for x in range(len(feature)):
if type(feature[x]) is np.ndarray:
feature[x] = avg(feature[x])
arr = key + "," + str(feature)[1:-1] + "\n"
f = open('data.csv', 'a')
lock.acquire()
f.write(arr)
lock.release()
f.close()
return feature
def shared_main(source, dest, display_result):
source_audio = _loader(source)
destination_audio = _loader(dest)
source_frame = FrameGenerator(source_audio, frameSize=2048, hopSize=512)
destination_frame = FrameGenerator(destination_audio,
frameSize=2048,
hopSize=512)
window = Windowing(type='hann') # window function
spectrum = Spectrum() # spectrum function
pitch_yin_fft = PitchYinFFT() # pitch extractor
pitch_saliennce = PitchSalience()
loudness = Loudness()
# draw_plot(source_frame, window, spectrum, pitch_yin_fft)
min_cost, match_result = compare(source_frame, destination_frame, window, \
spectrum, pitch_yin_fft, 5, 1, 1, display_result, loudness)
return min_cost, match_result
def feature_allframes(audio, beats, frame_indexer = None): # Initialise the algorithms w = Windowing(type = 'hann') loudness = Loudness() if frame_indexer is None: frame_indexer = range(1,len(beats) - 1) # Exclude first frame, because it has no predecessor to calculate difference with # 1 loudness value by default loudness_values = np.zeros((len(beats), 1)) # 1 difference value between loudness value cur and cur-1 # 1 difference value between loudness value cur and cur-4 # 1 difference value between differences above loudness_feature_vector = np.zeros((len(beats), 4)) # Step 1: Calculate framewise for all output frames # Calculate this for all frames where this frame, or its successor, is in the frame_indexer for i in [i for i in range(len(beats)) if (i in frame_indexer) or (i-1 in frame_indexer) or (i-2 in frame_indexer) or (i-3 in frame_indexer)]: SAMPLE_RATE = 44100 start_sample = int(beats[i] * SAMPLE_RATE) end_sample = int(beats[i+1] * SAMPLE_RATE) #print start_sample, end_sample frame = audio[start_sample : end_sample if (start_sample - end_sample) % 2 == 0 else end_sample - 1] loudness_values[i] = loudness(w(frame)) loudness_values = preprocessing.scale(loudness_values) # Step 2: construct feature vector for i in frame_indexer: loudness_feature_vector[i] = np.reshape(loudness_values[i:i+4], (4,)) # Include the raw values as absolute features result = loudness_feature_vector[frame_indexer] #~ print np.shape(result), np.shape(loudness_values), np.shape(loudness_differences) return result