def extractFeatures(self, eventsList, Fs, snr):
feature = []
for event in eventsList:
frame = event.getData()
F = audioFeatureExtraction.stFeatureExtraction(
frame, Fs, self.frameSize * Fs, self.frameStep * Fs)
raw_feature = F[:self.discard, :].T
tmp = []
for j in range(0, raw_feature.shape[1]
): # compute median and med for each columns
feature_column = raw_feature[:, j]
median = np.median(raw_feature[:, j])
median_absolute_deviation = np.median(
np.abs(feature_column - median))
tmp.append(median)
tmp.append(median_absolute_deviation)
tmp.append(event.getTarget()) # add class label
tmp.append(raw_feature.shape[0]) # add number of frame per signal
tmp.append(snr) # add snr
tmp.append(event.getId()) # add id
tmp.append(event.getBackground()) # add background type
feature.append(tmp)
return feature
def beatExtractionWrapper(wavFileName, plot):
if not os.path.isfile(wavFileName):
raise Exception("Input audio file not found!")
[Fs, x] = audioBasicIO.readAudioFile(wavFileName)
F = aF.stFeatureExtraction(x, Fs, 0.050 * Fs, 0.050 * Fs)
BPM, ratio = aF.beatExtraction(F, 0.050, plot)
print "Beat: {0:d} bpm ".format(int(BPM))
print "Ratio: {0:.2f} ".format(ratio)
def beatExtractionWrapper(wavFileName, plot):
if not os.path.isfile(wavFileName):
raise Exception("Input audio file not found!")
[Fs, x] = audioBasicIO.readAudioFile(wavFileName)
F = aF.stFeatureExtraction(x, Fs, 0.050 * Fs, 0.050 * Fs)
BPM, ratio = aF.beatExtraction(F, 0.050, plot)
print "Beat: {0:d} bpm ".format(int(BPM))
print "Ratio: {0:.2f} ".format(ratio)
def beatExtractionWrapper(wav_file, plot):
if not os.path.isfile(wav_file):
raise Exception("Input audio file not found!")
[fs, x] = audioBasicIO.readAudioFile(wav_file)
F, _ = aF.stFeatureExtraction(x, fs, 0.050 * fs, 0.050 * fs)
bpm, ratio = aF.beatExtraction(F, 0.050, plot)
print("Beat: {0:d} bpm ".format(int(bpm)))
print("Ratio: {0:.2f} ".format(ratio))
def musicThumbnailing(x, Fs, shortTermSize=1.0, shortTermStep=0.5, thumbnailSize=10.0): ''' This function detects instances of the most representative part of a music recording, also called "music thumbnails". A technique similar to the one proposed in [1], however a wider set of audio features is used instead of chroma features. In particular the following steps are followed: - Extract short-term audio features. Typical short-term window size: 1 second - Compute the self-silimarity matrix, i.e. all pairwise similarities between feature vectors - Apply a diagonal mask is as a moving average filter on the values of the self-similarty matrix. The size of the mask is equal to the desirable thumbnail length. - Find the position of the maximum value of the new (filtered) self-similarity matrix. The audio segments that correspond to the diagonial around that position are the selected thumbnails ARGUMENTS: - x: input signal - Fs: sampling frequency - shortTermSize: window size (in seconds) - shortTermStep: window step (in seconds) - thumbnailSize: desider thumbnail size (in seconds) RETURNS: - A1: beginning of 1st thumbnail (in seconds) - A2: ending of 1st thumbnail (in seconds) - B1: beginning of 2nd thumbnail (in seconds) - B2: ending of 2nd thumbnail (in seconds) USAGE EXAMPLE: import audioFeatureExtraction as aF [Fs, x] = basicIO.readAudioFile(inputFile) [A1, A2, B1, B2] = musicThumbnailing(x, Fs) [1] Bartsch, M. A., & Wakefield, G. H. (2005). Audio thumbnailing of popular music using chroma-based representations. Multimedia, IEEE Transactions on, 7(1), 96-104. ''' x = audioBasicIO.stereo2mono(x); # feature extraction: stFeatures = aF.stFeatureExtraction(x, Fs, Fs*shortTermSize, Fs*shortTermStep) # self-similarity matrix S = selfSimilarityMatrix(stFeatures) # moving filter: M = int(round(thumbnailSize / shortTermStep)) B = numpy.eye(M,M) S = scipy.signal.convolve2d(S, B, 'valid') # post-processing (remove main diagonal elements) MIN = numpy.min(S) for i in range(S.shape[0]): for j in range(S.shape[1]): if abs(i-j) < 5.0 / shortTermStep or i > j: S[i,j] = MIN; # find max position: maxVal = numpy.max(S) I = numpy.argmax(S) [I, J] = numpy.unravel_index(S.argmax(), S.shape) # expand: i1 = I; i2 = I j1 = J; j2 = J while i2-i1<M: if S[i1-1, j1-1] > S[i2+1,j2+1]: i1 -= 1 j1 -= 1 else: i2 += 1 j2 += 1 return (shortTermStep*i1, shortTermStep*i2, shortTermStep*j1, shortTermStep*j2, S)
def silenceRemoval(x, Fs, stWin, stStep, smoothWindow = 0.5, Weight = 0.5, plot = False):
'''
Event Detection (silence removal)
ARGUMENTS:
- x: the input audio signal
- Fs: sampling freq
- stWin, stStep: window size and step in seconds
- smoothWindow: (optinal) smooth window (in seconds)
- Weight: (optinal) weight factor (0 < Weight < 1) the higher, the more strict
- plot: (optinal) True if results are to be plotted
RETURNS:
- segmentLimits: list of segment limits in seconds (e.g [[0.1, 0.9], [1.4, 3.0]] means that
the resulting segments are (0.1 - 0.9) seconds and (1.4, 3.0) seconds
'''
if Weight>=1:
Weight = 0.99;
if Weight<=0:
Weight = 0.01;
# Step 1: feature extraction
x = audioBasicIO.stereo2mono(x); # convert to mono
ShortTermFeatures = aF.stFeatureExtraction(x, Fs, stWin*Fs, stStep*Fs) # extract short-term features
# Step 2: train binary SVM classifier of low vs high energy frames
EnergySt = ShortTermFeatures[1, :] # keep only the energy short-term sequence (2nd feature)
E = numpy.sort(EnergySt) # sort the energy feature values:
L1 = int(len(E)/10) # number of 10% of the total short-term windows
T1 = numpy.mean(E[0:L1]) # compute "lower" 10% energy threshold
T2 = numpy.mean(E[-L1:-1]) # compute "higher" 10% energy threshold
Class1 = ShortTermFeatures[:,numpy.where(EnergySt<T1)[0]] # get all features that correspond to low energy
Class2 = ShortTermFeatures[:,numpy.where(EnergySt>T2)[0]] # get all features that correspond to high energy
featuresSS = [Class1.T, Class2.T]; # form the binary classification task and ...
[featuresNormSS, MEANSS, STDSS] = aT.normalizeFeatures(featuresSS) # normalize and ...
SVM = aT.trainSVM(featuresNormSS, 1.0) # train the respective SVM probabilistic model (ONSET vs SILENCE)
# Step 3: compute onset probability based on the trained SVM
ProbOnset = []
for i in range(ShortTermFeatures.shape[1]): # for each frame
curFV = (ShortTermFeatures[:,i] - MEANSS) / STDSS # normalize feature vector
ProbOnset.append(SVM.pred_probability(curFV)[1]) # get SVM probability (that it belongs to the ONSET class)
ProbOnset = numpy.array(ProbOnset)
ProbOnset = smoothMovingAvg(ProbOnset, smoothWindow / stStep) # smooth probability
# Step 4A: detect onset frame indices:
ProbOnsetSorted = numpy.sort(ProbOnset) # find probability Threshold as a weighted average of top 10% and lower 10% of the values
Nt = ProbOnsetSorted.shape[0] / 10;
T = (numpy.mean( (1-Weight)*ProbOnsetSorted[0:Nt] ) + Weight*numpy.mean(ProbOnsetSorted[-Nt::]) )
MaxIdx = numpy.where(ProbOnset>T)[0]; # get the indices of the frames that satisfy the thresholding
i = 0;
timeClusters = []
segmentLimits = []
# Step 4B: group frame indices to onset segments
while i<len(MaxIdx): # for each of the detected onset indices
curCluster = [MaxIdx[i]]
if i==len(MaxIdx)-1:
break
while MaxIdx[i+1] - curCluster[-1] <= 2:
curCluster.append(MaxIdx[i+1])
i += 1
if i==len(MaxIdx)-1:
break
i += 1
timeClusters.append(curCluster)
segmentLimits.append([curCluster[0]*stStep, curCluster[-1]*stStep])
# Step 5: Post process: remove very small segments:
minDuration = 0.2;
segmentLimits2 = []
for s in segmentLimits:
if s[1] - s[0] > minDuration:
segmentLimits2.append(s)
segmentLimits = segmentLimits2;
if plot:
timeX = numpy.arange(0, x.shape[0] / float(Fs) , 1.0/Fs)
plt.subplot(2,1,1); plt.plot(timeX, x)
for s in segmentLimits:
plt.axvline(x=s[0]);
plt.axvline(x=s[1]);
plt.subplot(2,1,2); plt.plot(numpy.arange(0, ProbOnset.shape[0] * stStep, stStep), ProbOnset);
plt.title('Signal')
for s in segmentLimits:
plt.axvline(x=s[0]);
plt.axvline(x=s[1]);
plt.title('SVM Probability')
plt.show()
return segmentLimits
import audioBasicIO
import audioFeatureExtraction
import matplotlib.pyplot as plt
import numpy
print("COUNT 1\n")
[Fs1, x1] = audioBasicIO.readAudioFile("data/practice.wav")
F1 = audioFeatureExtraction.stFeatureExtraction(x1, Fs1, 0.050 * Fs1,
0.025 * Fs1)
# F1[12*420] MATRIX
print(len(F1[9:21]), len(F1[9:21][0]))
print("\n\nCOUNT 2\n")
[Fs2, x2] = audioBasicIO.readAudioFile("data/practice2.wav")
F2 = audioFeatureExtraction.stFeatureExtraction(x2, Fs2, 0.050 * Fs2,
0.025 * Fs2)
print(len(F2[9:21]), len(F2[9:21][0]))
size = min(len(F2[9:21][0]), len(F1[9:21][0]))
print(size, "\n")
print("\n\nCORRCOEF\n")
print(numpy.corrcoef(F1[9:21, 0:size], F2[9:21, 0:size]) * 0.5 + 0.5)
print("\n\nE Distance\n")
print(numpy.linalg.norm(F1[9:21, 0:size] - F2[9:21, 0:size]))
def silenceRemoval(x,
fs,
st_win,
st_step,
smoothWindow=0.5,
weight=0.5,
plot=False):
'''
Event Detection (silence removal)
ARGUMENTS:
- x: the input audio signal
- fs: sampling freq
- st_win, st_step: window size and step in seconds
- smoothWindow: (optinal) smooth window (in seconds)
- weight: (optinal) weight factor (0 < weight < 1) the higher, the more strict
- plot: (optinal) True if results are to be plotted
RETURNS:
- seg_limits: list of segment limits in seconds (e.g [[0.1, 0.9], [1.4, 3.0]] means that
the resulting segments are (0.1 - 0.9) seconds and (1.4, 3.0) seconds
'''
if weight >= 1:
weight = 0.99
if weight <= 0:
weight = 0.01
# Step 1: feature extraction
x = audioBasicIO.stereo2mono(x)
st_feats, _ = aF.stFeatureExtraction(x, fs, st_win * fs, st_step * fs)
# Step 2: train binary svm classifier of low vs high energy frames
# keep only the energy short-term sequence (2nd feature)
st_energy = st_feats[1, :]
en = numpy.sort(st_energy)
# number of 10% of the total short-term windows
l1 = int(len(en) / 10)
# compute "lower" 10% energy threshold
t1 = numpy.mean(en[0:l1]) + 0.000000000000001
# compute "higher" 10% energy threshold
t2 = numpy.mean(en[-l1:-1]) + 0.000000000000001
# get all features that correspond to low energy
class1 = st_feats[:, numpy.where(st_energy <= t1)[0]]
# get all features that correspond to high energy
class2 = st_feats[:, numpy.where(st_energy >= t2)[0]]
# form the binary classification task and ...
faets_s = [class1.T, class2.T]
# normalize and train the respective svm probabilistic model
# (ONSET vs SILENCE)
[faets_s_norm, means_s, stds_s] = aT.normalizeFeatures(faets_s)
svm = aT.trainSVM(faets_s_norm, 1.0)
# Step 3: compute onset probability based on the trained svm
prob_on_set = []
for i in range(st_feats.shape[1]):
# for each frame
cur_fv = (st_feats[:, i] - means_s) / stds_s
# get svm probability (that it belongs to the ONSET class)
prob_on_set.append(svm.predict_proba(cur_fv.reshape(1, -1))[0][1])
prob_on_set = numpy.array(prob_on_set)
# smooth probability:
prob_on_set = smoothMovingAvg(prob_on_set, smoothWindow / st_step)
# Step 4A: detect onset frame indices:
prog_on_set_sort = numpy.sort(prob_on_set)
# find probability Threshold as a weighted average
# of top 10% and lower 10% of the values
Nt = int(prog_on_set_sort.shape[0] / 10)
T = (numpy.mean((1 - weight) * prog_on_set_sort[0:Nt]) +
weight * numpy.mean(prog_on_set_sort[-Nt::]))
max_idx = numpy.where(prob_on_set > T)[0]
# get the indices of the frames that satisfy the thresholding
i = 0
time_clusters = []
seg_limits = []
# Step 4B: group frame indices to onset segments
while i < len(max_idx):
# for each of the detected onset indices
cur_cluster = [max_idx[i]]
if i == len(max_idx) - 1:
break
while max_idx[i + 1] - cur_cluster[-1] <= 2:
cur_cluster.append(max_idx[i + 1])
i += 1
if i == len(max_idx) - 1:
break
i += 1
time_clusters.append(cur_cluster)
seg_limits.append(
[cur_cluster[0] * st_step, cur_cluster[-1] * st_step])
# Step 5: Post process: remove very small segments:
min_dur = 0.2
seg_limits_2 = []
for s in seg_limits:
if s[1] - s[0] > min_dur:
seg_limits_2.append(s)
seg_limits = seg_limits_2
if plot:
timeX = numpy.arange(0, x.shape[0] / float(fs), 1.0 / fs)
plt.subplot(2, 1, 1)
plt.plot(timeX, x)
for s in seg_limits:
plt.axvline(x=s[0])
plt.axvline(x=s[1])
plt.subplot(2, 1, 2)
plt.plot(numpy.arange(0, prob_on_set.shape[0] * st_step, st_step),
prob_on_set)
plt.title('Signal')
for s in seg_limits:
plt.axvline(x=s[0])
plt.axvline(x=s[1])
plt.title('svm Probability')
plt.show()
return seg_limits
def main(argv):
if argv[1] == "-dirMp3toWAV": # convert mp3 to wav (batch)
if len(argv) == 5:
path = argv[2]
if argv[3] not in ["8000", "16000", "32000", "44100"]:
print "Error. Unsupported sampling rate (must be: 8000, 16000, 32000 or 44100)."
return
if argv[4] not in ["1", "2"]:
print "Error. Number of output channels must be 1 or 2"
return
if not os.path.isdir(path):
raise Exception("Input path not found!")
useMp3TagsAsNames = True
audioBasicIO.convertDirMP3ToWav(path, int(argv[3]), int(argv[4]),
useMp3TagsAsNames)
else:
print "Error.\nSyntax: " + argv[
0] + " -dirMp3toWAV <dirName> <sampling Freq> <numOfChannels>"
if argv[1] == "-dirWAVChangeFs": # convert mp3 to wav (batch)
if len(argv) == 5:
path = argv[2]
if argv[3] not in ["8000", "16000", "32000", "44100"]:
print "Error. Unsupported sampling rate (must be: 8000, 16000, 32000 or 44100)."
return
if argv[4] not in ["1", "2"]:
print "Error. Number of output channels must be 1 or 2"
return
if not os.path.isdir(path):
raise Exception("Input path not found!")
audioBasicIO.convertFsDirWavToWav(path, int(argv[3]), int(argv[4]))
else:
print "Error.\nSyntax: " + argv[
0] + " -dirMp3toWAV <dirName> <sampling Freq> <numOfChannels>"
elif argv[
1] == "-featureExtractionFile": # short-term and mid-term feature extraction to files (csv and numpy)
if len(argv) == 7:
wavFileName = argv[2]
if not os.path.isfile(wavFileName):
raise Exception("Input audio file not found!")
if not (uT.isNum(argv[3]) and uT.isNum(argv[4])
and uT.isNum(argv[5]) and uT.isNum(argv[6])):
raise Exception(
"Mid-term and short-term window sizes and steps must be numbers!"
)
mtWin = float(argv[3])
mtStep = float(argv[4])
stWin = float(argv[5])
stStep = float(argv[6])
outFile = wavFileName
aF.mtFeatureExtractionToFile(wavFileName, mtWin, mtStep, stWin,
stStep, outFile, True, True, True)
else:
print "Error.\nSyntax: " + argv[
0] + " -featureExtractionFile <wavFileName> <mtWin> <mtStep> <stWin> <stStep>"
elif argv[1] == "-beatExtraction":
if len(argv) == 4:
wavFileName = argv[2]
if not os.path.isfile(wavFileName):
raise Exception("Input audio file not found!")
if not (uT.isNum(argv[3])):
raise Exception("PLOT must be either 0 or 1")
if not ((int(argv[3]) == 0) or (int(argv[3]) == 1)):
raise Exception("PLOT must be either 0 or 1")
[Fs, x] = audioBasicIO.readAudioFile(wavFileName)
F = aF.stFeatureExtraction(x, Fs, 0.050 * Fs, 0.050 * Fs)
BPM, ratio = aF.beatExtraction(F, 0.050, int(argv[3]) == 1)
print "Beat: {0:d} bpm ".format(int(BPM))
print "Ratio: {0:.2f} ".format(ratio)
else:
print "Error.\nSyntax: " + argv[
0] + " -beatExtraction <wavFileName> <PLOT (0 or 1)>"
elif argv[
1] == '-featureExtractionDir': # same as -featureExtractionFile, in a batch mode (i.e. for each WAV file in the provided path)
if len(argv) == 7:
path = argv[2]
if not os.path.isdir(path):
raise Exception("Input path not found!")
if not (uT.isNum(argv[3]) and uT.isNum(argv[4])
and uT.isNum(argv[5]) and uT.isNum(argv[6])):
raise Exception(
"Mid-term and short-term window sizes and steps must be numbers!"
)
mtWin = float(argv[3])
mtStep = float(argv[4])
stWin = float(argv[5])
stStep = float(argv[6])
aF.mtFeatureExtractionToFileDir(path, mtWin, mtStep, stWin, stStep,
True, True, True)
else:
print "Error.\nSyntax: " + argv[
0] + " -featureExtractionDir <path> <mtWin> <mtStep> <stWin> <stStep>"
elif argv[
1] == '-featureVisualizationDir': # visualize the content relationships between recordings stored in a folder
if len(argv) == 3:
if not os.path.isdir(argv[2]):
raise Exception("Input folder not found!")
aV.visualizeFeaturesFolder(argv[2], "pca", "")
elif argv[
1] == '-fileSpectrogram': # show spectogram of a sound stored in a file
if len(argv) == 3:
wavFileName = argv[2]
if not os.path.isfile(wavFileName):
raise Exception("Input audio file not found!")
[Fs, x] = audioBasicIO.readAudioFile(wavFileName)
x = audioBasicIO.stereo2mono(x)
specgram, TimeAxis, FreqAxis = aF.stSpectogram(
x, Fs, round(Fs * 0.040), round(Fs * 0.040), True)
else:
print "Error.\nSyntax: " + argv[0] + " -fileSpectrogram <fileName>"
elif argv[
1] == '-fileChromagram': # show spectogram of a sound stored in a file
if len(argv) == 3:
wavFileName = argv[2]
if not os.path.isfile(wavFileName):
raise Exception("Input audio file not found!")
[Fs, x] = audioBasicIO.readAudioFile(wavFileName)
x = audioBasicIO.stereo2mono(x)
specgram, TimeAxis, FreqAxis = aF.stChromagram(
x, Fs, round(Fs * 0.040), round(Fs * 0.040), True)
else:
print "Error.\nSyntax: " + argv[0] + " -fileSpectrogram <fileName>"
elif argv[1] == "-trainClassifier": # Segment classifier training (OK)
if len(argv) > 6:
method = argv[2]
beatFeatures = (int(argv[3]) == 1)
listOfDirs = argv[4:len(argv) - 1]
modelName = argv[-1]
aT.featureAndTrain(listOfDirs,
1,
1,
aT.shortTermWindow,
aT.shortTermStep,
method.lower(),
modelName,
computeBEAT=beatFeatures)
else:
print "Error.\nSyntax: " + argv[
0] + " -trainClassifier <method(svm or knn)> <beat features> <directory 1> <directory 2> ... <directory N> <modelName>"
elif argv[1] == "-trainRegression": # Segment regression model
if len(argv) == 6:
method = argv[2]
beatFeatures = (int(argv[3]) == 1)
dirName = argv[4]
modelName = argv[5]
aT.featureAndTrainRegression(dirName,
1,
1,
aT.shortTermWindow,
aT.shortTermStep,
method.lower(),
modelName,
computeBEAT=beatFeatures)
else:
print "Error.\nSyntax: " + argv[
0] + " -trainRegression <method(svm or knn)> <beat features> <directory> <modelName>"
elif argv[1] == "-classifyFile": # Single File Classification (OK)
if len(argv) == 5:
modelType = argv[2]
modelName = argv[3]
inputFile = argv[4]
if modelType not in ["svm", "knn"]:
raise Exception("ModelType has to be either svm or knn!")
if not os.path.isfile(modelName):
raise Exception("Input modelName not found!")
if not os.path.isfile(inputFile):
raise Exception("Input audio file not found!")
[Result, P,
classNames] = aT.fileClassification(inputFile, modelName,
modelType)
print "{0:s}\t{1:s}".format("Class", "Probability")
for i, c in enumerate(classNames):
print "{0:s}\t{1:.2f}".format(c, P[i])
print "Winner class: " + classNames[int(Result)]
else:
print "Error.\nSyntax: " + argv[
0] + " -classifyFile <method(svm or knn)> <modelName> <fileName>"
elif argv[1] == "-regressionFile": # Single File Classification (OK)
if len(argv) == 5:
modelType = argv[2]
modelName = argv[3]
inputFile = argv[4]
if modelType not in ["svm", "knn"]:
raise Exception("ModelType has to be either svm or knn!")
if not os.path.isfile(inputFile):
raise Exception("Input audio file not found!")
R, regressionNames = aT.fileRegression(inputFile, modelName,
modelType)
for i in range(len(R)):
print "{0:s}\t{1:.3f}".format(regressionNames[i], R[i])
#print "{0:s}\t{1:.2f}".format(c,P[i])
else:
print "Error.\nSyntax: " + argv[
0] + " -regressionFile <method(svm or knn)> <modelName> <fileName>"
elif argv[1] == "-classifyFolder": # Directory classification (Ok)
if len(argv) == 6 or len(argv) == 5:
modelType = argv[2]
modelName = argv[3]
inputFolder = argv[4]
if len(argv) == 6:
outputMode = argv[5]
else:
outputMode = "0"
if modelType not in ["svm", "knn"]:
raise Exception("ModelType has to be either svm or knn!")
if outputMode not in ["0", "1"]:
raise Exception("outputMode has to be 0 or 1")
if not os.path.isfile(modelName):
raise Exception("Input modelName not found!")
files = '*.wav'
if os.path.isdir(inputFolder):
strFilePattern = os.path.join(inputFolder, files)
else:
strFilePattern = inputFolder + files
wavFilesList = []
wavFilesList.extend(glob.glob(strFilePattern))
wavFilesList = sorted(wavFilesList)
if len(wavFilesList) == 0:
print "No WAV files found!"
return
Results = []
for wavFile in wavFilesList:
[Result, P,
classNames] = aT.fileClassification(wavFile, modelName,
modelType)
Result = int(Result)
Results.append(Result)
if outputMode == "1":
print "{0:s}\t{1:s}".format(wavFile, classNames[Result])
Results = numpy.array(Results)
# print distribution of classes:
[Histogram,
_] = numpy.histogram(Results,
bins=numpy.arange(len(classNames) + 1))
for i, h in enumerate(Histogram):
print "{0:20s}\t\t{1:d}".format(classNames[i], h)
else:
print "Error.\nSyntax: " + argv[
0] + " -classifyFolder <method(svm or knn)> <modelName> <folderName> <outputMode(0 or 1)"
elif argv[
1] == "-regressionFolder": # Regression applied on the WAV files of a folder
if len(argv) == 5:
modelType = argv[2]
modelName = argv[3]
inputFolder = argv[4]
if modelType not in ["svm", "knn"]:
raise Exception("ModelType has to be either svm or knn!")
files = '*.wav'
if os.path.isdir(inputFolder):
strFilePattern = os.path.join(inputFolder, files)
else:
strFilePattern = inputFolder + files
wavFilesList = []
wavFilesList.extend(glob.glob(strFilePattern))
wavFilesList = sorted(wavFilesList)
if len(wavFilesList) == 0:
print "No WAV files found!"
return
Results = []
for wavFile in wavFilesList:
R, regressionNames = aT.fileRegression(wavFile, modelName,
modelType)
Results.append(R)
Results = numpy.array(Results)
for i, r in enumerate(regressionNames):
[Histogram, bins] = numpy.histogram(Results[:, i])
centers = (bins[0:-1] + bins[1::]) / 2.0
plt.subplot(len(regressionNames), 1, i)
plt.plot(centers, Histogram)
plt.title(r)
plt.show()
# for h in Histogram:
# print "{0:20d}".format(h),
# if outputMode=="1":
# for i,h in enumerate(Histogram):
# print "{0:20s}\t\t{1:d}".format(classNames[i], h)
else:
print "Error.\nSyntax: " + argv[
0] + " -regressionFolder <method(svm or knn)> <modelName> <folderName>"
elif argv[1] == '-trainHMMsegmenter_fromfile':
if len(argv) == 7:
wavFile = argv[2]
gtFile = argv[3]
hmmModelName = argv[4]
if not uT.isNum(argv[5]):
print "Error: mid-term window size must be float!"
return
if not uT.isNum(argv[6]):
print "Error: mid-term window step must be float!"
return
mtWin = float(argv[5])
mtStep = float(argv[6])
if not os.path.isfile(wavFile):
print "Error: wavfile does not exist!"
return
if not os.path.isfile(gtFile):
print "Error: groundtruth does not exist!"
return
aS.trainHMM_fromFile(wavFile, gtFile, hmmModelName, mtWin, mtStep)
else:
print "Error.\nSyntax: " + argv[
0] + " -trainHMMsegmenter_fromfile <wavFilePath> <gtSegmentFilePath> <hmmModelFileName> <mtWin> <mtStep>"
elif argv[1] == '-trainHMMsegmenter_fromdir':
if len(argv) == 6:
dirPath = argv[2]
hmmModelName = argv[3]
if not uT.isNum(argv[4]):
print "Error: mid-term window size must be float!"
if not uT.isNum(argv[5]):
print "Error: mid-term window step must be float!"
mtWin = float(argv[4])
mtStep = float(argv[5])
aS.trainHMM_fromDir(dirPath, hmmModelName, mtWin, mtStep)
else:
print "Error.\nSyntax: " + argv[
0] + " -trainHMMsegmenter_fromdir <dirPath> <hmmModelFileName> <mtWin> <mtStep>"
elif argv[
1] == "-segmentClassifyFileHMM": # HMM-based segmentation-classification
if len(argv) == 4:
hmmModelName = argv[2]
wavFile = argv[3]
gtFile = wavFile.replace('.wav', '.segments')
aS.hmmSegmentation(wavFile,
hmmModelName,
PLOT=True,
gtFileName=gtFile)
else:
print "Error.\nSyntax: " + argv[
0] + " -segmentClassifyHMM <hmmModelName> <fileName>"
elif argv[
1] == '-segmentClassifyFile': # Segmentation-classification (fix-sized segment using knn or svm)
if (len(argv) == 5):
modelType = argv[2]
modelName = argv[3]
inputWavFile = argv[4]
if modelType not in ["svm", "knn"]:
raise Exception("ModelType has to be either svm or knn!")
if not os.path.isfile(modelName):
raise Exception("Input modelName not found!")
if not os.path.isfile(inputWavFile):
raise Exception("Input audio file not found!")
gtFile = inputWavFile.replace('.wav', '.segments')
aS.mtFileClassification(inputWavFile, modelName, modelType, True,
gtFile)
else:
print "Error.\nSyntax: " + argv[
0] + " -segmentClassifyFile <method(svm or knn)> <modelName> <fileName>"
elif argv[1] == "-segmentationEvaluation":
if len(argv) == 5:
methodName = argv[2]
modelName = argv[3]
dirName = argv[4]
aS.evaluateSegmentationClassificationDir(dirName, modelName,
methodName)
else:
print "Error.\nSyntax: " + argv[
0] + " -segmentationEvaluation <method(svm or knn)> <modelName> <directoryName>"
elif argv[1] == "-silenceRemoval":
if len(argv) == 5:
inputFile = argv[2]
if not os.path.isfile(inputFile):
raise Exception("Input audio file not found!")
smoothingWindow = float(argv[3])
weight = float(argv[4])
[Fs,
x] = audioBasicIO.readAudioFile(inputFile) # read audio signal
segmentLimits = aS.silenceRemoval(x, Fs, 0.05, 0.05,
smoothingWindow, weight,
False) # get onsets
for i, s in enumerate(segmentLimits):
strOut = "{0:s}_{1:.3f}-{2:.3f}.wav".format(
inputFile[0:-4], s[0], s[1])
wavfile.write(strOut, Fs, x[int(Fs * s[0]):int(Fs * s[1])])
else:
print "Error.\nSyntax: " + argv[
0] + " -silenceRemoval <inputFile> <smoothinWindow(secs)> <Threshold Weight>"
elif argv[
1] == '-speakerDiarization': # speaker diarization (from file): TODO
inputFile = argv[2]
nSpeakers = int(argv[3])
useLDA = (int(argv[4]) == 1)
if useLDA:
aS.speakerDiarization(inputFile, nSpeakers, PLOT=True)
else:
aS.speakerDiarization(inputFile, nSpeakers, LDAdim=0, PLOT=True)
#print speechLimits
elif argv[1] == "-speakerDiarizationScriptEval":
dir = argv[2]
listOfLDAs = [int(l) for l in argv[3::]]
aS.speakerDiarizationEvaluateScript(dir, listOfLDAs)
elif argv[1] == '-thumbnail': # music thumbnailing (OK)
if len(argv) == 4:
inputFile = argv[2]
stWindow = 1.0
stStep = 1.0
if not os.path.isfile(inputFile):
raise Exception("Input audio file not found!")
[Fs, x] = audioBasicIO.readAudioFile(inputFile) # read file
if Fs == -1: # could not read file
return
try:
thumbnailSize = float(argv[3])
except ValueError:
print "Thumbnail size must be a float (in seconds)"
return
[A1, A2, B1, B2, Smatrix] = aS.musicThumbnailing(
x, Fs, stWindow, stStep,
thumbnailSize) # find thumbnail endpoints
# write thumbnails to WAV files:
thumbnailFileName1 = inputFile.replace(".wav", "_thumb1.wav")
thumbnailFileName2 = inputFile.replace(".wav", "_thumb2.wav")
wavfile.write(thumbnailFileName1, Fs, x[int(Fs * A1):int(Fs * A2)])
wavfile.write(thumbnailFileName2, Fs, x[int(Fs * B1):int(Fs * B2)])
print "1st thumbnail (stored in file {0:s}): {1:4.1f}sec -- {2:4.1f}sec".format(
thumbnailFileName1, A1, A2)
print "2nd thumbnail (stored in file {0:s}): {1:4.1f}sec -- {2:4.1f}sec".format(
thumbnailFileName2, B1, B2)
# Plot self-similarity matrix:
fig = plt.figure()
ax = fig.add_subplot(111, aspect='auto')
plt.imshow(Smatrix)
# Plot best-similarity diagonal:
Xcenter = (A1 / stStep + A2 / stStep) / 2.0
Ycenter = (B1 / stStep + B2 / stStep) / 2.0
e1 = matplotlib.patches.Ellipse((Ycenter, Xcenter),
thumbnailSize * 1.4,
3,
angle=45,
linewidth=3,
fill=False)
ax.add_patch(e1)
plt.plot([B1, Smatrix.shape[0]], [A1, A1],
color='k',
linestyle='--',
linewidth=2)
plt.plot([B2, Smatrix.shape[0]], [A2, A2],
color='k',
linestyle='--',
linewidth=2)
plt.plot([B1, B1], [A1, Smatrix.shape[0]],
color='k',
linestyle='--',
linewidth=2)
plt.plot([B2, B2], [A2, Smatrix.shape[0]],
color='k',
linestyle='--',
linewidth=2)
plt.xlim([0, Smatrix.shape[0]])
plt.ylim([Smatrix.shape[1], 0])
ax.yaxis.set_label_position("right")
ax.yaxis.tick_right()
plt.xlabel('frame no')
plt.ylabel('frame no')
plt.title('Self-similarity matrix')
plt.show()
else:
print "Error.\nSyntax: " + argv[
0] + " -thumbnail <filename> <thumbnailsize(seconds)>"
import audioBasicIO
import audioFeatureExtraction
import matplotlib.pyplot as plt
import audioTrainTest as aT
plot = False
[Fs, x] = audioBasicIO.readAudioFile("data/Heavy.wav")
F = audioFeatureExtraction.stFeatureExtraction(x, Fs, 0.050 * Fs, 0.025 * Fs)
# ZCR: The rate of sign-changes of the signal during the duration of a particular frame.
if plot:
plt.subplot(2, 2, 1)
plt.plot(F[0, :])
plt.xlabel('Frame no')
plt.ylabel('ZCR')
plt.subplot(2, 2, 2)
plt.plot(F[1, :])
plt.xlabel('Frame no')
plt.ylabel('Energy')
plt.subplot(2, 2, 3)
plt.plot(F[2, :])
plt.xlabel('Frame no')
plt.ylabel('Entropy of Energy')
plt.subplot(2, 2, 4)
plt.plot(F[3, :])
plt.xlabel('Frame no')
plt.ylabel('Spectral Centroid')
plt.show()
Result, P, classNames = aT.fileClassification("data/Heavy.wav",
"data/svmMusicGenre3", "svm")
print Result
def main(argv):
if argv[1] == "-dirMp3toWAV": # convert mp3 to wav (batch)
if len(argv)==5:
path = argv[2]
if argv[3] not in ["8000", "16000", "32000", "44100"]:
print "Error. Unsupported sampling rate (must be: 8000, 16000, 32000 or 44100)."; return
if argv[4] not in ["1","2"]:
print "Error. Number of output channels must be 1 or 2"; return
if not os.path.isdir(path):
raise Exception("Input path not found!")
useMp3TagsAsNames = True
audioBasicIO.convertDirMP3ToWav(path, int(argv[3]), int(argv[4]), useMp3TagsAsNames)
else:
print "Error.\nSyntax: " + argv[0] + " -dirMp3toWAV <dirName> <sampling Freq> <numOfChannels>"
if argv[1] == "-dirWAVChangeFs": # convert mp3 to wav (batch)
if len(argv)==5:
path = argv[2]
if argv[3] not in ["8000", "16000", "32000", "44100"]:
print "Error. Unsupported sampling rate (must be: 8000, 16000, 32000 or 44100)."; return
if argv[4] not in ["1","2"]:
print "Error. Number of output channels must be 1 or 2"; return
if not os.path.isdir(path):
raise Exception("Input path not found!")
audioBasicIO.convertFsDirWavToWav(path, int(argv[3]), int(argv[4]))
else:
print "Error.\nSyntax: " + argv[0] + " -dirMp3toWAV <dirName> <sampling Freq> <numOfChannels>"
elif argv[1] == "-featureExtractionFile": # short-term and mid-term feature extraction to files (csv and numpy)
if len(argv)==7:
wavFileName = argv[2]
if not os.path.isfile(wavFileName):
raise Exception("Input audio file not found!")
if not (uT.isNum(argv[3]) and uT.isNum(argv[4]) and uT.isNum(argv[5]) and uT.isNum(argv[6])):
raise Exception("Mid-term and short-term window sizes and steps must be numbers!")
mtWin = float(argv[3])
mtStep = float(argv[4])
stWin = float(argv[5])
stStep = float(argv[6])
outFile = wavFileName
aF.mtFeatureExtractionToFile(wavFileName, mtWin, mtStep, stWin, stStep, outFile, True, True, True)
else:
print "Error.\nSyntax: " + argv[0] + " -featureExtractionFile <wavFileName> <mtWin> <mtStep> <stWin> <stStep>"
elif argv[1] == "-beatExtraction":
if len(argv)==4:
wavFileName = argv[2]
if not os.path.isfile(wavFileName):
raise Exception("Input audio file not found!")
if not (uT.isNum(argv[3])):
raise Exception("PLOT must be either 0 or 1")
if not ( (int(argv[3]) == 0) or (int(argv[3]) == 1) ):
raise Exception("PLOT must be either 0 or 1")
[Fs, x] = audioBasicIO.readAudioFile(wavFileName);
F = aF.stFeatureExtraction(x, Fs, 0.050*Fs, 0.050*Fs);
BPM, ratio = aF.beatExtraction(F, 0.050, int(argv[3])==1)
print "Beat: {0:d} bpm ".format(int(BPM))
print "Ratio: {0:.2f} ".format(ratio)
else:
print "Error.\nSyntax: " + argv[0] + " -beatExtraction <wavFileName> <PLOT (0 or 1)>"
elif argv[1] == '-featureExtractionDir': # same as -featureExtractionFile, in a batch mode (i.e. for each WAV file in the provided path)
if len(argv)==7:
path = argv[2]
if not os.path.isdir(path):
raise Exception("Input path not found!")
if not (uT.isNum(argv[3]) and uT.isNum(argv[4]) and uT.isNum(argv[5]) and uT.isNum(argv[6])):
raise Exception("Mid-term and short-term window sizes and steps must be numbers!")
mtWin = float(argv[3])
mtStep = float(argv[4])
stWin = float(argv[5])
stStep = float(argv[6])
aF.mtFeatureExtractionToFileDir(path, mtWin, mtStep, stWin, stStep, True, True, True)
else:
print "Error.\nSyntax: " + argv[0] + " -featureExtractionDir <path> <mtWin> <mtStep> <stWin> <stStep>"
elif argv[1] == '-featureVisualizationDir': # visualize the content relationships between recordings stored in a folder
if len(argv)==3:
if not os.path.isdir(argv[2]):
raise Exception("Input folder not found!")
aV.visualizeFeaturesFolder(argv[2], "pca", "")
elif argv[1] == '-fileSpectrogram': # show spectogram of a sound stored in a file
if len(argv)==3:
wavFileName = argv[2]
if not os.path.isfile(wavFileName):
raise Exception("Input audio file not found!")
[Fs, x] = audioBasicIO.readAudioFile(wavFileName)
x = audioBasicIO.stereo2mono(x)
specgram, TimeAxis, FreqAxis = aF.stSpectogram(x, Fs, round(Fs*0.040), round(Fs*0.040), True)
else:
print "Error.\nSyntax: " + argv[0] + " -fileSpectrogram <fileName>"
elif argv[1] == '-fileChromagram': # show spectogram of a sound stored in a file
if len(argv)==3:
wavFileName = argv[2]
if not os.path.isfile(wavFileName):
raise Exception("Input audio file not found!")
[Fs, x] = audioBasicIO.readAudioFile(wavFileName)
x = audioBasicIO.stereo2mono(x)
specgram, TimeAxis, FreqAxis = aF.stChromagram(x, Fs, round(Fs*0.040), round(Fs*0.040), True)
else:
print "Error.\nSyntax: " + argv[0] + " -fileSpectrogram <fileName>"
elif argv[1] == "-trainClassifier": # Segment classifier training (OK)
if len(argv)>6:
method = argv[2]
beatFeatures = (int(argv[3])==1)
listOfDirs = argv[4:len(argv)-1]
modelName = argv[-1]
aT.featureAndTrain(listOfDirs, 1, 1, aT.shortTermWindow, aT.shortTermStep, method.lower(), modelName, computeBEAT = beatFeatures)
else:
print "Error.\nSyntax: " + argv[0] + " -trainClassifier <method(svm or knn)> <beat features> <directory 1> <directory 2> ... <directory N> <modelName>"
elif argv[1] == "-trainRegression": # Segment regression model
if len(argv)==6:
method = argv[2]
beatFeatures = (int(argv[3])==1)
dirName = argv[4]
modelName = argv[5]
aT.featureAndTrainRegression(dirName, 1, 1, aT.shortTermWindow, aT.shortTermStep, method.lower(), modelName, computeBEAT = beatFeatures)
else:
print "Error.\nSyntax: " + argv[0] + " -trainRegression <method(svm or knn)> <beat features> <directory> <modelName>"
elif argv[1] == "-classifyFile": # Single File Classification (OK)
if len(argv)==5:
modelType = argv[2]
modelName = argv[3]
inputFile = argv[4]
if modelType not in ["svm", "knn"]:
raise Exception("ModelType has to be either svm or knn!")
if not os.path.isfile(modelName):
raise Exception("Input modelName not found!")
if not os.path.isfile(inputFile):
raise Exception("Input audio file not found!")
[Result, P, classNames] = aT.fileClassification(inputFile, modelName, modelType)
print "{0:s}\t{1:s}".format("Class","Probability")
for i,c in enumerate(classNames):
print "{0:s}\t{1:.2f}".format(c,P[i])
print "Winner class: " + classNames[int(Result)]
else:
print "Error.\nSyntax: " + argv[0] + " -classifyFile <method(svm or knn)> <modelName> <fileName>"
elif argv[1] == "-regressionFile": # Single File Classification (OK)
if len(argv)==5:
modelType = argv[2]
modelName = argv[3]
inputFile = argv[4]
if modelType not in ["svm", "knn"]:
raise Exception("ModelType has to be either svm or knn!")
if not os.path.isfile(inputFile):
raise Exception("Input audio file not found!")
R, regressionNames = aT.fileRegression(inputFile, modelName, modelType)
for i in range(len(R)):
print "{0:s}\t{1:.3f}".format(regressionNames[i], R[i])
#print "{0:s}\t{1:.2f}".format(c,P[i])
else:
print "Error.\nSyntax: " + argv[0] + " -regressionFile <method(svm or knn)> <modelName> <fileName>"
elif argv[1] == "-classifyFolder": # Directory classification (Ok)
if len(argv)==6 or len(argv)==5:
modelType = argv[2]
modelName = argv[3]
inputFolder = argv[4]
if len(argv)==6:
outputMode = argv[5]
else:
outputMode = "0"
if modelType not in ["svm", "knn"]:
raise Exception("ModelType has to be either svm or knn!")
if outputMode not in ["0","1"]:
raise Exception("outputMode has to be 0 or 1")
if not os.path.isfile(modelName):
raise Exception("Input modelName not found!")
files = '*.wav'
if os.path.isdir(inputFolder):
strFilePattern = os.path.join(inputFolder, files)
else:
strFilePattern = inputFolder + files
wavFilesList = []
wavFilesList.extend(glob.glob(strFilePattern))
wavFilesList = sorted(wavFilesList)
if len(wavFilesList)==0:
print "No WAV files found!"
return
Results = []
for wavFile in wavFilesList:
[Result, P, classNames] = aT.fileClassification(wavFile, modelName, modelType)
Result = int(Result)
Results.append(Result)
if outputMode=="1":
print "{0:s}\t{1:s}".format(wavFile,classNames[Result])
Results = numpy.array(Results)
# print distribution of classes:
[Histogram, _] = numpy.histogram(Results, bins=numpy.arange(len(classNames)+1))
for i,h in enumerate(Histogram):
print "{0:20s}\t\t{1:d}".format(classNames[i], h)
else:
print "Error.\nSyntax: " + argv[0] + " -classifyFolder <method(svm or knn)> <modelName> <folderName> <outputMode(0 or 1)"
elif argv[1] == "-regressionFolder": # Regression applied on the WAV files of a folder
if len(argv)==5:
modelType = argv[2]
modelName = argv[3]
inputFolder = argv[4]
if modelType not in ["svm", "knn"]:
raise Exception("ModelType has to be either svm or knn!")
files = '*.wav'
if os.path.isdir(inputFolder):
strFilePattern = os.path.join(inputFolder, files)
else:
strFilePattern = inputFolder + files
wavFilesList = []
wavFilesList.extend(glob.glob(strFilePattern))
wavFilesList = sorted(wavFilesList)
if len(wavFilesList)==0:
print "No WAV files found!"
return
Results = []
for wavFile in wavFilesList:
R, regressionNames = aT.fileRegression(wavFile, modelName, modelType)
Results.append(R)
Results = numpy.array(Results)
for i, r in enumerate(regressionNames):
[Histogram, bins] = numpy.histogram(Results[:, i])
centers = (bins[0:-1] + bins[1::]) / 2.0
plt.subplot(len(regressionNames), 1, i);
plt.plot(centers, Histogram)
plt.title(r)
plt.show()
# for h in Histogram:
# print "{0:20d}".format(h),
# if outputMode=="1":
# for i,h in enumerate(Histogram):
# print "{0:20s}\t\t{1:d}".format(classNames[i], h)
else:
print "Error.\nSyntax: " + argv[0] + " -regressionFolder <method(svm or knn)> <modelName> <folderName>"
elif argv[1] == '-trainHMMsegmenter_fromfile':
if len(argv)==7:
wavFile = argv[2]
gtFile = argv[3]
hmmModelName = argv[4]
if not uT.isNum(argv[5]):
print "Error: mid-term window size must be float!"; return
if not uT.isNum(argv[6]):
print "Error: mid-term window step must be float!"; return
mtWin = float(argv[5])
mtStep = float(argv[6])
if not os.path.isfile(wavFile):
print "Error: wavfile does not exist!"; return
if not os.path.isfile(gtFile):
print "Error: groundtruth does not exist!"; return
aS.trainHMM_fromFile(wavFile, gtFile, hmmModelName, mtWin, mtStep)
else:
print "Error.\nSyntax: " + argv[0] + " -trainHMMsegmenter_fromfile <wavFilePath> <gtSegmentFilePath> <hmmModelFileName> <mtWin> <mtStep>"
elif argv[1] == '-trainHMMsegmenter_fromdir':
if len(argv)==6:
dirPath = argv[2]
hmmModelName = argv[3]
if not uT.isNum(argv[4]):
print "Error: mid-term window size must be float!"
if not uT.isNum(argv[5]):
print "Error: mid-term window step must be float!"
mtWin = float(argv[4])
mtStep = float(argv[5])
aS.trainHMM_fromDir(dirPath, hmmModelName, mtWin, mtStep)
else:
print "Error.\nSyntax: " + argv[0] + " -trainHMMsegmenter_fromdir <dirPath> <hmmModelFileName> <mtWin> <mtStep>"
elif argv[1] == "-segmentClassifyFileHMM": # HMM-based segmentation-classification
if len(argv)==4:
hmmModelName = argv[2]
wavFile = argv[3]
gtFile = wavFile.replace('.wav', '.segments');
aS.hmmSegmentation(wavFile, hmmModelName, PLOT = True, gtFileName = gtFile)
else:
print "Error.\nSyntax: " + argv[0] + " -segmentClassifyHMM <hmmModelName> <fileName>"
elif argv[1] == '-segmentClassifyFile': # Segmentation-classification (fix-sized segment using knn or svm)
if (len(argv)==5):
modelType = argv[2]
modelName = argv[3]
inputWavFile = argv[4]
if modelType not in ["svm", "knn"]:
raise Exception("ModelType has to be either svm or knn!")
if not os.path.isfile(modelName):
raise Exception("Input modelName not found!")
if not os.path.isfile(inputWavFile):
raise Exception("Input audio file not found!")
gtFile = inputWavFile.replace('.wav', '.segments');
aS.mtFileClassification(inputWavFile, modelName, modelType, True, gtFile)
else:
print "Error.\nSyntax: " + argv[0] + " -segmentClassifyFile <method(svm or knn)> <modelName> <fileName>"
elif argv[1] == "-segmentationEvaluation":
if len(argv)==5:
methodName = argv[2]
modelName = argv[3]
dirName = argv[4]
aS.evaluateSegmentationClassificationDir(dirName, modelName, methodName)
else:
print "Error.\nSyntax: " + argv[0] + " -segmentationEvaluation <method(svm or knn)> <modelName> <directoryName>"
elif argv[1] == "-silenceRemoval":
if len(argv)==5:
inputFile = argv[2]
if not os.path.isfile(inputFile):
raise Exception("Input audio file not found!")
smoothingWindow = float(argv[3])
weight = float(argv[4])
[Fs, x] = audioBasicIO.readAudioFile(inputFile) # read audio signal
segmentLimits = aS.silenceRemoval(x, Fs, 0.05, 0.05, smoothingWindow, weight, False) # get onsets
for i, s in enumerate(segmentLimits):
strOut = "{0:s}_{1:.3f}-{2:.3f}.wav".format(inputFile[0:-4], s[0], s[1])
wavfile.write( strOut, Fs, x[int(Fs*s[0]):int(Fs*s[1])])
else:
print "Error.\nSyntax: " + argv[0] + " -silenceRemoval <inputFile> <smoothinWindow(secs)> <Threshold Weight>"
elif argv[1] == '-speakerDiarization': # speaker diarization (from file): TODO
inputFile = argv[2]
nSpeakers = int(argv[3])
useLDA = (int(argv[4])==1)
if useLDA:
aS.speakerDiarization(inputFile, nSpeakers, PLOT = True);
else:
aS.speakerDiarization(inputFile, nSpeakers, LDAdim = 0, PLOT = True);
#print speechLimits
elif argv[1] == "-speakerDiarizationScriptEval":
dir = argv[2]
listOfLDAs = [int(l) for l in argv[3::]]
aS.speakerDiarizationEvaluateScript(dir, listOfLDAs)
elif argv[1] == '-thumbnail': # music thumbnailing (OK)
if len(argv)==4:
inputFile = argv[2]
stWindow = 1.0
stStep = 1.0
if not os.path.isfile(inputFile):
raise Exception("Input audio file not found!")
[Fs, x] = audioBasicIO.readAudioFile(inputFile) # read file
if Fs == -1: # could not read file
return
try:
thumbnailSize = float(argv[3])
except ValueError:
print "Thumbnail size must be a float (in seconds)"
return
[A1, A2, B1, B2, Smatrix] = aS.musicThumbnailing(x, Fs, stWindow, stStep, thumbnailSize) # find thumbnail endpoints
# write thumbnails to WAV files:
thumbnailFileName1 = inputFile.replace(".wav","_thumb1.wav")
thumbnailFileName2 = inputFile.replace(".wav","_thumb2.wav")
wavfile.write(thumbnailFileName1, Fs, x[int(Fs*A1):int(Fs*A2)])
wavfile.write(thumbnailFileName2, Fs, x[int(Fs*B1):int(Fs*B2)])
print "1st thumbnail (stored in file {0:s}): {1:4.1f}sec -- {2:4.1f}sec".format(thumbnailFileName1, A1, A2)
print "2nd thumbnail (stored in file {0:s}): {1:4.1f}sec -- {2:4.1f}sec".format(thumbnailFileName2, B1, B2)
# Plot self-similarity matrix:
fig = plt.figure()
ax = fig.add_subplot(111, aspect='auto')
plt.imshow(Smatrix)
# Plot best-similarity diagonal:
Xcenter = (A1/stStep + A2/stStep) / 2.0
Ycenter = (B1/stStep + B2/stStep) / 2.0
e1 = matplotlib.patches.Ellipse((Ycenter, Xcenter), thumbnailSize * 1.4, 3,
angle=45, linewidth=3, fill=False)
ax.add_patch(e1)
plt.plot([B1, Smatrix.shape[0]], [A1, A1], color='k', linestyle='--', linewidth=2)
plt.plot([B2, Smatrix.shape[0]], [A2, A2], color='k', linestyle='--', linewidth=2)
plt.plot([B1, B1], [A1, Smatrix.shape[0]], color='k', linestyle='--', linewidth=2)
plt.plot([B2, B2], [A2, Smatrix.shape[0]], color='k', linestyle='--', linewidth=2)
plt.xlim([0, Smatrix.shape[0]])
plt.ylim([Smatrix.shape[1], 0])
ax.yaxis.set_label_position("right")
ax.yaxis.tick_right()
plt.xlabel('frame no')
plt.ylabel('frame no')
plt.title('Self-similarity matrix')
plt.show()
else:
print "Error.\nSyntax: " + argv[0] + " -thumbnail <filename> <thumbnailsize(seconds)>"
import random as rn
import math
import operator
import numpy as np
import audioBasicIO
import audioFeatureExtraction
dire = r"G:\5th sem\ee320 DSP\project\csv\newdata\dataset\testdata\angry"
c = 1
for filename in os.walk(dire):
for x in filename[2]:
label1 = []
label = []
file = filename[0] + "\\" + str(x)
[Fs, x] = audioBasicIO.readAudioFile(file)
F = audioFeatureExtraction.stFeatureExtraction(x, Fs, 0.50 * Fs,
0.25 * Fs)
for i in range(len(F[0])):
label1.append(3)
label.append(label1)
G = np.append(F, label, axis=0)
loc = r"G:\5th sem\ee320 DSP\project\csv\newdata\dataset\testdata\angrycsv\an" + str(
c) + ".csv"
c = c + 1
np.savetxt(loc, G, delimiter=",")
dire = r"G:\5th sem\ee320 DSP\project\csv\newdata\dataset\testdata\sad"
c = 1
for filename in os.walk(dire):
for x in filename[2]:
label = []
height = 20
k = 0
col, avg, med, std, maxm, minm = [], [], [], [], [], []
emo_labels, signal_data, filename = dirWav(
'/media/shreya/New Volume1/datasets/EMO-DB/wav/', '*.wav')
output = np.asarray(emo_labels)
features = []
feat = []
feature = []
length_features = []
for i in signal_data:
temp = audioFeatureExtraction.stFeatureExtraction(i, 16000, 1024, 1024)
print 'temp', temp.shape
avg.append(temp.mean(axis=1))
med.append(np.median(temp, axis=1))
std.append(np.std(temp, axis=1))
maxm.append(np.amax(temp, axis=1))
minm.append(np.amin(temp, axis=1))
mean = np.asarray(avg)
median = np.asarray(med)
maximum = np.asarray(maxm)
minimum = np.asarray(minm)
standard_deviation = np.asarray(std)
def reduce_zeroOneNorm(arr):
#!/usr/bin/env python2.7
import audioBasicIO
import audioFeatureExtraction
import matplotlib.pyplot as plt
[Fs_x, x] = audioBasicIO.readAudioFile("emer/1.wav")
x = audioBasicIO.stereo2mono(x)
F_x = audioFeatureExtraction.stFeatureExtraction(x, Fs_x, 0.050 * Fs_x,
0.025 * Fs_x)
[Fs_y, y] = audioBasicIO.readAudioFile("nonemer/9.wav")
y = audioBasicIO.stereo2mono(y)
F_y = audioFeatureExtraction.stFeatureExtraction(y, Fs_y, 0.050 * Fs_y,
0.025 * Fs_y)
plt.subplot(2, 1, 1)
plt.plot(F_x[0, :])
plt.xlabel('emer')
plt.ylabel('ZCR')
plt.subplot(2, 1, 2)
plt.plot(F_y[0, :])
plt.xlabel('nonemer')
plt.ylabel('ZCR')
plt.show()
plt.subplot(2, 1, 1)
plt.plot(F_x[1, :])
plt.xlabel('emer')
plt.ylabel('Energy')
plt.subplot(2, 1, 2)
plt.plot(F_y[1, :])
plt.xlabel('nonemer')
def analyzeFeatures(self, data):
values, features = aF.stFeatureExtraction(data, self.Fs, 0.1 * self.Fs, 0.1 * self.Fs)
for index, feature in enumerate(features):
print("%s:\t%s" % (feature, features[index]))
def main(argv):
if argv[1] == "-shortTerm":
for i in range(nExp):
[Fs, x] = audioBasicIO.readAudioFile("diarizationExample.wav")
duration = x.shape[0] / float(Fs)
t1 = time.clock()
F = audioFeatureExtraction.stFeatureExtraction(
x, Fs, 0.050 * Fs, 0.050 * Fs)
t2 = time.clock()
perTime1 = duration / (t2 - t1)
print("short-term feature extraction: {0:.1f} x realtime".format(perTime1))
elif argv[1] == "-classifyFile":
for i in range(nExp):
[Fs, x] = audioBasicIO.readAudioFile("diarizationExample.wav")
duration = x.shape[0] / float(Fs)
t1 = time.clock()
aT.fileClassification("diarizationExample.wav", "svmSM", "svm")
t2 = time.clock()
perTime1 = duration / (t2 - t1)
print("Mid-term feature extraction + classification \t {0:.1f} x realtime".format(perTime1))
elif argv[1] == "-mtClassify":
for i in range(nExp):
[Fs, x] = audioBasicIO.readAudioFile("diarizationExample.wav")
duration = x.shape[0] / float(Fs)
t1 = time.clock()
[flagsInd, classesAll, acc] = aS.mtFileClassification(
"diarizationExample.wav", "svmSM", "svm", False, '')
t2 = time.clock()
perTime1 = duration / (t2 - t1)
print("Fix-sized classification - segmentation \t {0:.1f} x realtime".format(perTime1))
elif argv[1] == "-hmmSegmentation":
for i in range(nExp):
[Fs, x] = audioBasicIO.readAudioFile("diarizationExample.wav")
duration = x.shape[0] / float(Fs)
t1 = time.clock()
aS.hmmSegmentation('diarizationExample.wav',
'hmmRadioSM', False, '')
t2 = time.clock()
perTime1 = duration / (t2 - t1)
print("HMM-based classification - segmentation \t {0:.1f} x realtime".format(perTime1))
elif argv[1] == "-silenceRemoval":
for i in range(nExp):
[Fs, x] = audioBasicIO.readAudioFile("diarizationExample.wav")
duration = x.shape[0] / float(Fs)
t1 = time.clock()
[Fs, x] = audioBasicIO.readAudioFile("diarizationExample.wav")
segments = aS.silenceRemoval(
x, Fs, 0.050, 0.050, smoothWindow=1.0, Weight=0.3, plot=False)
t2 = time.clock()
perTime1 = duration / (t2 - t1)
print("Silence removal \t {0:.1f} x realtime".format(perTime1))
elif argv[1] == "-thumbnailing":
for i in range(nExp):
[Fs1, x1] = audioBasicIO.readAudioFile("scottish.wav")
duration1 = x1.shape[0] / float(Fs1)
t1 = time.clock()
[A1, A2, B1, B2, Smatrix] = aS.musicThumbnailing(
x1, Fs1, 1.0, 1.0, 15.0) # find thumbnail endpoints
t2 = time.clock()
perTime1 = duration1 / (t2 - t1)
print("Thumbnail \t {0:.1f} x realtime".format(perTime1))
elif argv[1] == "-diarization-noLDA":
for i in range(nExp):
[Fs1, x1] = audioBasicIO.readAudioFile("diarizationExample.wav")
duration1 = x1.shape[0] / float(Fs1)
t1 = time.clock()
aS.speakerDiarization("diarizationExample.wav",
4, LDAdim=0, PLOT=False)
t2 = time.clock()
perTime1 = duration1 / (t2 - t1)
print("Diarization \t {0:.1f} x realtime".format(perTime1))
elif argv[1] == "-diarization-LDA":
for i in range(nExp):
[Fs1, x1] = audioBasicIO.readAudioFile("diarizationExample.wav")
duration1 = x1.shape[0] / float(Fs1)
t1 = time.clock()
aS.speakerDiarization("diarizationExample.wav", 4, PLOT=False)
t2 = time.clock()
perTime1 = duration1 / (t2 - t1)
print("Diarization \t {0:.1f} x realtime".format(perTime1))
