def Get_Frames(wavfilepath, flength, fshift): wavFilePath = "/home/matsui-pc/matsui/sound/newsound_section01~10/" + wavfilepath wavefile = AudioFile.open(wavFilePath) # WAVファイルの読み込み # print len(wavefile) 'フレームシフトを利用する場合' frames = wavefile.frames(flength, fshift, np.hamming) # フレーム化(フレームシフトを追加したもの)(フレーム数が倍になっていたためきちんと動いている) 'フレームシフトを利用しない場合' "pymirのFrame.pyを変更" # frames = wavefile.frames(flength, np.hamming) # フレーム化 # print "############# frames ###################" # print len(frames) return frames
def calcLPCC(wav_path, frame_size=1024, lpcc_order=12):
wavData = AudioFile.open(wav_path)
lpcc_feature = []
# Decomposing into frames
# Fixed frame size
windowFunction = numpy.hamming
fixedFrames = wavData.frames(frame_size, windowFunction)
for frame_item in fixedFrames:
frame_lpcc_feature = frame_item.lpcc(lpcc_order)
drop_flag = False
for number in frame_lpcc_feature:
if math.isinf(number) or math.isnan(number):
drop_flag = True
break
if not drop_flag:
lpcc_feature.append(frame_lpcc_feature)
# print lpcc_feature
# print len(lpcc_feature)
lpcc_feature = numpy.array(lpcc_feature)
return lpcc_feature
def Get_Centroid(wavfilepath, fs, flength, fshift, freqbin): wavFilePath = "/home/matsui-pc/matsui/sound/newsound_section01~10/" + wavfilepath wavefile = AudioFile.open(wavFilePath) # WAVファイルの読み込み slength = len(wavefile) # 音声の長さ # print "******************" # print slength fnum = int(math.floor((slength - flength) / fshift + 1)) """ pymir/Frame.py """ frames = wavefile.frames(flength, fshift, np.hamming) # フレーム化 """ 関数preprocess """ preprocess_result = preprocess(wavefile, fs, flength, fshift, fnum) spectrum_complex = preprocess_result[0] spectrum_abs = preprocess_result[1] powerspec = spectrum_abs ** 2 """ 関数Centroid """ centroid = Centroid(powerspec, freqbin, fs, wavfilepath) return centroid
def visioning(name, art):
filename = "tempaudio/%s" % name
# filename = "tempaudio/test.mp3" ##for testing
audiofile = AudioFile.open(filename)
y, sr = librosa.load(filename)
tempo, beat_frames = librosa.beat.beat_track(y=y, sr=sr)
beat_times = librosa.frames_to_time(beat_frames, sr=sr)
# print tempo
# print beat_times
numberbeat = len(beat_frames)
frames = audiofile.frames(8820, numpy.hamming)
size = len(frames)
width = size / 2
height = width
# print height
# print width
visual = Image.new("RGB", (height, width), "black")
draw = ImageDraw.Draw(visual)
frameIndex = 0
startIndex = 0
# getting the chords and colors from the modified pitch module
for frame in frames:
spectrum = frame.spectrum()
chroma = spectrum.chroma()
chord, coloring, mode = Pitch.getChord(chroma)
endIndex = startIndex + len(frame)
startTime = startIndex / frame.sampleRate
endTime = endIndex / frame.sampleRate
# print startTime
start = startTime / randint(1,2)
end = endTime * randint(1,2)
defx = (height/randint(1,3) - endTime)
defy = (width/randint(1,3) - startTime)
x = (defy, startTime)
w = (endTime, defx)
# x = (int(startTime), randint(1, height))
# w = (randint(1, width), int(endTime))
# start = height / randint(1,2)
# end = width / randint(1,2)
# print "%s | %s " % (chord, coloring)
frameIndex = frameIndex + 1
startIndex = startIndex + len(frame)
if mode == 0: ## chord is minor
randomnumber = randint(0,3)
if randomnumber==0:
draw.pieslice((x, w), start, end, fill = coloring)
elif randomnumber==1:
draw.line((x, w), fill = coloring)
elif randomnumber==2:
draw.polygon((x, w), fill = coloring)
else:
draw.rectangle((x, w), fill = coloring)
elif mode == 1: ## chord is major
randomnumber = randint(0,3)
if randomnumber==0:
draw.arc((x, w), start, end, coloring)
elif randomnumber==1:
draw.chord((x, w), start, end, fill = coloring)
elif randomnumber==2:
draw.ellipse((x, w), fill = coloring)
else:
draw.point((x, w), fill = coloring)
i= 0 ## Beat creates a white point in the immage to show disruption
while i < numberbeat:
if i+1 > numberbeat:
break
x = (beat_times[i], randint(1,height))
w = ((randint(1, width), beat_times[i]))
draw.point((x, w), fill = "white")
i += 1
## Tempo of song effects smoothness of the image
tempo1= int(tempo)
visual = visual.filter(ImageFilter.EDGE_ENHANCE)
if 50<=tempo1<90:
visual = visual.filter(ImageFilter.SMOOTH_MORE)
elif 90<=tempo1<110:
visual = visual.filter(ImageFilter.SMOOTH)
elif tempo1>=130:
for i in range(0,5):
visual = visual.filter(ImageFilter.EDGE_ENHANCE)
else:
pass
path= "static/picture/%s.png" % art
visual.save(path, "PNG")
def FeatureVectorForAudioFile(filename): # jank random parameter version if version == 1: featureVector = [] wavData = AudioFile.open(filename) spectrumData = wavData.spectrum() featureVector.append(wavData.rms()) featureVector.append(spectrumData.mean()) featureVector.append(wavData.zcr()) featureVector.append(spectrumData.centroid()) featureVector.append(spectrumData.variance()) featureVector.append(spectrumData.rolloff()) mfcc = spectrumData.mfcc2() for i in range(12): featureVector.append(mfcc[i]) featureVector = NormalizeVector(featureVector) return featureVector # more thought out - frame-by-frame and mean/std of rms,pitch,mfcc if version == 2: featureVector = [] wavData = AudioFile.open(filename) windowFunction = np.hamming fixedFrames = wavData.frames(1024, windowFunction) rms = [] mfcc = [] spectralMean = [] # spectralCentroid = [] # spectralRolloff = [] # spectralSkewness = [] # spectralSpread = [] # spectralVariance = [] for frame in fixedFrames: try: spectrum = frame.spectrum() rms.append(frame.rms()) mfcc.append(spectrum.mfcc2()) spectralMean.append(spectrum.mean()) except: pass # spectralCentroid.append(spectrum.centroid()) # spectralRolloff.append(spectrum.rolloff()) # spectralSkewness.append(spectrum.skewnness()) # spectralSpread.append(spectrum.spread()) # spectralVariance.append(spectrum.variance()) # # featureVector.append(wavData.rms()) # featureVector.append(wavSpectrum.mean()) # featureVector.append(wavData.zcr()) # featureVector.append(wavSpectrum.centroid()) # featureVector.append(wavSpectrum.variance()) # featureVector.append(wavSpectrum.rolloff()) #normalize vectors rms = NormalizeVector(rms) for mfccInTime in mfcc: mfccInTime = NormalizeVector(mfccInTime) spectralMean = NormalizeVector(spectralMean) # calculate mean of all features rmsMean = np.mean(rms, axis = 0) mfccMean = np.mean(mfcc, axis = 0) spectralMeanMean = np.mean(spectralMean, axis = 0) # calculate std of all features rmsStd = np.std(rms, axis = 0) mfccStd = np.std(mfcc, axis = 0) spectralMeanStd = np.std(spectralMean, axis = 0) # add to feature vector featureVector.append(rmsMean) featureVector.append(rmsStd) featureVector.append(spectralMeanMean) featureVector.append(spectralMeanStd) for i in range(12): featureVector.append(mfccMean[i]) featureVector.append(mfccStd[i]) return featureVector
from pymir import AudioFile
import glob
import numpy as np
import librosa
import os
from python_speech_features import mfcc
wavs = []
maxSize = 949
for i in range(1, 2):
for y in range(1, 2):
dirName = 'Data/' + str(i) + '/' + str(y) + '/sorted/'
for filename in glob.glob(dirName + '*.wav'):
features = []
wavData = AudioFile.open(filename)
fixedFrames = wavData.frames(1024)
print(len(fixedFrames))
if ((len(fixedFrames)) < 350):
for frame in fixedFrames:
features.append(frame.lpcc())
featuresNP = np.array(features)
print(featuresNP.shape)
# wavs.append(featuresNP.ravel())
# print (dirName + ' '+ str(len(fixedFrames)))
# print(len(wavs))
# np.savetxt('lpc/lpcFrameLens512.csv',wavs,delimiter=',')
# print((fixedFrames[1].lpc()) ) # LPC, with order = len(fixedFrames[0])-1
# print(len(fixedFrames))
# print((fixedFrames[0].lpc()))
from pymir import AudioFile
from pymir import Onsets
from pymir import SpectralFlux
import numpy as np
import csv
feature = []
wavData = AudioFile.open("dataset_audio/9.wav")
#mp3Data = AudioFile.open("test-stereo.mp3")
fixedFrames = wavData.frames(44688)
windowFunction = np.hamming
#fixedFrames = AudioFile.frames(1024,windowFunction)
#energyOnsets = Onsets.onsetsByEnergy(wavData)
#framesFromOnsets = wavData.framesFromOnsets(energyOnsets)
#print(fixedFrames[0].cqt()) # Constant Q Transform
#print(fixedFrames[0].dct()) # Discrete Cosine Transform
#print(np.sum(fixedFrames[0].energy(windowSize = 256))) # Energy
# fixedFrames[0].play() # Playback using pyAudio
#print(np.size(fixedFrames))
#fixedFrames[0].plot() # Plot using matplotlib
#print(fixedFrames[0].rms()) # Root-mean-squared amplitude
#print(fixedFrames[0].zcr())
# Compute the spectra of each frame
i = 0
spectra = [f.spectrum() for f in fixedFrames]
for root, directories, filenames in os.walk(folderPath):
for filename in filenames:
path = os.path.join(root,filename)
if fnmatch.fnmatch(filename, '*.wav'):
job_list.append(str(path))
return job_list
job_list = make_job_list('/Users/kamal/Developpement/Explosound/features_extraction_/ExplosoundSamples-master')
analyzed = 0
audiofile_features = {}
for i in range(30):
try:
audiofile = AudioFile.open(job_list[i])
spectrum = audiofile.spectrum()
features = [audiofile.rms(), spectrum.centroid(), str(spectrum.flatness())]
audiofile_features[job_list[i]] = features
except:
pass
else:
pass
finally:
pass
rms = []
centroid = []
flatness = []
for i in audiofile_features.keys():
Currently under development
Last updated: 9 December 2012
"""
import sys
sys.path.append('..')
from pymir import AudioFile
from pymir import Energy
from pymir import Onsets
import matplotlib.pyplot as plt
filename = "../audio_files/drum_loop_01.wav"
print "Opening File: " + filename
audiofile = AudioFile.open(filename)
plt.plot(audiofile)
plt.show()
# Time-based methods
print "Finding onsets using Energy function (temporal domain)"
o = Onsets.onsetsByEnergy(audiofile)
print o
frames = audiofile.framesFromOnsets(o)
for i in range(0, len(frames)):
print "Frame " + str(i)
plt.plot(frames[i])
plt.show()
def get_features(job_list, force_rescan):
print "get_features"
if (force_rescan == True):
print ">force_rescan on " + str(len(job_list)) + " jobs"
analysed = 0
audiofile_features = {}
for i in range(len(job_list)):
try:
print "> " + str(i) + " / " + str(len(job_list)),
audiofile = AudioFile.open(job_list[i])
print ",s",
spectrum = audiofile.spectrum()
rms = audiofile.rms()
print ",rms",
centroid = spectrum.centroid
print ",centroid",
flatness = -1
try:
flatness = str(spectrum.flatness())
pass
except:
pass
finally:
pass
print ",flatness",
features = [rms, centroid, flatness]
print ",done!"
audiofile_features[job_list[i]] = features
analysed = analysed + 1
except:
#print "Exception: " + job_list[i]
pass
else:
pass
finally:
pass
print "<force_rescan " + str(analysed) + "/" + str(
len(job_list)) + " scanned."
rms = []
centroid = []
flatness = []
for i in audiofile_features.keys():
rms.append(audiofile_features[i][0])
centroid.append(audiofile_features[i][1])
flatness.append(float(audiofile_features[i][2]))
#serialisation avec pickle
#centrer et reduire
rms = ppr.scale(rms)
centroid = np.log(centroid)
m = max(centroid)
n = min(centroid)
centroid = [(c - n) / (m - n) for c in centroid]
flatness = ppr.scale(flatness)
m = max(flatness)
n = min(flatness)
flatness = [(c - n) / (m - n) for c in flatness]
filtered_features = []
for i in range(len(rms)):
filtered_features.append(
[job_list[i], rms[i], centroid[i], flatness[i]])
return filtered_features
delimiter=',',
quotechar='"',
quoting=csv.QUOTE_MINIMAL)
# header = ['File','SC_mu','SC_sigma','Bandwidth_mean', 'Bandwidth_sigma','ZCR', 'STE','Weiner']
# outwriter.writerow(header)
for root, _, files in os.walk(directory):
print "# of files: " + str(len(files))
ctr = 0
for f in files:
print str(ctr) + " :" + f
ctr += 1
fullpath = os.path.join(root, f)
row = []
row.append(f)
wavData = AudioFile.open(fullpath)
windowFunction = numpy.hamming
fixedFrames = wavData.frames(1024, windowFunction)
#
# Temporal Features
#
#
ZCR = []
for i in range(0, len(fixedFrames)):
zcr = fixedFrames[i].zcr()
if numpy.isnan(numpy.min(zcr)):
zcr = 0
ZCR.append(zcr)
def load(path): return AudioFile.open(path)
"""
from __future__ import division
import sys
sys.path.append('..')
from pymir import AudioFile
from pymir import Pitch
from pymir import Onsets
import matplotlib.pyplot as plt
# Load the audio
print "Loading Audio"
audiofile = AudioFile.open("../audio_files/test-stereo.mp3")
plt.plot(audiofile)
plt.show()
print "Finding onsets using Spectral Flux (spectral domain)"
o = Onsets.onsetsByFlux(audiofile)
print o
print "Extracting Frames"
frames = audiofile.framesFromOnsets(o)
#for i in range(0, len(frames)):
# print "Frame " + str(i)
# plt.plot(frames[i])
# plt.show()
"""
Tests of different onset detection methods
Currently under development
Last updated: 9 December 2012
"""
import sys
sys.path.append('..')
from pymir import AudioFile
from pymir import Energy
from pymir import Onsets
import matplotlib.pyplot as plt
import numpy
filename = "../audio_files/drum_loop_01.wav"
print "Opening File: " + filename
audiofile = AudioFile.open(filename)
#plt.plot(audiofile)
#plt.show()
frames = audiofile.frames(2048, numpy.hamming)
print len(frames)
def get_features(job_list, force_rescan):
print "get_features"
if(force_rescan == True):
print ">force_rescan on " + str(len(job_list)) + " jobs"
analysed = 0
audiofile_features = {}
for i in range(len(job_list)):
try:
print "> " + str(i) + " / " + str(len(job_list)),
audiofile = AudioFile.open(job_list[i])
print ",s",
spectrum = audiofile.spectrum()
rms = audiofile.rms()
print ",rms",
centroid = spectrum.centroid
print ",centroid",
flatness = -1
try:
flatness = str(spectrum.flatness())
pass
except:
pass
finally:
pass
print ",flatness",
features = [rms, centroid, flatness]
print ",done!"
audiofile_features[job_list[i]] = features
analysed = analysed + 1
except:
#print "Exception: " + job_list[i]
pass
else:
pass
finally:
pass
print "<force_rescan " + str(analysed) + "/" + str(len(job_list)) + " scanned."
rms = []
centroid = []
flatness = []
for i in audiofile_features.keys():
rms.append(audiofile_features[i][0])
centroid.append(audiofile_features[i][1])
flatness.append(float(audiofile_features[i][2]))
#serialisation avec pickle
#centrer et reduire
rms = ppr.scale(rms)
centroid = np.log(centroid)
m = max(centroid)
n = min(centroid)
centroid = [(c-n)/(m-n) for c in centroid]
flatness = ppr.scale(flatness)
m = max(flatness)
n = min(flatness)
flatness = [(c-n)/(m-n) for c in flatness]
filtered_features = []
for i in range(len(rms)):
filtered_features.append([job_list[i], rms[i], centroid[i], flatness[i]])
return filtered_features
with open('BirdsDataset.csv', 'w') as csvfile:
outwriter = csv.writer(csvfile, delimiter=',',quotechar='"', quoting=csv.QUOTE_MINIMAL)
# header = ['File','SC_mu','SC_sigma','Bandwidth_mean', 'Bandwidth_sigma','ZCR', 'STE','Weiner']
# outwriter.writerow(header)
for root, _, files in os.walk(directory):
print "# of files: " + str(len(files))
ctr = 0
for f in files:
print str(ctr) + " :" + f
ctr += 1
fullpath = os.path.join(root, f)
row = []
row.append(f)
wavData = AudioFile.open(fullpath)
windowFunction = numpy.hamming
fixedFrames = wavData.frames(1024, windowFunction)
#
# Temporal Features
#
#
ZCR = []
for i in range(0,len(fixedFrames)):
zcr = fixedFrames[i].zcr()
if numpy.isnan(numpy.min(zcr)):
prev_ste = frames[i].energy(windowSize)[0];
print "The result is written on %s" % output
try:
target = sys.argv[1]
except:
print "Usage: %s [wav file] " % sys.argv[0]
exit(0)
#TODO read from configuration file
frameSize = 882 # 20ms
freq = 44100
token = os.path.basename(target).split(".")
filename = ".".join(token[0:len(token)-1])
wav = AudioFile.open(target)
# Data set 2 - Using sliding Window
sampling = [s for s in frange(frameSize/4.0,len(wav),frameSize/4.0)]
frames = slidingWindow(wav,sampling,frameSize)
output = '%s-sampling.csv' % filename
writeCsv(sampling,frames,output)
Last updated: 9 December 2012
"""
from __future__ import division
import sys
sys.path.append('..')
from pymir import AudioFile
from pymir import Pitch
from pymir import Onsets
import matplotlib.pyplot as plt
# Load the audio
print "Loading Audio"
audiofile = AudioFile.open("../audio_files/test-stereo.mp3")
#audiofile = AudioFile.open("/Users/jsawruk/mir-samples/foo-fighters/01-The-Pretender.mp3")
plt.plot(audiofile)
plt.show()
#audiofile = audiofile[:100000]
print "Finding onsets using Spectral Flux (spectral domain)"
o = Onsets.onsetsByFlux(audiofile)
print o
print "Extracting Frames"
frames = audiofile.framesFromOnsets(o)
#for i in range(0, len(frames)):
# print "Frame " + str(i)
def load(path):
return AudioFile.open(path)
