def getCepsMatrixFromData(t,y):
'''
funkcja wyznacz wspolczynniki MFCC z danych otrzymanych jako parametry
t: czas podany jako tablica kolejnych wartosci
y: sygnal podany jako tablica wartosci
'''
##########
y = sigproc.preemp(y)
fr, wordspower, wordszeros, wordsdetect, ITL ,ITU, word_fr, word_y = RecordModule.detectSingleWord(t,y)
##################
MelFeat = mealfeat.MelFeatures()
ceps_matrix = MelFeat.calcMelMatrixFeatures(word_y)
return ceps_matrix
def calcMelMatrixFeatures(self, data):
''' funkcja oblicza i zwraca macierz wspolczynnikow MFCC '''
x = sigproc.preemp(data, self.a) #self.preemph(data,self.a)
# fr, wordspower, wordszeros, wordsdetect, ITL ,ITU, word_fr, word_y = RecordModule.detectSingleWord(range(len(x)), x)
# x = word_y
self.fs = 44100.0
outTuple = self.stft(x, self.t1, self.t2, self.fs)
X = outTuple[0]
nfft = outTuple[1]
numWindows = outTuple[2]
wts = self.filtbank(self.numFilts, self.minfrq, self.maxfrq,
self.width, nfft)
Xp = pow(X, 2)
wts = pow(wts.transpose(), 2)
P = np.dot(wts, Xp)
Q = np.log(P)
C = self.dct(Q, self.numcepBasic)
C_cmn = self.cmn(C)
R_cmn = self.idct(C_cmn, 128) #second parameter is length of iDCT
d1 = self.deltas(C_cmn, self.del_w)
d2 = self.deltas(d1, self.dbl_del_w)
C_out = np.zeros((3 * self.numcepBasic, numWindows))
C_out[0:self.numcepBasic, :] = C_cmn
C_out[self.numcepBasic:2 * self.numcepBasic] = d1
C_out[2 * self.numcepBasic:3 * self.numcepBasic] = d2
# C_out[0,:] = 0
# C_out = C_out[:,0:self.numcepsBands]
# usrednienie wartosci spektrum dla danych wspolcz. w danm przedziale czasu
sizeBand = int(len(C_out.T) / self.numcepsBands)
if (sizeBand == 0):
sizeBand = 1
lpBand = 0
C_out2 = [[0 for x in range(self.numcepsBands)]
for y in range(self.numallceps)]
if (False): #srednia z przedzialu
for i in range(self.numallceps):
lpBand = 0
for j in range(len(C_out.T)):
C_out2[i][lpBand] += C_out[i][j]
if (j % sizeBand == (sizeBand - 1)):
C_out2[i][lpBand] = C_out2[i][lpBand] / sizeBand
# print("b:",lpBand)
if (lpBand < self.numcepsBands - 1):
lpBand += 1
if (True): #amplituda z przedzialu
for kk in range(self.numallceps):
for ll in range(self.numcepsBands):
# print('len',sizeBand,'ff',ll*sizeBand,ll*sizeBand+sizeBand)
amplMax = max(C_out[kk][ll * sizeBand:ll * sizeBand +
sizeBand])
amplMin = min(C_out[kk][ll * sizeBand:ll * sizeBand +
sizeBand])
if amplMax > abs(amplMin):
C_out2[kk][ll] = amplMax
else:
C_out2[kk][ll] = amplMin
return C_out2
ylabel('|Y(freq)|')
def readWav(filename, Fs):
'''
funkcja czyta plik dzwiekowy
'''
rate, data = read(filename)
y = numpy.atleast_2d(data)[0]
lungime = len(y)
timp = len(y) / Fs
t = linspace(0, timp, len(y))
return t, y
if __name__ == '__main__':
Fs = 44100.0
# sampling rate
filename = "learn_set//wlacz//6.wav"
t, y = readWav(filename, Fs)
y = sigproc.preemp(y, 0.97)
fr, wordspower, wordszeros, wordsdetect, ITL, ITU, word_fr, word_y = RecordModule.detectSingleWord(
t, y)
pylab.subplot(211)
pylab.title(filename)
pylab.plot(t, y)
pylab.subplot(212)
print(word_y.shape)
plotSpectrum(word_y, Fs)
pylab.show()
if __name__ == '__main__':
print("please speak a word into the microphone")
# filename = "learn_set//wlacz//"+str( 12 )+".wav"
filename = 'test.wav'
# record_to_file(filename)
# print("done - result written to ", filename)
# t,y = PlotModule.readWav(filename, RATE)
t,y = getSpeechFromMic()
print("done")
# t,y = getSpeechFromMic()
y = sigproc.preemp(y,0.97)
fr, wordspower, wordszeros, wordsdetect, ITL ,ITU, word_fr, word_y = detectSingleWord(t,y)
pylab.subplot(611)
pylab.title(filename)
pylab.plot(t, y, 'b')
pylab.subplot(612)
pylab.title('sygnal mocy komendy ')
pylab.plot(fr, (wordspower), 'r')
arrITL = array('f', range(len(wordspower)) )
for i in range(len(wordspower)):
arrITL[i] = ITL
def calcMelMatrixFeatures(self, data):
''' funkcja oblicza i zwraca macierz wspolczynnikow MFCC '''
x = sigproc.preemp(data, self.a)#self.preemph(data,self.a)
# fr, wordspower, wordszeros, wordsdetect, ITL ,ITU, word_fr, word_y = RecordModule.detectSingleWord(range(len(x)), x)
# x = word_y
self.fs = 44100.0
outTuple = self.stft(x,self.t1,self.t2,self.fs)
X = outTuple[0]
nfft = outTuple[1]
numWindows = outTuple[2]
wts = self.filtbank(self.numFilts, self.minfrq, self.maxfrq,
self.width, nfft)
Xp = pow(X,2)
wts = pow(wts.transpose(),2)
P = np.dot(wts,Xp)
Q = np.log(P);
C = self.dct(Q,self.numcepBasic)
C_cmn = self.cmn(C);
R_cmn = self.idct(C_cmn,128) #second parameter is length of iDCT
d1 = self.deltas(C_cmn,self.del_w)
d2 = self.deltas(d1,self.dbl_del_w)
C_out = np.zeros((3*self.numcepBasic,numWindows))
C_out[0:self.numcepBasic,:] = C_cmn
C_out[self.numcepBasic:2*self.numcepBasic] = d1
C_out[2*self.numcepBasic:3*self.numcepBasic] = d2
# C_out[0,:] = 0
# C_out = C_out[:,0:self.numcepsBands]
# usrednienie wartosci spektrum dla danych wspolcz. w danm przedziale czasu
sizeBand = int(len(C_out.T) / self.numcepsBands)
if(sizeBand==0):
sizeBand=1
lpBand = 0
C_out2 = [[0 for x in range(self.numcepsBands)] for y in range(self.numallceps)]
if(False): #srednia z przedzialu
for i in range(self.numallceps):
lpBand = 0
for j in range(len(C_out.T)):
C_out2[i][lpBand] += C_out[i][j]
if (j%sizeBand == (sizeBand-1)):
C_out2[i][lpBand] = C_out2[i][lpBand]/sizeBand
# print("b:",lpBand)
if(lpBand<self.numcepsBands-1):
lpBand+=1
if(True): #amplituda z przedzialu
for kk in range(self.numallceps):
for ll in range(self.numcepsBands) :
# print('len',sizeBand,'ff',ll*sizeBand,ll*sizeBand+sizeBand)
amplMax = max(C_out[kk][ll*sizeBand:ll*sizeBand+sizeBand])
amplMin = min(C_out[kk][ll*sizeBand:ll*sizeBand+sizeBand])
if amplMax > abs(amplMin) :
C_out2[kk][ll] = amplMax
else :
C_out2[kk][ll] = amplMin
return C_out2