def transformation_synthesis(
inputFile,
fs,
hfreq,
hmag,
mYst,
freqScaling=np.array([0, 1.2, 2.01, 1.2, 2.679, .7, 3.146, .7]),
freqStretching=np.array([0, 1, 2.01, 1, 2.679, 1.5, 3.146, 1.5]),
timbrePreservation=1,
timeScaling=np.array([0, 0, 2.138, 2.138 - 1.0, 3.146, 3.146])):
"""
transform the analysis values returned by the analysis function and synthesize the sound
inputFile: name of input file
fs: sampling rate of input file
hfreq, hmag: harmonic frequencies and magnitudes
mYst: stochastic residual
freqScaling: frequency scaling factors, in time-value pairs (value of 1 no scaling)
freqStretching: frequency stretching factors, in time-value pairs (value of 1 no stretching)
timbrePreservation: 1 preserves original timbre, 0 it does not
timeScaling: time scaling factors, in time-value pairs
"""
# size of fft used in synthesis
Ns = 512
# hop size (has to be 1/4 of Ns)
H = 128
# frequency scaling of the harmonics
hfreqt, hmagt = HT.harmonicFreqScaling(hfreq, hmag, freqScaling,
freqStretching, timbrePreservation,
fs)
# time scaling the sound
yhfreq, yhmag, ystocEnv = HPST.hpsTimeScale(hfreqt, hmagt, mYst,
timeScaling)
# synthesis from the trasformed hps representation
y, yh, yst = HPS.hpsModelSynth(yhfreq, yhmag, np.array([]), ystocEnv, Ns,
H, fs)
# write output sound
outputFile = 'output_sounds/' + os.path.basename(
inputFile)[:-4] + '_hpsModelTransformation.wav'
UF.wavwrite(y, fs, outputFile)
# create figure to plot
plt.figure(figsize=(12, 6))
# frequency range to plot
maxplotfreq = 15000.0
# plot spectrogram of transformed stochastic compoment
plt.subplot(2, 1, 1)
numFrames = int(ystocEnv[:, 0].size)
sizeEnv = int(ystocEnv[0, :].size)
frmTime = H * np.arange(numFrames) / float(fs)
binFreq = (.5 * fs) * np.arange(sizeEnv * maxplotfreq /
(.5 * fs)) / sizeEnv
plt.pcolormesh(
frmTime, binFreq,
np.transpose(ystocEnv[:, :int(sizeEnv * maxplotfreq / (.5 * fs)) + 1]))
plt.autoscale(tight=True)
# plot transformed harmonic on top of stochastic spectrogram
if (yhfreq.shape[1] > 0):
harms = yhfreq * np.less(yhfreq, maxplotfreq)
harms[harms == 0] = np.nan
numFrames = int(harms[:, 0].size)
frmTime = H * np.arange(numFrames) / float(fs)
plt.plot(frmTime, harms, color='k', ms=3, alpha=1)
plt.xlabel('time (sec)')
plt.ylabel('frequency (Hz)')
plt.autoscale(tight=True)
plt.title('harmonics + stochastic spectrogram')
# plot the output sound
plt.subplot(2, 1, 2)
plt.plot(np.arange(y.size) / float(fs), y)
plt.axis([0, y.size / float(fs), min(y), max(y)])
plt.ylabel('amplitude')
plt.xlabel('time (sec)')
plt.title('output sound: y')
plt.tight_layout()
plt.show()
def transformation_synthesis(inputFile, fs, hfreq, hmag, mYst, freqScaling = np.array([0, 1.2, 2.01, 1.2, 2.679, .7, 3.146, .7]),
freqStretching = np.array([0, 1, 2.01, 1, 2.679, 1.5, 3.146, 1.5]), timbrePreservation = 1,
timeScaling = np.array([0, 0, 2.138, 2.138-1.0, 3.146, 3.146])):
"""
transform the analysis values returned by the analysis function and synthesize the sound
inputFile: name of input file
fs: sampling rate of input file
hfreq, hmag: harmonic frequencies and magnitudes
mYst: stochastic residual
freqScaling: frequency scaling factors, in time-value pairs (value of 1 no scaling)
freqStretching: frequency stretching factors, in time-value pairs (value of 1 no stretching)
timbrePreservation: 1 preserves original timbre, 0 it does not
timeScaling: time scaling factors, in time-value pairs
"""
# size of fft used in synthesis
Ns = 512
# hop size (has to be 1/4 of Ns)
H = 128
# frequency scaling of the harmonics
hfreqt, hmagt = HT.harmonicFreqScaling(hfreq, hmag, freqScaling, freqStretching, timbrePreservation, fs)
# time scaling the sound
yhfreq, yhmag, ystocEnv = HPST.hpsTimeScale(hfreqt, hmagt, mYst, timeScaling)
# synthesis from the trasformed hps representation
y, yh, yst = HPS.hpsModelSynth(yhfreq, yhmag, np.array([]), ystocEnv, Ns, H, fs)
# write output sound
outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_hpsModelTransformation.wav'
UF.wavwrite(y,fs, outputFile)
# create figure to plot
plt.figure(figsize=(12, 6))
# frequency range to plot
maxplotfreq = 15000.0
# plot spectrogram of transformed stochastic compoment
plt.subplot(2,1,1)
numFrames = int(ystocEnv[:,0].size)
sizeEnv = int(ystocEnv[0,:].size)
frmTime = H*np.arange(numFrames)/float(fs)
binFreq = (.5*fs)*np.arange(sizeEnv*maxplotfreq/(.5*fs))/sizeEnv
plt.pcolormesh(frmTime, binFreq, np.transpose(ystocEnv[:,:sizeEnv*maxplotfreq/(.5*fs)+1]))
plt.autoscale(tight=True)
# plot transformed harmonic on top of stochastic spectrogram
if (yhfreq.shape[1] > 0):
harms = yhfreq*np.less(yhfreq,maxplotfreq)
harms[harms==0] = np.nan
numFrames = int(harms[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
plt.plot(frmTime, harms, color='k', ms=3, alpha=1)
plt.xlabel('time (sec)')
plt.ylabel('frequency (Hz)')
plt.autoscale(tight=True)
plt.title('harmonics + stochastic spectrogram')
# plot the output sound
plt.subplot(2,1,2)
plt.plot(np.arange(y.size)/float(fs), y)
plt.axis([0, y.size/float(fs), min(y), max(y)])
plt.ylabel('amplitude')
plt.xlabel('time (sec)')
plt.title('output sound: y')
plt.tight_layout()
plt.show()
def transformation_synthesis_stereo(inputFile, fs, hfreq, hmag, mYst, freqScaling = np.array([0, 1.2, 2.01, 1.2, 2.679, .7, 3.146, .7]),
freqStretching = np.array([0, 1, 2.01, 1, 2.679, 1.5, 3.146, 1.5]), timbrePreservation = 1,
timeScaling = np.array([0, 0, 2.138, 2.138-1.0, 3.146, 3.146]),
inputSound=[]):
"""
transform the analysis values returned by the analysis function and synthesize the sound
inputFile: name of input file
fs: sampling rate of input file
hfreq, hmag: harmonic frequencies and magnitudes
mYst: stochastic residual
freqScaling: tuple (for L and R channels) of arrays of frequency scaling factors, in time-value pairs (value of 1 no scaling)
freqStretching: tuple (for L and R channels) of arrays of frequency stretching factors, in time-value pairs (value of 1 no stretching)
timbrePreservation: tuple (for L and R channels) of arrays of 1 preserves original timbre, 0 it does not
timeScaling: tuple (for L and R channels) of arrays of time scaling factors, in time-value pairs
inputSound: original sound, used for auto-attenuation of the output sound
"""
# size of fft used in synthesis
Ns = 512
# hop size (has to be 1/4 of Ns)
H = 128
# frequency scaling of the harmonics
hfreqtLeft , hmagtLeft = HT.harmonicFreqScaling(hfreq, hmag, freqScaling[0], freqStretching[0], timbrePreservation[0], fs)
hfreqtRight, hmagtRight = HT.harmonicFreqScaling(hfreq, hmag, freqScaling[1], freqStretching[1], timbrePreservation[1], fs)
# time scaling the sound
yhfreqLeft , yhmagLeft , ystocEnvLeft = HPST.hpsTimeScale(hfreqtLeft , hmagtLeft , mYst, timeScaling[0])
yhfreqRight, yhmagRight, ystocEnvRight = HPST.hpsTimeScale(hfreqtRight, hmagtRight, mYst, timeScaling[1])
# synthesis from the trasformed hps representation
yLeft, yhLeft, ystLeft = HPS.hpsModelSynth(yhfreqLeft , yhmagLeft , np.array([]), ystocEnvLeft , Ns, H, fs)
yRight, yhRight, ystRight = HPS.hpsModelSynth(yhfreqRight, yhmagRight, np.array([]), ystocEnvRight, Ns, H, fs)
# write output sound
outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_hpsModelTransformation.wav'
UF.wavwriteStereo(yLeft, yRight, fs, outputFile, inputSound)
# create figure to plot
plt.figure(figsize=(12, 12))
# frequency range to plot
maxplotfreq = 15000.0
def plotTransformedSignals(channelName,subplot1,subplot2,yhfreq,ystocEnv,y):
# plot spectrogram of transformed stochastic compoment
plt.subplot(subplot1)
numFrames = int(ystocEnv[:,0].size)
sizeEnv = int(ystocEnv[0,:].size)
frmTime = H*np.arange(numFrames)/float(fs)
binFreq = (.5*fs)*np.arange(sizeEnv*maxplotfreq/(.5*fs))/sizeEnv
plt.pcolormesh(frmTime, binFreq, np.transpose(ystocEnv[:,:sizeEnv*maxplotfreq/(.5*fs)+1]))
plt.autoscale(tight=True)
# plot transformed harmonic on top of stochastic spectrogram
if (yhfreq.shape[1] > 0):
harms = yhfreq*np.less(yhfreq,maxplotfreq)
harms[harms==0] = np.nan
numFrames = int(harms[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
plt.plot(frmTime, harms, color='k', ms=3, alpha=1)
plt.xlabel('time (sec)')
plt.ylabel('frequency (Hz)')
plt.autoscale(tight=True)
plt.title('harmonics + stochastic spectrogram (' + channelName + ')')
# plot the output sound
plt.subplot(subplot2)
plt.plot(np.arange(y.size)/float(fs), y)
plt.axis([0, y.size/float(fs), min(y), max(y)])
plt.ylabel('amplitude')
plt.xlabel('time (sec)')
plt.title('output sound: y (' + channelName + ')')
plotTransformedSignals('Left' , 411, 413, yhfreqLeft, ystocEnvLeft, yLeft)
plotTransformedSignals('Right', 412, 414, yhfreqRight, ystocEnvRight, yRight)
plt.tight_layout()
plt.show()
3.524,
3.609,
3.609,
4.573,
4.573,
4.7,
4.7,
5.8 - 0.4,
5.8 - 0.4,
5.902 - 0.4,
5.902 - 0.4,
8.483,
8.483,
]
)
yhfreq, yhmag, ystocEnv = HPST.hpsTimeScale(hfreqt, hmagt, mYst, timeScaling)
y, yh, yst = HPS.hpsModelSynth(yhfreq, yhmag, np.array([]), ystocEnv, Ns, H, fs)
mY, pY = STFT.stftAnal(y, fs, w, N, H)
UF.wavwrite(y, fs, "cello-phrase-weird-transformation.wav")
# plot the spectrogram of the transformed sound
plt.figure(10, figsize=(16, 4.5))
maxplotfreq = 5000.0
numFrames = int(mY[:, 0].size)
frmTime = H * np.arange(numFrames) / float(fs)
binFreq = fs * np.arange(N * maxplotfreq / fs) / N
plt.pcolormesh(frmTime, binFreq, np.transpose(mY[:, : N * maxplotfreq / fs + 1]))
plt.xlabel("time (sec)")
plt.ylabel("frequency (Hz)")
plt.title("spectrogram of transformed sound")
plt.autoscale(tight=True)
minf0 = 350
maxf0 = 700
f0et = 5
harmDevSlope = 0.01
stocf = 0.1
Ns = 512
H = 128
(fs, x) = UF.wavread(inputFile)
w = get_window(window, M)
hfreq, hmag, hphase, mYst = HPS.hpsModelAnal(x, fs, w, N, H, t, nH, minf0,
maxf0, f0et, harmDevSlope,
minSineDur, Ns, stocf)
timeScaling = np.array([0, 0, 2.138, 2.138 - 1.5, 3.146, 3.146])
yhfreq, yhmag, ystocEnv = HPST.hpsTimeScale(hfreq, hmag, mYst, timeScaling)
y, yh, yst = HPS.hpsModelSynth(yhfreq, yhmag, np.array([]), ystocEnv, Ns, H,
fs)
UF.wavwrite(y, fs, 'hps-transformation.wav')
plt.figure(figsize=(12, 9))
maxplotfreq = 14900.0
# plot the input sound
plt.subplot(4, 1, 1)
plt.plot(np.arange(x.size) / float(fs), x)
plt.axis([0, x.size / float(fs), min(x), max(x)])
plt.title('x (sax-phrase-short.wav')
