def onset_test(noise_ffts, noise_phs, ind, mY, pY, M, H, recType, outDir, name,
fs):
"""
Synthesis of a sound using the short-time Fourier transform
mY: magnitude spectra, pY: phase spectra, M: window size, H: hop-size
returns y: output sound
"""
for i in ind:
mY[i - 3:i + 3, :] = noise_ffts[97:103, :]
pY[i - 3:i + 3, :] = noise_phs[97:103, :]
hM1 = (M + 1) // 2 # half analysis window size by rounding
hM2 = M // 2 # half analysis window size by floor
nFrames = mY[:, 0].size # number of frames
y = np.zeros(nFrames * H + hM1 + hM2) # initialize output array
pin = hM1
for i in range(nFrames): # iterate over all frames
y1 = DFT.dftSynth(mY[i, :], pY[i, :], M) # compute idft
y[pin - hM1:pin +
hM2] += H * y1 # overlap-add to generate output sound
pin += H # advance sound pointer
y = np.delete(
y,
range(hM2)) # delete half of first window which was added in stftAnal
y = np.delete(y, range(
y.size - hM1,
y.size)) # delete the end of the sound that was added in stftAnal
os.chdir('/home/tgoodall/sms-tools/software/models/Overtone_Arrays/' +
recType + '/' + outDir)
outputFile = name + '.wav'
UF.wavwrite(y, fs, outputFile)
return y
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
print("Reading WAV...")
(sampleRate, samples) = wavread(inputFile)
step = M
print("Sample Rate:",sampleRate,"kHz/16")
print("Sample Size:",len(samples))
base = os.path.basename(inputFile)
outputFile = "%s_downsampled.wav" % base.replace(".wav", "")
print("Downsampling...")
downsampled = hopSamples(samples, step)
print("Writing downsampled WAV...")
flen=len(samples)
nlen=len(downsampled)
newRate = sampleRate/(flen/nlen)
wavwrite(downsampled, sampleRate, outputFile)
print("Done:", outputFile)
def computeModel(inputFile, B, M, window = 'hanning', t = -90):
bands = range(len(B))
fs, x = UF.wavread(inputFile)
w = [get_window(window, M[i]) for i in bands]
N = (2**np.ceil(np.log2(B))).astype(int)
y_combined = SMMR.sineModelMultiRes(x, fs, w, N, t, B)
#y, y_combined = SMMR.sineModelMultiRes_combined(x, fs, w, N, t, B)
# output sound file name
outputFileInputFile = 'output_sounds/' + os.path.basename(inputFile)
#outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_sineModel.wav'
outputFile_combined = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_sineModelMultiRes.wav'
# write the synthesized sound obtained from the sinusoidal synthesis
UF.wavwrite(x, fs, outputFileInputFile)
#UF.wavwrite(y, fs, outputFile)
UF.wavwrite(y_combined, fs, outputFile_combined)
plt.figure()
plt.plot(x)
plt.plot(y_combined)
plt.show()
def main(inputFile = '../../sounds/piano.wav', window = 'hamming', M = 1024, N = 1024, H = 512): """ analysis/synthesis using the STFT inputFile: input sound file (monophonic with sampling rate of 44100) window: analysis window type (choice of rectangular, hanning, hamming, blackman, blackmanharris) M: analysis window size N: fft size (power of two, bigger or equal than M) H: hop size (at least 1/2 of analysis window size to have good overlap-add) """ # read input sound (monophonic with sampling rate of 44100) fs, x = UF.wavread(inputFile) # compute analysis window w = get_window(window, M) # compute the magnitude and phase spectrogram mX, pX = STFT.stftAnal(x, fs, w, N, H) # perform the inverse stft y = STFT.stftSynth(mX, pX, M, H) # output sound file (monophonic with sampling rate of 44100) outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_stft.wav' # write the sound resulting from the inverse stft UF.wavwrite(y, fs, outputFile) return x, fs, mX, pX, y
def getJawaab(ipFile = '../dataset/testInputs/testInput_1.wav', ipulsePos = getPulsePosFromAnn('../dataset/testInputs/testInput_1.csv'), strokeModels = None, oFile = './tablaOutput.wav', randomFlag = 1):
# If poolFeats are not built, give an error!
if strokeModels == None:
print "Train models first before calling getJawaab() ..."
opulsePos = None
strokeSeq = None
oFile = None
ts = None
else:
print "Getting jawaab..."
pulsePeriod = np.median(np.diff(ipulsePos))
print pulsePeriod
fss, audioIn = UF.wavread(ipFile)
if randomFlag == 1:
strokeSeq, tStamps, opulsePos = genRandomComposition(pulsePeriod, pieceDur = len(audioIn)/params.Fs, strokeModels = strokeModels)
else:
invCmat = getInvCovarianceMatrix(strokeModels)
strokeSeq, tStamps, opulsePos = genSimilarComposition(pulsePeriod, pieceDur = len(audioIn)/params.Fs, strokeModels = strokeModels, iAudioFile = ipFile, iPos = ipulsePos,invC = invCmat)
print strokeSeq
print tStamps
print opulsePos
if oFile != None:
audio = genAudioFromStrokeSeq(strokeModels,strokeSeq,tStamps)
audio = audio/(np.max(audio) + 0.01)
UF.wavwrite(audio, params.Fs, oFile)
return opulsePos, strokeSeq, tStamps, oFile
def transformation_synthesis(inputFile, fs, hfreq, hmag, freqScaling = np.array([0, 2.0, 1, .3]),
freqStretching = np.array([0, 1, 1, 1.5]), timbrePreservation = 1,
timeScaling = np.array([0, .0, .671, .671, 1.978, 1.978+1.0])):
# transform the analysis values returned by the analysis function and synthesize the sound
# inputFile: name of input file
# fs: sampling rate of input file
# tfreq, tmag: sinusoidal frequencies and magnitudes
# freqScaling: frequency scaling factors, in time-value pairs
# freqStretchig: frequency stretching factors, in time-value pairs
# 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
yhfreq, yhmag = HT.harmonicFreqScaling(hfreq, hmag, freqScaling, freqStretching, timbrePreservation, fs)
# time scale the sound
yhfreq, yhmag = ST.sineTimeScaling(yhfreq, yhmag, timeScaling)
# synthesis
y = SM.sineModelSynth(yhfreq, yhmag, np.array([]), Ns, H, fs)
# write output sound
outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_harmonicModelTransformation.wav'
UF.wavwrite(y,fs, outputFile)
# --------- plotting --------------------
# create figure to plot
plt.figure(figsize=(12, 6))
# frequency range to plot
maxplotfreq = 15000.0
plt.subplot(2,1,1)
# plot the transformed sinusoidal frequencies
tracks = yhfreq*np.less(yhfreq, maxplotfreq)
tracks[tracks<=0] = np.nan
numFrames = int(tracks[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
plt.plot(frmTime, tracks, color='k')
plt.title('transformed harmonic tracks')
plt.autoscale(tight=True)
# 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, tfreq, tmag, freqScaling = np.array([0, 2.0, 1, .3]),
timeScaling = np.array([0, .0, .671, .671, 1.978, 1.978+1.0])):
"""
Transform the analysis values returned by the analysis function and synthesize the sound
inputFile: name of input file; fs: sampling rate of input file
tfreq, tmag: sinusoidal frequencies and magnitudes
freqScaling: frequency scaling factors, in time-value pairs
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 sinusoidal tracks
ytfreq = ST.sineFreqScaling(tfreq, freqScaling)
# time scale the sinusoidal tracks
ytfreq, ytmag = ST.sineTimeScaling(ytfreq, tmag, timeScaling)
# synthesis
y = SM.sineModelSynth(ytfreq, ytmag, np.array([]), Ns, H, fs)
# write output sound
outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_sineModelTransformation.wav'
UF.wavwrite(y,fs, outputFile)
# create figure to plot
plt.figure(figsize=(12, 6))
# frequency range to plot
maxplotfreq = 15000.0
# plot the transformed sinusoidal frequencies
if (ytfreq.shape[1] > 0):
plt.subplot(2,1,1)
tracks = np.copy(ytfreq)
tracks = tracks*np.less(tracks, maxplotfreq)
tracks[tracks<=0] = np.nan
numFrames = int(tracks[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
plt.plot(frmTime, tracks)
plt.title('transformed sinusoidal tracks')
plt.autoscale(tight=True)
# 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 main(inputFile='../../sounds/ocean.wav', H=256, stocf=.1):
# ------- analysis parameters -------------------
# inputFile: input sound file (monophonic with sampling rate of 44100)
# H: hop size
# stocf: decimation factor used for the stochastic approximation
# --------- computation -----------------
# read input sound
(fs, x) = UF.wavread(inputFile)
# compute stochastic model
mYst = STM.stochasticModelAnal(x, H, stocf)
# synthesize sound from stochastic model
y = STM.stochasticModelSynth(mYst, H)
outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_stochasticModel.wav'
# write output sound
UF.wavwrite(y, fs, outputFile)
# --------- plotting --------------------
# create figure to plot
plt.figure(figsize=(12, 9))
# plot the input sound
plt.subplot(3,1,1)
plt.plot(np.arange(x.size)/float(fs), x)
plt.axis([0, x.size/float(fs), min(x), max(x)])
plt.ylabel('amplitude')
plt.xlabel('time (sec)')
plt.title('input sound: x')
# plot stochastic representation
plt.subplot(3,1,2)
numFrames = int(mYst[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
binFreq = np.arange(stocf*H)*float(fs)/(stocf*2*H)
plt.pcolormesh(frmTime, binFreq, np.transpose(mYst))
plt.autoscale(tight=True)
plt.xlabel('time (sec)')
plt.ylabel('frequency (Hz)')
plt.title('stochastic approximation')
# plot the output sound
plt.subplot(3,1,3)
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.tight_layout()
plt.show()
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
## Your code here
(fs, x) = wavread(inputFile)
wavwrite(inputFile + '_downsampled.wav', sampling_rate=M)
def extractHarmSpec(inputFile='../../sounds/ocean.wav',
H=256,
N=512,
stocf=.1):
"""
inputFile: input sound file (monophonic with sampling rate of 44100)
H: hop size, N: fft size
stocf: decimation factor used for the stochastic approximation (bigger than 0, maximum 1)
"""
# read input sound
(fs, x) = UF.wavread(inputFile)
# compute stochastic model
stocEnv = STM.stochasticModelAnal(x, H, N, stocf)
# synthesize sound from stochastic model
y = STM.stochasticModelSynth(stocEnv, H, N)
outputFile = 'output_sounds/' + os.path.basename(
inputFile)[:-4] + '_stochasticModel.wav'
# write output sound
UF.wavwrite(y, fs, outputFile)
# create figure to plot
plt.figure(figsize=(12, 9))
# plot the input sound
plt.subplot(3, 1, 1)
plt.plot(np.arange(x.size) / float(fs), x)
plt.axis([0, x.size / float(fs), min(x), max(x)])
plt.ylabel('amplitude')
plt.xlabel('time (sec)')
plt.title('input sound: x')
# plot stochastic representation
plt.subplot(3, 1, 2)
numFrames = int(stocEnv[:, 0].size)
frmTime = H * np.arange(numFrames) / float(fs)
binFreq = np.arange(stocf * N / 2) * float(fs) / (stocf * N)
plt.pcolormesh(frmTime, binFreq, np.transpose(stocEnv))
plt.autoscale(tight=True)
plt.xlabel('time (sec)')
plt.ylabel('frequency (Hz)')
plt.title('stochastic approximation')
# plot the output sound
plt.subplot(3, 1, 3)
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.tight_layout()
plt.show()
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
## Your code here
fs, x = wavread(inputFile)
y = hopSamples(x, M)
wavwrite(y, fs, 'test.wav')
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
fs, x = wavread(inputFile)
wavwrite(hopSamples(x, M), int(fs / M),
os.path.splitext(inputFile)[0] + "_downsampled.wav")
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
## Your code here
fs, x = wavread(inputFile)
y = hopAsmples(x, M)
wavwrite(y, fs / M, os.path.basename(inputFile)[0:-4] + '_downsampled.wav')
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
# Your code here
fs, x = wavread(inputFile)
y = hopSamples(x, M)
wavwrite(y, floor(fs / M), "result.wav")
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
## Your code here
(fs, data) = wavread(inputFile)
newData = data[::M]
wavwrite(newData, fs / M, inputFile + "_downsampled.wav")
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
(fs, x) = wavread(inputFile)
newSamples = hopSamples(x, M)
wavwrite(newSamples, int(fs / M),
os.path.basename(inputFile)[0:-4] + '_downsampled.wav')
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
## Your code here
fs, x = wavread(inputFile)
new_x = hopSamples(x, M)
wavwrite(new_x, fs // M, re.sub('.wav', '_downsampled.wav', inputFile))
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
outputFile = inputFile.rstrip('.wav') + '_downsampled.wav'
print outputFile
(sr, data) = wavread(inputFile)
downsampleData = hopSamples(data, M)
wavwrite(downsampleData, sr / M, outputFile)
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
## Your code here
fs,samples=wavread(inputFile)
downsampled=hopSamples(samples,M)
wavwrite(downsampled,fs,"{}_downsampled.wav".format(inputFile[:-4]))
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
## Your code here
fileNameParts = os.path.splitext(inputFile)
fs, x = wavread(inputFile)
wavwrite(hopSamples(x, M), fs,
fileNameParts[0] + '_downsampled' + fileNameParts[1])
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
## Your code here
samplingRate, samples = wavread(inputFile)
downsampledSamples = hopSamples(samples, M)
wavwrite(downsampledSamples, M,
inputFile.replace('.wav', '_downsampled.wav'))
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
(fs, x) = wavread(inputFile)
dsfs = fs / M
new_array = x[::M]
downsampled = inputFile.replace('.wav', '_downsampled.wav')
wavwrite(new_array, dsfs, downsampled)
def testAudio ():
samplingRate = 44100
freq = 30000
duration = 2.0
samples = np.arange(duration*samplingRate)
signal = np.sin(2*np.pi*freq*samples/samplingRate)
##print(signal)
#plt.plot(duration, signal)
#plt.show()
wavwrite(signal, samplingRate, "testa.wav")
return
def downsampleAudio(inputFile,M): (fs, x) = UF.wavread(inputFile) x.astype(int) x_array = np.array(x) x_array_slice = x_array[::M] # equivalent to: x_array_slice[0:x_array.size:M] outputFile_name = 'downsampled_' + inputFile[13:] outputFile_path = '../../sounds/output_sounds/' name_and_path = outputFile_path + outputFile_name UF.wavwrite(x_array_slice, fs, name_and_path )
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
fs, x = wavread(inputFile)
y = hopSamples(x, M)
basename, extension = inputFile.rsplit(".", 1)
outputFile = basename + "_downsampled." + extension
wavwrite(y, fs / M, outputFile)
def writeSound(y, fs, name):
'''
writes a constructed sound to a file if the sound is 16bit,
the program uses the utilFunctions module to write the sound,
otherwise, it uses the python library sound and writes at
24bits.
'''
outPutAttack = name
if fs == 44100:
UF.wavwrite(y, fs, outPutAttack)
else:
sf.write(outPutAttack, y, fs, subtype="PCM_24")
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
x = wavread(inputFile)[1]
fs = wavread(inputFile)[0]
a = hopSamples(x, M)
file_name = inputFile.replace('.wav', '_downsampled.wav')
print(file_name)
wavwrite(a, fs/M, file_name)
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
## Your code here
fs, x = wavread(inputFile)
if fs <> 44100:
print "Sample rate must be 44100."
ds = hopSamples(x, M)
wavwrite(ds, fs / M, inputFile[:-4] + "_downsampled.wav")
def main(inputFile='../../sounds/ocean.wav', H=256, N=512, stocf=.1):
"""
inputFile: input sound file (monophonic with sampling rate of 44100)
H: hop size, N: fft size
stocf: decimation factor used for the stochastic approximation (bigger than 0, maximum 1)
"""
# read input sound
(fs, x) = UF.wavread(inputFile)
# compute stochastic model
stocEnv = STM.stochasticModelAnal(x, H, N, stocf)
# synthesize sound from stochastic model
y = STM.stochasticModelSynth(stocEnv, H, N)
outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_stochasticModel.wav'
# write output sound
UF.wavwrite(y, fs, outputFile)
# create figure to plot
plt.figure(figsize=(12, 9))
# plot the input sound
plt.subplot(3,1,1)
plt.plot(np.arange(x.size)/float(fs), x)
plt.axis([0, x.size/float(fs), min(x), max(x)])
plt.ylabel('amplitude')
plt.xlabel('time (sec)')
plt.title('input sound: x')
# plot stochastic representation
plt.subplot(3,1,2)
numFrames = int(stocEnv[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
binFreq = np.arange(stocf*(N/2+1))*float(fs)/(stocf*N)
plt.pcolormesh(frmTime, binFreq, np.transpose(stocEnv))
plt.autoscale(tight=True)
plt.xlabel('time (sec)')
plt.ylabel('frequency (Hz)')
plt.title('stochastic approximation')
# plot the output sound
plt.subplot(3,1,3)
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.tight_layout()
plt.show(block=False)
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
## Your code here
(sampleRate, dataArray) = wavread(inputFile)
downSampleByM = dataArray[::M]
outputRate = sampleRate/M
wavwrite(downSampleByM, outputRate, 'test%s_downsampled.wav' %(M))
return
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
## Your code here
(fs, x) = wavread(inputFile)
##Start downsampling
x = x[::M]
##New sampling rate
fs = fs/float(M)
wavwrite(x,fs,'output_downsampled.wav')
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
## Your code here
outputFile = inputFile[0:inputFile.rfind('.')] + "_downsampled.wav"
(fs, x) = wavread(inputFile)
fs = int(fs / M)
y = hopSamples(x, M)
wavwrite(y, fs, outputFile)
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
print('Reading file...')
fs, x = wavread(inputFile)
print('Sample rate: ', fs)
print('Number of samples: ', len(x))
y = x[::M]
newFs = fs / M
wavwrite(y, newFs, 'downsampled.wav')
def main(inputFile='../../sounds/vignesh.wav',
window='blackman',
M=1201,
N=2048,
t=-90,
minSineDur=0.1,
nH=100,
minf0=130,
maxf0=300,
f0et=7,
harmDevSlope=0.01):
"""
Analysis and synthesis using the harmonic model
inputFile: input sound file (monophonic with sampling rate of 44100)
window: analysis window type (rectangular, hanning, hamming, blackman, blackmanharris)
M: analysis window size; N: fft size (power of two, bigger or equal than M)
t: magnitude threshold of spectral peaks; minSineDur: minimum duration of sinusoidal tracks
nH: maximum number of harmonics; minf0: minimum fundamental frequency in sound
maxf0: maximum fundamental frequency in sound; f0et: maximum error accepted in f0 detection algorithm
harmDevSlope: allowed deviation of harmonic tracks, higher harmonics could have higher allowed deviation
"""
# size of fft used in synthesis
Ns = 512
# hop size (has to be 1/4 of Ns)
H = 128
# read input sound
(fs, x) = UF.wavread(inputFile)
# compute analysis window
w = get_window(window, M)
# detect harmonics of input sound
hfreq, hmag, hphase = HM.harmonicModelAnal(x, fs, w, N, H, t, nH, minf0,
maxf0, f0et, harmDevSlope,
minSineDur)
# synthesize the harmonics
y = SM.sineModelSynth(hfreq, hmag, hphase, Ns, H, fs)
# output sound file (monophonic with sampling rate of 44100)
outputFile = 'output_sounds/' + os.path.basename(
inputFile)[:-4] + '_harmonicModel.wav'
# write the sound resulting from harmonic analysis
UF.wavwrite(y, fs, outputFile)
return x, fs, hfreq, y
def writeExampleFiles():
"""
A convenience function: writes out example files, some of them with optimal parameters found by exploreSineModelMultiRes()
"""
inputFile='../../sounds/orchestra.wav'
fs, x = UF.wavread(inputFile)
W = np.array(['blackmanharris'])
M = np.array([1001])
N = np.array([4096])
B = np.array([ ])
T = np.array([-90])
Ns = 512
best = Best()
y = best.calculateAndUpdate(x, fs, Ns, W, M, N, B, T)
outputFile = inputFile[:-4] + '_optimizedSineModel.wav'
print '->',outputFile
UF.wavwrite(y, fs, outputFile)
inputFile='../../sounds/121061__thirsk__160-link-strings-2-mono.wav'
fs, x = UF.wavread(inputFile)
W = np.array(['hamming','hamming','hamming'])
M = np.array([3001,1501,751])
N = np.array([16384,8192,4096])
B = np.array([2756.25,5512.5])
T = np.array([-90,-90,-90])
Ns = 512
best = Best()
y = best.calculateAndUpdate(x, fs, Ns, W, M, N, B, T)
outputFile = inputFile[:-4] + '_optimizedSineModel.wav'
print '->',outputFile
UF.wavwrite(y, fs, outputFile)
inputFile='../../sounds/orchestra.wav'
fs, x = UF.wavread(inputFile)
W = np.array(['hamming','hamming','hamming'])
M = np.array([3001,1501,751])
N = np.array([16384,8192,4096])
B = np.array([2756.25,5512.5])
T = np.array([-90,-90,-90])
Ns = 512
best = Best()
y = best.calculateAndUpdate(x, fs, Ns, W, M, N, B, T)
outputFile = inputFile[:-4] + '_nonOptimizedSineModel.wav'
print '->',outputFile
UF.wavwrite(y, fs, outputFile)
inputFile='../../sounds/121061__thirsk__160-link-strings-2-mono.wav'
fs, x = UF.wavread(inputFile)
W = np.array(['blackmanharris'])
M = np.array([1001])
N = np.array([4096])
B = np.array([ ])
T = np.array([-90])
Ns = 512
best = Best()
y = best.calculateAndUpdate(x, fs, Ns, W, M, N, B, T)
outputFile = inputFile[:-4] + '_nonOptimizedSineModel.wav'
print '->',outputFile
UF.wavwrite(y, fs, outputFile)
def downsampleAudio(inputFile, M):
"""
Inputs:
inputFile: file name of the wav file (including path)
M: downsampling factor (positive integer)
"""
fs, x = wavread(inputFile)
y = hopSamples(x, M)
dirname = os.path.dirname(inputFile)
file, ext = os.path.basename(inputFile).split('.')
outputFile = os.path.join(dirname, file + '_downsampled.' + ext)
wavwrite(y, int(fs / M), outputFile)
def main(inputFile='../../sounds/ocean.wav', H=256, N=512, stocf=.1): """ inputFile: input sound file (monophonic with sampling rate of 44100) H: hop size, N: fft size stocf: decimation factor used for the stochastic approximation (bigger than 0, maximum 1) """ # read input sound (fs, x) = UF.wavread(inputFile) # compute stochastic model stocEnv = STM.stochasticModelAnal(x, H, N, stocf) # synthesize sound from stochastic model y = STM.stochasticModelSynth(stocEnv, H, N) outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_stochasticModel.wav' # write output sound UF.wavwrite(y, fs, outputFile) return x, fs, stocEnv, y
def main(inputFile='../../sounds/bendir.wav', window='hamming', M=2001, N=2048, t=-80, minSineDur=0.02, maxnSines=150, freqDevOffset=10, freqDevSlope=0.001): """ Perform analysis/synthesis using the sinusoidal model inputFile: input sound file (monophonic with sampling rate of 44100) window: analysis window type (rectangular, hanning, hamming, blackman, blackmanharris) M: analysis window size; N: fft size (power of two, bigger or equal than M) t: magnitude threshold of spectral peaks; minSineDur: minimum duration of sinusoidal tracks maxnSines: maximum number of parallel sinusoids freqDevOffset: frequency deviation allowed in the sinusoids from frame to frame at frequency 0 freqDevSlope: slope of the frequency deviation, higher frequencies have bigger deviation """ # size of fft used in synthesis Ns = 512 # hop size (has to be 1/4 of Ns) H = 128 # read input sound fs, x = UF.wavread(inputFile) # compute analysis window w = get_window(window, M) # analyze the sound with the sinusoidal model tfreq, tmag, tphase = SM.sineModelAnal(x, fs, w, N, H, t, maxnSines, minSineDur, freqDevOffset, freqDevSlope) # synthesize the output sound from the sinusoidal representation y = SM.sineModelSynth(tfreq, tmag, tphase, Ns, H, fs) # output sound file name outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_sineModel.wav' # write the synthesized sound obtained from the sinusoidal synthesis UF.wavwrite(y, fs, outputFile) return x,fs,tfreq,y
def main(inputFile='../../sounds/vignesh.wav', window='blackman', M=1201, N=2048, t=-90, minSineDur=0.1, nH=100, minf0=130, maxf0=300, f0et=7, harmDevSlope=0.01): """ Analysis and synthesis using the harmonic model inputFile: input sound file (monophonic with sampling rate of 44100) window: analysis window type (rectangular, hanning, hamming, blackman, blackmanharris) M: analysis window size; N: fft size (power of two, bigger or equal than M) t: magnitude threshold of spectral peaks; minSineDur: minimum duration of sinusoidal tracks nH: maximum number of harmonics; minf0: minimum fundamental frequency in sound maxf0: maximum fundamental frequency in sound; f0et: maximum error accepted in f0 detection algorithm harmDevSlope: allowed deviation of harmonic tracks, higher harmonics could have higher allowed deviation """ # size of fft used in synthesis Ns = 512 # hop size (has to be 1/4 of Ns) H = 128 # read input sound (fs, x) = UF.wavread(inputFile) # compute analysis window w = get_window(window, M) # detect harmonics of input sound hfreq, hmag, hphase = HM.harmonicModelAnal(x, fs, w, N, H, t, nH, minf0, maxf0, f0et, harmDevSlope, minSineDur) # synthesize the harmonics y = SM.sineModelSynth(hfreq, hmag, hphase, Ns, H, fs) # output sound file (monophonic with sampling rate of 44100) outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_harmonicModel.wav' # write the sound resulting from harmonic analysis UF.wavwrite(y, fs, outputFile) return x,fs,hfreq,y
def main(inputFile='../../sounds/bendir.wav', window='hamming', M=2001, N=2048, t=-80, minSineDur=0.02, maxnSines=150, freqDevOffset=10, freqDevSlope=0.001): """ inputFile: input sound file (monophonic with sampling rate of 44100) window: analysis window type (rectangular, hanning, hamming, blackman, blackmanharris) M: analysis window size N: fft size (power of two, bigger or equal than M) t: magnitude threshold of spectral peaks minSineDur: minimum duration of sinusoidal tracks maxnSines: maximum number of parallel sinusoids freqDevOffset: frequency deviation allowed in the sinusoids from frame to frame at frequency 0 freqDevSlope: slope of the frequency deviation, higher frequencies have bigger deviation """ # size of fft used in synthesis Ns = 512 # hop size (has to be 1/4 of Ns) H = 128 # read input sound (fs, x) = UF.wavread(inputFile) # compute analysis window w = get_window(window, M) # perform sinusoidal plus residual analysis tfreq, tmag, tphase, xr = SPR.sprModelAnal(x, fs, w, N, H, t, minSineDur, maxnSines, freqDevOffset, freqDevSlope) # compute spectrogram of residual mXr, pXr = STFT.stftAnal(xr, fs, w, N, H) # sum sinusoids and residual y, ys = SPR.sprModelSynth(tfreq, tmag, tphase, xr, Ns, H, fs) # output sound file (monophonic with sampling rate of 44100) outputFileSines = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_sprModel_sines.wav' outputFileResidual = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_sprModel_residual.wav' outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_sprModel.wav' # write sounds files for sinusoidal, residual, and the sum UF.wavwrite(ys, fs, outputFileSines) UF.wavwrite(xr, fs, outputFileResidual) UF.wavwrite(y, fs, outputFile) return x, fs, mXr, tfreq, y
def main(inputFile='../../sounds/sax-phrase-short.wav', window='blackman', M=601, N=1024, t=-100, minSineDur=0.1, nH=100, minf0=350, maxf0=700, f0et=5, harmDevSlope=0.01): """ Perform analysis/synthesis using the harmonic plus residual model inputFile: input sound file (monophonic with sampling rate of 44100) window: analysis window type (rectangular, hanning, hamming, blackman, blackmanharris) M: analysis window size; N: fft size (power of two, bigger or equal than M) t: magnitude threshold of spectral peaks; minSineDur: minimum duration of sinusoidal tracks nH: maximum number of harmonics; minf0: minimum fundamental frequency in sound maxf0: maximum fundamental frequency in sound; f0et: maximum error accepted in f0 detection algorithm harmDevSlope: allowed deviation of harmonic tracks, higher harmonics have higher allowed deviation """ # size of fft used in synthesis Ns = 512 # hop size (has to be 1/4 of Ns) H = 128 # read input sound (fs, x) = UF.wavread(inputFile) # compute analysis window w = get_window(window, M) # find harmonics and residual hfreq, hmag, hphase, xr = HPR.hprModelAnal(x, fs, w, N, H, t, minSineDur, nH, minf0, maxf0, f0et, harmDevSlope) # compute spectrogram of residual mXr, pXr = STFT.stftAnal(xr, fs, w, N, H) # synthesize hpr model y, yh = HPR.hprModelSynth(hfreq, hmag, hphase, xr, Ns, H, fs) # output sound file (monophonic with sampling rate of 44100) outputFileSines = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_hprModel_sines.wav' outputFileResidual = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_hprModel_residual.wav' outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_hprModel.wav' # write sounds files for harmonics, residual, and the sum UF.wavwrite(yh, fs, outputFileSines) UF.wavwrite(xr, fs, outputFileResidual) UF.wavwrite(y, fs, outputFile) return x, fs, mXr,hfreq, y
def main(inputFile='../../sounds/sax-phrase-short.wav', window='blackman', M=601, N=1024, t=-100, minSineDur=0.1, nH=100, minf0=350, maxf0=700, f0et=5, harmDevSlope=0.01, stocf=0.1): """ inputFile: input sound file (monophonic with sampling rate of 44100) window: analysis window type (rectangular, hanning, hamming, blackman, blackmanharris) M: analysis window size; N: fft size (power of two, bigger or equal than M) t: magnitude threshold of spectral peaks; minSineDur: minimum duration of sinusoidal tracks nH: maximum number of harmonics; minf0: minimum fundamental frequency in sound maxf0: maximum fundamental frequency in sound; f0et: maximum error accepted in f0 detection algorithm harmDevSlope: allowed deviation of harmonic tracks, higher harmonics have higher allowed deviation stocf: decimation factor used for the stochastic approximation """ # size of fft used in synthesis Ns = 512 # hop size (has to be 1/4 of Ns) H = 128 # read input sound (fs, x) = UF.wavread(inputFile) # compute analysis window w = get_window(window, M) # compute the harmonic plus stochastic model of the whole sound hfreq, hmag, hphase, stocEnv = HPS.hpsModelAnal(x, fs, w, N, H, t, nH, minf0, maxf0, f0et, harmDevSlope, minSineDur, Ns, stocf) # synthesize a sound from the harmonic plus stochastic representation y, yh, yst = HPS.hpsModelSynth(hfreq, hmag, hphase, stocEnv, Ns, H, fs) # output sound file (monophonic with sampling rate of 44100) outputFileSines = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_hpsModel_sines.wav' outputFileStochastic = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_hpsModel_stochastic.wav' outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_hpsModel.wav' # write sounds files for harmonics, stochastic, and the sum UF.wavwrite(yh, fs, outputFileSines) UF.wavwrite(yst, fs, outputFileStochastic) UF.wavwrite(y, fs, outputFile) return x, fs, hfreq, stocEnv, y
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')
# plot spectrogram stochastic compoment
plt.subplot(4,1,2)
numFrames = int(mYst[:,0].size)
def analysis(inputFile='../../sounds/sax-phrase-short.wav', window='blackman', M=601, N=1024, t=-100,
minSineDur=0.1, nH=100, minf0=350, maxf0=700, f0et=5, harmDevSlope=0.01, stocf=0.1):
"""
Analyze a sound with the harmonic plus stochastic model
inputFile: input sound file (monophonic with sampling rate of 44100)
window: analysis window type (rectangular, hanning, hamming, blackman, blackmanharris)
M: analysis window size
N: fft size (power of two, bigger or equal than M)
t: magnitude threshold of spectral peaks
minSineDur: minimum duration of sinusoidal tracks
nH: maximum number of harmonics
minf0: minimum fundamental frequency in sound
maxf0: maximum fundamental frequency in sound
f0et: maximum error accepted in f0 detection algorithm
harmDevSlope: allowed deviation of harmonic tracks, higher harmonics have higher allowed deviation
stocf: decimation factor used for the stochastic approximation
returns inputFile: input file name; fs: sampling rate of input file,
hfreq, hmag: harmonic frequencies, magnitude; mYst: stochastic residual
"""
# size of fft used in synthesis
Ns = 512
# hop size (has to be 1/4 of Ns)
H = 128
# read input sound
(fs, x) = UF.wavread(inputFile)
# compute analysis window
w = get_window(window, M)
# compute the harmonic plus stochastic model of the whole sound
hfreq, hmag, hphase, mYst = HPS.hpsModelAnal(x, fs, w, N, H, t, nH, minf0, maxf0, f0et, harmDevSlope, minSineDur, Ns, stocf)
# synthesize the harmonic plus stochastic model without original phases
y, yh, yst = HPS.hpsModelSynth(hfreq, hmag, np.array([]), mYst, Ns, H, fs)
# write output sound
outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_hpsModel.wav'
UF.wavwrite(y,fs, outputFile)
# create figure to plot
plt.figure(figsize=(12, 9))
# frequency range to plot
maxplotfreq = 15000.0
# plot the input sound
plt.subplot(3,1,1)
plt.plot(np.arange(x.size)/float(fs), x)
plt.axis([0, x.size/float(fs), min(x), max(x)])
plt.ylabel('amplitude')
plt.xlabel('time (sec)')
plt.title('input sound: x')
# plot spectrogram stochastic compoment
plt.subplot(3,1,2)
numFrames = int(mYst[:,0].size)
sizeEnv = int(mYst[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(mYst[:,:sizeEnv*maxplotfreq/(.5*fs)+1]))
plt.autoscale(tight=True)
# plot harmonic on top of stochastic spectrogram
if (hfreq.shape[1] > 0):
harms = hfreq*np.less(hfreq,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(3,1,3)
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(block=False)
return inputFile, fs, hfreq, hmag, mYst
def exploreSineModelMultiRes(inputFile='../../sounds/orchestra.wav'):
"""
inputFile (string) = wav file including the path
"""
fs, x = UF.wavread(inputFile) # read input sound
# First, let's check whether the new code returns same result as old one for mono-resolution case
verifySineModelMultiRes()
# Let's find optimal parameters in a reasonable range
windows =['hanning', 'hamming', 'blackman', 'blackmanharris']
best = Best()
for k in range(5,80,5):
m = k * 100 + 1 # Window size in samples
for window in windows: # Window type
for t in range(-90,-100,-10): # Threshold
for Ns in [512]: # size of fft used in synthesis
n = 2
while n < m: n = n * 2 # size of fft used in analysis
for nPower in range(0,3): # try out the analysis window closest to window size, and some larger ones
for nAdditionalResolutions in range(0,4): # try out multi-resolution analysis windows
W = np.array([window])
M = np.array([m])
N = np.array([n])
B = np.array([ ])
T = np.array([t])
log_m = np.log(float(m))
log_n = np.log(float(n))
log_f = np.log(fs/2.0)
log_step = np.log(2)
executeStep = True
continueAddingResolutions = True
for additionalResolution in range(0,nAdditionalResolutions):
if continueAddingResolutions:
scaledM = int(np.exp(log_m - log_step*(additionalResolution+1)))
if scaledM % 2 == 0: scaledM = scaledM + 1
scaledN = int(np.exp(log_n - log_step*(additionalResolution+1)))
if scaledN < scaledM: scaledN = scaledM
appropriateScaledN = 2
while appropriateScaledN < scaledN: appropriateScaledN = appropriateScaledN * 2
frequencyBoundary = np.exp(log_f - (log_step*(nAdditionalResolutions - additionalResolution)))
if scaledM < Ns:
continueAddingResolutions = False
if additionalResolution == 0: executeStep = False
else:
W = np.append(W,window)
M = np.append(M,scaledM)
N = np.append(N,appropriateScaledN)
B = np.append(B,frequencyBoundary)
T = np.append(T,t)
if executeStep:
best.calculateAndUpdate(x, fs, Ns, W, M, N, B, T)
n = n * 2
print 'FILE:',inputFile
print 'BEST:','diff =',best.diff,'for W =',best.W,', M =',best.M,', N =',best.N,', B =',best.B,', T =',best.T,', Ns =',best.Ns
y_best = best.calculateAndUpdate(x, fs, best.Ns, best.W, best.M, best.N, best.B, best.T)
outputFile = inputFile[:-4] + '_optimizedSineModel.wav'
UF.wavwrite(y_best, fs, outputFile)
# compute the FO and the harmonics
t = -97
minf0 = 310
maxf0 = 450
f0et = 4
nH = 70
harmDevSlope = 0.01
Ns = H * 4
minSineDur = 0.3
hfreq, hmag, hphase = HM.harmonicModelAnal(x, fs, w, N, H, t, nH, minf0, maxf0, f0et, harmDevSlope, minSineDur)
hfreqt = copy.copy(hfreq)
hfreqt[:, 1:] = 0
yf0 = 4 * SM.sineModelSynth(hfreqt, hmag, hphase, Ns, H, fs)
yh = SM.sineModelSynth(hfreq, hmag, hphase, Ns, H, fs)
UF.wavwrite(yf0, fs, "cello-phrase-f0.wav")
UF.wavwrite(yh, fs, "cello-phrase-harmonics.wav")
# plot the F0 on top of the spectrogram
plt.figure(3, figsize=(16, 4.5))
maxplotfreq = 5000.0
harms = hfreq * np.less(hfreq, maxplotfreq)
harms[harms[:, 0] == 0] = np.nan
numFrames = int(mX[:, 0].size)
frmTime = H * np.arange(numFrames) / float(fs)
binFreq = fs * np.arange(N * maxplotfreq / fs) / N
plt.pcolormesh(frmTime, binFreq, np.transpose(mX[:, : N * maxplotfreq / fs + 1]))
plt.plot(frmTime, harms[:, 0], linewidth=3, color="0")
plt.xlabel("time (sec)")
plt.ylabel("frequency (Hz)")
plt.title("spectrogram + fundamental frequency")
def main(inputFile='../../sounds/sax-phrase-short.wav', window='blackman', M=601, N=1024, t=-100,
minSineDur=0.1, nH=100, minf0=350, maxf0=700, f0et=5, harmDevSlope=0.01):
"""
Perform analysis/synthesis using the harmonic plus residual model
inputFile: input sound file (monophonic with sampling rate of 44100)
window: analysis window type (rectangular, hanning, hamming, blackman, blackmanharris)
M: analysis window size; N: fft size (power of two, bigger or equal than M)
t: magnitude threshold of spectral peaks; minSineDur: minimum duration of sinusoidal tracks
nH: maximum number of harmonics; minf0: minimum fundamental frequency in sound
maxf0: maximum fundamental frequency in sound; f0et: maximum error accepted in f0 detection algorithm
harmDevSlope: allowed deviation of harmonic tracks, higher harmonics have higher allowed deviation
"""
# size of fft used in synthesis
Ns = 512
# hop size (has to be 1/4 of Ns)
H = 128
# read input sound
(fs, x) = UF.wavread(inputFile)
# compute analysis window
w = get_window(window, M)
# find harmonics and residual
hfreq, hmag, hphase, xr = HPR.hprModelAnal(x, fs, w, N, H, t, minSineDur, nH, minf0, maxf0, f0et, harmDevSlope)
# compute spectrogram of residual
mXr, pXr = STFT.stftAnal(xr, w, N, H)
# synthesize hpr model
y, yh = HPR.hprModelSynth(hfreq, hmag, hphase, xr, Ns, H, fs)
# output sound file (monophonic with sampling rate of 44100)
outputFileSines = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_hprModel_sines.wav'
outputFileResidual = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_hprModel_residual.wav'
outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_hprModel.wav'
# write sounds files for harmonics, residual, and the sum
UF.wavwrite(yh, fs, outputFileSines)
UF.wavwrite(xr, fs, outputFileResidual)
UF.wavwrite(y, fs, outputFile)
# create figure to plot
plt.figure(figsize=(12, 9))
# frequency range to plot
maxplotfreq = 5000.0
# plot the input sound
plt.subplot(3,1,1)
plt.plot(np.arange(x.size)/float(fs), x)
plt.axis([0, x.size/float(fs), min(x), max(x)])
plt.ylabel('amplitude')
plt.xlabel('time (sec)')
plt.title('input sound: x')
# plot the magnitude spectrogram of residual
plt.subplot(3,1,2)
maxplotbin = int(N*maxplotfreq/fs)
numFrames = int(mXr[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
binFreq = np.arange(maxplotbin+1)*float(fs)/N
plt.pcolormesh(frmTime, binFreq, np.transpose(mXr[:,:maxplotbin+1]))
plt.autoscale(tight=True)
# plot harmonic frequencies on residual spectrogram
if (hfreq.shape[1] > 0):
harms = hfreq*np.less(hfreq,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(s)')
plt.ylabel('frequency(Hz)')
plt.autoscale(tight=True)
plt.title('harmonics + residual spectrogram')
# plot the output sound
plt.subplot(3,1,3)
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.ion()
plt.show()
binFreq = np.arange(maxplotbin+1)*float(fs)/N
plt.pcolormesh(frmTime, binFreq, np.transpose(mX[:,:maxplotbin+1]))
plt.autoscale(tight=True)
plt.subplot(4,1,3)
numFrames = int(ytfreq[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
tracks = ytfreq*np.less(ytfreq, maxplotfreq)
tracks[tracks<=0] = np.nan
plt.plot(frmTime, tracks, color='k', lw=1)
plt.autoscale(tight=True)
plt.title('mY + time-scaled sine frequencies')
maxplotbin = int(N*maxplotfreq/fs)
numFrames = int(mY[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
binFreq = np.arange(maxplotbin+1)*float(fs)/N
plt.pcolormesh(frmTime, binFreq, np.transpose(mY[:,:maxplotbin+1]))
plt.autoscale(tight=True)
plt.subplot(4,1,4)
plt.plot(np.arange(y.size)/float(fs), y, 'b')
plt.axis([0,y.size/float(fs),min(y),max(y)])
plt.title('y')
plt.tight_layout()
UF.wavwrite(y, fs, 'mridangam-sineModelTimeScale.wav')
plt.savefig('sineModelTimeScale-mridangam.png')
plt.show()
plt.subplot(311)
numFrames = int(mX[:,0].size)
frmTime = H1*np.arange(numFrames)/float(fs)
binFreq = fs*np.arange(N1*maxplotfreq/fs)/N1
plt.pcolormesh(frmTime, binFreq, np.transpose(mX[:,:int(N1*maxplotfreq/fs+1)]))
plt.title('mX (orchestra.wav)')
plt.autoscale(tight=True)
plt.subplot(312)
numFrames = int(mX2[:,0].size)
frmTime = H1*np.arange(numFrames)/float(fs)
N = 2*mX2[0,:].size
binFreq = fs*np.arange(N*maxplotfreq/fs)/N
plt.pcolormesh(frmTime, binFreq, np.transpose(mX2[:,:int(N*maxplotfreq/fs+1)]))
plt.title('mX2 (speech-male.wav)')
plt.autoscale(tight=True)
plt.subplot(313)
numFrames = int(mY[:,0].size)
frmTime = H1*np.arange(numFrames)/float(fs)
binFreq = fs*np.arange(N1*maxplotfreq/fs)/N1
plt.pcolormesh(frmTime, binFreq, np.transpose(mY[:,:int(N1*maxplotfreq/fs+1)]))
plt.title('mY')
plt.autoscale(tight=True)
plt.tight_layout()
UF.wavwrite(y, fs, 'orchestra-speech-stftMorph.wav')
plt.savefig('stftMorph-orchestra.png')
plt.show()
def main(inputFile='../../sounds/bendir.wav', window='hamming', M=2001, N=2048, t=-80, minSineDur=0.02,
maxnSines=150, freqDevOffset=10, freqDevSlope=0.001):
"""
Perform analysis/synthesis using the sinusoidal model
inputFile: input sound file (monophonic with sampling rate of 44100)
window: analysis window type (rectangular, hanning, hamming, blackman, blackmanharris)
M: analysis window size; N: fft size (power of two, bigger or equal than M)
t: magnitude threshold of spectral peaks; minSineDur: minimum duration of sinusoidal tracks
maxnSines: maximum number of parallel sinusoids
freqDevOffset: frequency deviation allowed in the sinusoids from frame to frame at frequency 0
freqDevSlope: slope of the frequency deviation, higher frequencies have bigger deviation
"""
# size of fft used in synthesis
Ns = 512
# hop size (has to be 1/4 of Ns)
H = 128
# read input sound
fs, x = UF.wavread(inputFile)
# compute analysis window
w = get_window(window, M)
# analyze the sound with the sinusoidal model
tfreq, tmag, tphase = SM.sineModelAnal(x, fs, w, N, H, t, maxnSines, minSineDur, freqDevOffset, freqDevSlope)
# synthesize the output sound from the sinusoidal representation
y = SM.sineModelSynth(tfreq, tmag, tphase, Ns, H, fs)
# output sound file name
outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_sineModel.wav'
# write the synthesized sound obtained from the sinusoidal synthesis
UF.wavwrite(y, fs, outputFile)
# create figure to show plots
plt.figure(figsize=(12, 9))
# frequency range to plot
maxplotfreq = 5000.0
# plot the input sound
plt.subplot(3,1,1)
plt.plot(np.arange(x.size)/float(fs), x)
plt.axis([0, x.size/float(fs), min(x), max(x)])
plt.ylabel('amplitude')
plt.xlabel('time (sec)')
plt.title('input sound: x')
# plot the sinusoidal frequencies
plt.subplot(3,1,2)
if (tfreq.shape[1] > 0):
numFrames = tfreq.shape[0]
frmTime = H*np.arange(numFrames)/float(fs)
tfreq[tfreq<=0] = np.nan
plt.plot(frmTime, tfreq)
plt.axis([0, x.size/float(fs), 0, maxplotfreq])
plt.title('frequencies of sinusoidal tracks')
# plot the output sound
plt.subplot(3,1,3)
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(block=False)
def estimateF0(inputFile = '../../sounds/cello-double-2.wav'):
"""
Function to estimate fundamental frequency (f0) in an audio signal. This function also plots the
f0 contour on the spectrogram and synthesize the f0 contour.
Input:
inputFile (string): wav file including the path
Output:
f0 (numpy array): array of the estimated fundamental frequency (f0) values
"""
### Change these analysis parameter values
window = "blackman"
M = 4401
N = 8192
f0et = 7
t = -90.0
minf0 = 140
maxf0 = 210
### Do not modify the code below
H = 256 #fix hop size
fs, x = UF.wavread(inputFile) #reading inputFile
w = get_window(window, M) #obtaining analysis window
f0 = HM.f0Detection(x, fs, w, N, H, t, minf0, maxf0, f0et) #estimating F0
startFrame = np.floor(0.5*fs/H)
endFrame = np.ceil(4.0*fs/H)
f0[:startFrame] = 0
f0[endFrame:] = 0
y = UF.sinewaveSynth(f0, 0.8, H, fs)
UF.wavwrite(y, fs, 'synthF0Contour.wav')
## Code for plotting the f0 contour on top of the spectrogram
# frequency range to plot
maxplotfreq = 500.0
fontSize = 16
plot = 1 # plot = 1 plots the f0 contour, otherwise saves it to a file.
fig = plt.figure()
ax = fig.add_subplot(111)
mX, pX = stft.stftAnal(x, fs, w, N, H) #using same params as used for analysis
mX = np.transpose(mX[:,:int(N*(maxplotfreq/fs))+1])
timeStamps = np.arange(mX.shape[1])*H/float(fs)
binFreqs = np.arange(mX.shape[0])*fs/float(N)
plt.pcolormesh(timeStamps, binFreqs, mX)
plt.plot(timeStamps, f0, color = 'k', linewidth=1.5)
plt.plot([0.5, 0.5], [0, maxplotfreq], color = 'b', linewidth=1.5)
plt.plot([4.0, 4.0], [0, maxplotfreq], color = 'b', linewidth=1.5)
plt.autoscale(tight=True)
plt.ylabel('Frequency (Hz)', fontsize = fontSize)
plt.xlabel('Time (s)', fontsize = fontSize)
plt.legend(('f0',))
xLim = ax.get_xlim()
yLim = ax.get_ylim()
ax.set_aspect((xLim[1]-xLim[0])/(2.0*(yLim[1]-yLim[0])))
if plot == 1: #save the plot too!
plt.autoscale(tight=True)
plt.show()
else:
fig.tight_layout()
fig.savefig('f0_over_Spectrogram.png', dpi=150, bbox_inches='tight')
return f0
def analysis(inputFile='../../sounds/vignesh.wav', window='blackman', M=1201, N=2048, t=-90,
minSineDur=0.1, nH=100, minf0=130, maxf0=300, f0et=7, harmDevSlope=0.01):
"""
Analyze a sound with the harmonic model
inputFile: input sound file (monophonic with sampling rate of 44100)
window: analysis window type (rectangular, hanning, hamming, blackman, blackmanharris)
M: analysis window size
N: fft size (power of two, bigger or equal than M)
t: magnitude threshold of spectral peaks
minSineDur: minimum duration of sinusoidal tracks
nH: maximum number of harmonics
minf0: minimum fundamental frequency in sound
maxf0: maximum fundamental frequency in sound
f0et: maximum error accepted in f0 detection algorithm
harmDevSlope: allowed deviation of harmonic tracks, higher harmonics have higher allowed deviation
returns inputFile: input file name; fs: sampling rate of input file, tfreq,
tmag: sinusoidal frequencies and magnitudes
"""
# size of fft used in synthesis
Ns = 512
# hop size (has to be 1/4 of Ns)
H = 128
# read input sound
fs, x = UF.wavread(inputFile)
# compute analysis window
w = get_window(window, M)
# compute the harmonic model of the whole sound
hfreq, hmag, hphase = HM.harmonicModelAnal(x, fs, w, N, H, t, nH, minf0, maxf0, f0et, harmDevSlope, minSineDur)
# synthesize the sines without original phases
y = SM.sineModelSynth(hfreq, hmag, np.array([]), Ns, H, fs)
# output sound file (monophonic with sampling rate of 44100)
outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_harmonicModel.wav'
# write the sound resulting from the inverse stft
UF.wavwrite(y, fs, outputFile)
# create figure to show plots
plt.figure(figsize=(12, 9))
# frequency range to plot
maxplotfreq = 5000.0
# plot the input sound
plt.subplot(3,1,1)
plt.plot(np.arange(x.size)/float(fs), x)
plt.axis([0, x.size/float(fs), min(x), max(x)])
plt.ylabel('amplitude')
plt.xlabel('time (sec)')
plt.title('input sound: x')
if (hfreq.shape[1] > 0):
plt.subplot(3,1,2)
tracks = np.copy(hfreq)
numFrames = tracks.shape[0]
frmTime = H*np.arange(numFrames)/float(fs)
tracks[tracks<=0] = np.nan
plt.plot(frmTime, tracks)
plt.axis([0, x.size/float(fs), 0, maxplotfreq])
plt.title('frequencies of harmonic tracks')
# plot the output sound
plt.subplot(3,1,3)
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(block=False)
return inputFile, fs, hfreq, hmag
plt.axis([0,x.size/float(fs),min(x),max(x)])
plt.subplot(412)
numFrames = int(mX[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
binFreq = np.arange(mX[0,:].size)*float(fs)/N
plt.pcolormesh(frmTime, binFreq, np.transpose(mX))
plt.title('mX, M=1024, N=1024, H=512')
plt.autoscale(tight=True)
plt.subplot(413)
numFrames = int(pX[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
binFreq = np.arange(pX[0,:].size)*float(fs)/N
plt.pcolormesh(frmTime, binFreq, np.diff(np.transpose(pX),axis=0))
plt.title('pX derivative, M=1024, N=1024, H=512')
plt.autoscale(tight=True)
plt.subplot(414)
plt.plot(np.arange(y.size)/float(fs), y,'b')
plt.axis([0,y.size/float(fs),min(y),max(y)])
plt.title('y')
plt.tight_layout()
plt.savefig('stft-system.png')
UF.wavwrite(y, fs, 'piano-stft.wav')
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 exploreSineModel(inputFile='../sms-tools/sounds/multisines.wav'):
"""
Input:
inputFile (string) = wav file including the path
Output:
return True
Discuss on the forum!
"""
window='hamming' # Window type
M=3001 # Window size in sample
N=4096 # FFT Size
t=-80 # Threshold
minSineDur=0.02 # minimum duration of a sinusoid
maxnSines=15 # Maximum number of sinusoids at any time frame
freqDevOffset=10 # minimum frequency deviation at 0Hz
freqDevSlope=0.001 # slope increase of minimum frequency deviation
Ns = 512 # size of fft used in synthesis
H = 128 # hop size (has to be 1/4 of Ns)
fs, x = UF.wavread(inputFile) # read input sound
w = get_window(window, M) # compute analysis window
# analyze the sound with the sinusoidal model
tfreq, tmag, tphase = SM.sineModelAnal(x, fs, w, N, H, t, maxnSines, minSineDur, freqDevOffset, freqDevSlope)
# synthesize the output sound from the sinusoidal representation
y = SM.sineModelSynth(tfreq, tmag, tphase, Ns, H, fs)
# output sound file name
outputFile = os.path.basename(inputFile)[:-4] + '_sineModel.wav'
# write the synthesized sound obtained from the sinusoidal synthesis
UF.wavwrite(y, fs, outputFile)
# create figure to show plots
plt.figure(figsize=(12, 9))
# frequency range to plot
maxplotfreq = 5000.0
# plot the input sound
plt.subplot(3,1,1)
plt.plot(np.arange(x.size)/float(fs), x)
plt.axis([0, x.size/float(fs), min(x), max(x)])
plt.ylabel('amplitude')
plt.xlabel('time (sec)')
plt.title('input sound: x')
# plot the sinusoidal frequencies
plt.subplot(3,1,2)
if (tfreq.shape[1] > 0):
numFrames = tfreq.shape[0]
frmTime = H*np.arange(numFrames)/float(fs)
tfreq[tfreq<=0] = np.nan
plt.plot(frmTime, tfreq)
plt.axis([0, x.size/float(fs), 0, maxplotfreq])
plt.title('frequencies of sinusoidal tracks')
# plot the output sound
plt.subplot(3,1,3)
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()
return True
def main(inputFile = '../../sounds/piano.wav', window = 'hamming', M = 1024, N = 1024, H = 512):
"""
analysis/synthesis using the STFT
inputFile: input sound file (monophonic with sampling rate of 44100)
window: analysis window type (choice of rectangular, hanning, hamming, blackman, blackmanharris)
M: analysis window size
N: fft size (power of two, bigger or equal than M)
H: hop size (at least 1/2 of analysis window size to have good overlap-add)
"""
# read input sound (monophonic with sampling rate of 44100)
fs, x = UF.wavread(inputFile)
# compute analysis window
w = get_window(window, M)
# compute the magnitude and phase spectrogram
mX, pX = STFT.stftAnal(x, w, N, H)
# perform the inverse stft
y = STFT.stftSynth(mX, pX, M, H)
# output sound file (monophonic with sampling rate of 44100)
outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_stft.wav'
# write the sound resulting from the inverse stft
UF.wavwrite(y, fs, outputFile)
# create figure to plot
plt.figure(figsize=(12, 9))
# frequency range to plot
maxplotfreq = 5000.0
# 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.ylabel('amplitude')
plt.xlabel('time (sec)')
plt.title('input sound: x')
# plot magnitude spectrogram
plt.subplot(4,1,2)
numFrames = int(mX[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
binFreq = fs*np.arange(N*maxplotfreq/fs)/N
plt.pcolormesh(frmTime, binFreq, np.transpose(mX[:,:N*maxplotfreq/fs+1]))
plt.xlabel('time (sec)')
plt.ylabel('frequency (Hz)')
plt.title('magnitude spectrogram')
plt.autoscale(tight=True)
# plot the phase spectrogram
plt.subplot(4,1,3)
numFrames = int(pX[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
binFreq = fs*np.arange(N*maxplotfreq/fs)/N
plt.pcolormesh(frmTime, binFreq, np.transpose(np.diff(pX[:,:N*maxplotfreq/fs+1],axis=1)))
plt.xlabel('time (sec)')
plt.ylabel('frequency (Hz)')
plt.title('phase spectrogram (derivative)')
plt.autoscale(tight=True)
# plot the output sound
plt.subplot(4,1,4)
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 analysis(inputFile='../../sounds/mridangam.wav', window='hamming', M=801, N=2048, t=-90,
minSineDur=0.01, maxnSines=150, freqDevOffset=20, freqDevSlope=0.02):
"""
Analyze a sound with the sine model
inputFile: input sound file (monophonic with sampling rate of 44100)
window: analysis window type (rectangular, hanning, hamming, blackman, blackmanharris)
M: analysis window size; N: fft size (power of two, bigger or equal than M)
t: magnitude threshold of spectral peaks; minSineDur: minimum duration of sinusoidal tracks
maxnSines: maximum number of parallel sinusoids
freqDevOffset: frequency deviation allowed in the sinusoids from frame to frame at frequency 0
freqDevSlope: slope of the frequency deviation, higher frequencies have bigger deviation
returns inputFile: input file name; fs: sampling rate of input file,
tfreq, tmag: sinusoidal frequencies and magnitudes
"""
# size of fft used in synthesis
Ns = 512
# hop size (has to be 1/4 of Ns)
H = 128
# read input sound
(fs, x) = UF.wavread(inputFile)
# compute analysis window
w = get_window(window, M)
# compute the sine model of the whole sound
tfreq, tmag, tphase = SM.sineModelAnal(x, fs, w, N, H, t, maxnSines, minSineDur, freqDevOffset, freqDevSlope)
# synthesize the sines without original phases
y = SM.sineModelSynth(tfreq, tmag, np.array([]), Ns, H, fs)
# output sound file (monophonic with sampling rate of 44100)
outputFile = 'output_sounds/' + os.path.basename(inputFile)[:-4] + '_sineModel.wav'
# write the sound resulting from the inverse stft
UF.wavwrite(y, fs, outputFile)
# create figure to show plots
plt.figure(figsize=(12, 9))
# frequency range to plot
maxplotfreq = 5000.0
# plot the input sound
plt.subplot(3,1,1)
plt.plot(np.arange(x.size)/float(fs), x)
plt.axis([0, x.size/float(fs), min(x), max(x)])
plt.ylabel('amplitude')
plt.xlabel('time (sec)')
plt.title('input sound: x')
# plot the sinusoidal frequencies
if (tfreq.shape[1] > 0):
plt.subplot(3,1,2)
tracks = np.copy(tfreq)
tracks = tracks*np.less(tracks, maxplotfreq)
tracks[tracks<=0] = np.nan
numFrames = int(tracks[:,0].size)
frmTime = H*np.arange(numFrames)/float(fs)
plt.plot(frmTime, tracks)
plt.axis([0, x.size/float(fs), 0, maxplotfreq])
plt.title('frequencies of sinusoidal tracks')
# plot the output sound
plt.subplot(3,1,3)
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(block=False)
return inputFile, fs, tfreq, tmag