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 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 marked as XX
window = 'blackman'
M = 4001
N = 4096
f0et = 11
t = -80
minf0 = 130
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
### Method 1
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
fig = plt.figure()
ax = fig.add_subplot(111)
mX, pX = stft.stftAnal(x, 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
(fs, x) = UF.wavread('../../../sounds/piano.wav')
w = np.blackman(1501)
N = 2048
t = -90
minf0 = 100
maxf0 = 300
f0et = 1
maxnpeaksTwm = 4
H = 128
x1 = x[int(1.5*fs):int(1.8*fs)]
plt.figure(1, figsize=(9, 7))
mX, pX = STFT.stftAnal(x, w, N, H)
f0 = HM.f0Detection(x, fs, w, N, H, t, minf0, maxf0, f0et)
f0 = UF.cleaningTrack(f0, 5)
yf0 = UF.sinewaveSynth(f0, .8, H, fs)
f0[f0==0] = np.nan
maxplotfreq = 800.0
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[:,:int(N*maxplotfreq/fs+1)]))
plt.autoscale(tight=True)
plt.plot(frmTime, f0, linewidth=2, color='k')
plt.autoscale(tight=True)
plt.title('mX + f0 (piano.wav), TWM')
plt.tight_layout()
plt.savefig('f0Twm-piano.png')
UF.wavwrite(yf0, fs, 'f0Twm-piano.wav')
(fs, x) = UF.wavread('../../../sounds/piano.wav')
w = np.blackman(1501)
N = 2048
t = -90
minf0 = 100
maxf0 = 300
f0et = 1
maxnpeaksTwm = 4
H = 128
x1 = x[int(1.5 * fs):int(1.8 * fs)]
plt.figure(1, figsize=(9, 7))
mX, pX = STFT.stftAnal(x, w, N, H)
f0 = HM.f0Detection(x, fs, w, N, H, t, minf0, maxf0, f0et)
f0 = UF.cleaningTrack(f0, 5)
yf0 = UF.sinewaveSynth(f0, .8, H, fs)
f0[f0 == 0] = np.nan
maxplotfreq = 800.0
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[:, :int(N * maxplotfreq / fs + 1)]))
plt.autoscale(tight=True)
plt.plot(frmTime, f0, linewidth=2, color='k')
plt.autoscale(tight=True)
plt.title('mX + f0 (piano.wav), TWM')
plt.tight_layout()
plt.savefig('f0Twm-piano.png')
salience = run_pitch_salience_function(peak_frequencies, peak_magnitudes)
salience_peaks_bins, salience_peaks_saliences = run_pitch_salience_function_peaks(salience)
pool.add('allframes_salience_peaks_bins', salience_peaks_bins)
pool.add('allframes_salience_peaks_saliences', salience_peaks_saliences)
contours_bins, contours_saliences, contours_start_times, duration = run_pitch_contours(
pool['allframes_salience_peaks_bins'],
pool['allframes_salience_peaks_saliences'])
pitch, confidence = run_pitch_contours_melody(contours_bins,
contours_saliences,
contours_start_times,
duration)
yf0 = UF.sinewaveSynth(pitch, .6, hopSize, sampleRate)
figure(1, figsize=(9, 6))
mX, pX = STFT.stftAnal(audio, sampleRate, hamming(frameSize), frameSize, hopSize)
maxplotfreq = 3000.0
numFrames = int(mX[:,0].size)
frmTime = hopSize*arange(numFrames)/float(sampleRate)
binFreq = sampleRate*arange(frameSize*maxplotfreq/sampleRate)/frameSize
plt.pcolormesh(frmTime, binFreq, np.transpose(mX[:,:frameSize*maxplotfreq/sampleRate+1]))
plt.autoscale(tight=True)
offset = .5 * frameSize/sampleRate
time = hopSize*arange(size(pitch))/float(sampleRate)
pitch[pitch==0]=nan
plot(time, pitch, color='k', linewidth = 2)
salience = run_pitch_salience_function(peak_frequencies, peak_magnitudes)
salience_peaks_bins, salience_peaks_saliences = run_pitch_salience_function_peaks(
salience)
pool.add('allframes_salience_peaks_bins', salience_peaks_bins)
pool.add('allframes_salience_peaks_saliences', salience_peaks_saliences)
contours_bins, contours_saliences, contours_start_times, duration = run_pitch_contours(
pool['allframes_salience_peaks_bins'],
pool['allframes_salience_peaks_saliences'])
pitch, confidence = run_pitch_contours_melody(contours_bins,
contours_saliences,
contours_start_times, duration)
yf0 = UF.sinewaveSynth(pitch, .6, hopSize, sampleRate)
figure(1, figsize=(9, 6))
mX, pX = STFT.stftAnal(audio, hamming(frameSize), frameSize, hopSize)
maxplotfreq = 3000.0
numFrames = int(mX[:, 0].size)
frmTime = hopSize * arange(numFrames) / float(sampleRate)
binFreq = sampleRate * arange(frameSize * maxplotfreq / sampleRate) / frameSize
plt.pcolormesh(frmTime, binFreq,
np.transpose(mX[:, :frameSize * maxplotfreq / sampleRate + 1]))
plt.autoscale(tight=True)
offset = .5 * frameSize / sampleRate
time = hopSize * arange(size(pitch)) / float(sampleRate)
pitch[pitch == 0] = nan
