def test_reconstruct_sound():
fs, x = audio.read_wav(sound_path("sax-phrase-short.wav"))
window_size, fft_size, hop_size = 4001, 4096, 2048
window = get_window('hamming', window_size)
xtfreq, xtmag, xtphase = sine.from_audio(
x, fs, window, fft_size, hop_size,
t=-80, maxnSines=100, minSineDur=.01, freqDevOffset=20, freqDevSlope=0.01)
x_reconstructed = sine.to_audio(xtfreq, xtmag, xtphase, fft_size, hop_size, fs)
assert 138746 == len(x)
expected_frame_count = int(math.ceil(float(len(x)) / hop_size))
assert expected_frame_count == len(xtfreq)
assert expected_frame_count == len(xtmag)
assert expected_frame_count == len(xtphase)
assert xtfreq.shape[1] <= 100
# statistics of the model for regression testing without explicitly storing the whole data
assert np.allclose(945.892990545, xtfreq.mean())
assert np.allclose(-30.3138495002, xtmag.mean())
assert np.allclose(1.34449391701, xtphase.mean())
# TODO: this is completely off, it should be equal to len(x)!
assert 69 * 2048 == len(x_reconstructed)
assert np.allclose(0.010812475879315771, rmse(x, x_reconstructed[:len(x)]))
def analysis(inputFile=demo_sound_path('mridangam.wav'),
window='hamming',
M=801,
N=2048,
t=-90,
minSineDur=0.01,
maxnSines=150,
freqDevOffset=20,
freqDevSlope=0.02,
interactive=True,
plotFile=False):
"""
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) = audio.read_wav(inputFile)
# compute analysis window
w = get_window(window, M)
# compute the sine model of the whole sound
tfreq, tmag, tphase = sine.from_audio(x, fs, w, N, H, t, maxnSines,
minSineDur, freqDevOffset,
freqDevSlope)
# synthesize the sines without original phases
y = sine.to_audio(tfreq, tmag, np.array([]), Ns, H, fs)
# output sound file (monophonic with sampling rate of 44100)
outputFile = 'output_sounds/' + strip_file(inputFile) + '_sineModel.wav'
# write the sound resulting from the inverse stft
audio.write_wav(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.shape[0])
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()
if interactive:
plt.show(block=False)
if plotFile:
plt.savefig('output_plots/%s_sine_transformation_analysis.png' %
files.strip_file(inputFile))
return inputFile, fs, tfreq, tmag
from smst.models import sine, stft
plt.figure(1, figsize=(9, 7))
plt.subplot(211)
(fs, x) = audio.read_wav('../../../sounds/vibraphone-C6.wav')
w = np.blackman(401)
N = 512
H = 100
t = -100
minSineDur = .02
maxnSines = 150
freqDevOffset = 20
freqDevSlope = 0.01
mX, pX = stft.from_audio(x, w, N, H)
tfreq, tmag, tphase = sine.from_audio(x, fs, w, N, H, t, maxnSines, minSineDur,
freqDevOffset, freqDevSlope)
maxplotfreq = 10000.0
maxplotbin = int(N * maxplotfreq / fs)
numFrames = int(mX.shape[0])
frmTime = H * np.arange(numFrames) / float(fs)
binFreq = np.arange(maxplotbin + 1) * float(fs) / N
plt.pcolormesh(frmTime, binFreq, np.transpose(mX[:, :maxplotbin + 1]))
plt.autoscale(tight=True)
tracks = tfreq * np.less(tfreq, maxplotfreq)
tracks[tracks <= 0] = np.nan
plt.plot(frmTime, tracks, color='k', lw=1.5)
plt.autoscale(tight=True)
plt.title('mX + sine frequencies (vibraphone-C6.wav)')
from smst.utils import audio
from smst.models import sine
(fs, x) = audio.read_wav('../../../sounds/mridangam.wav')
x1 = x[:int(1.49 * fs)]
w = np.hamming(801)
N = 2048
t = -90
minSineDur = .005
maxnSines = 150
freqDevOffset = 20
freqDevSlope = 0.02
Ns = 512
H = Ns / 4
sfreq, smag, sphase = sine.from_audio(x1, fs, w, N, H, t, maxnSines,
minSineDur, freqDevOffset, freqDevSlope)
timeScale = np.array([
.01, .0, .03, .03, .335, .8, .355, .82, .671, 1.0, .691, 1.02, .858, 1.1,
.878, 1.12, 1.185, 1.8, 1.205, 1.82, 1.49, 2.0
])
L = sfreq.shape[0] # number of input frames
maxInTime = max(timeScale[::2]) # maximum value used as input times
maxOutTime = max(timeScale[1::2]) # maximum value used in output times
outL = int(L * maxOutTime / maxInTime) # number of output frames
inFrames = L * timeScale[::2] / maxInTime # input time values in frames
outFrames = outL * timeScale[1::2] / maxOutTime # output time values in frames
timeScalingEnv = interp1d(outFrames, inFrames,
fill_value=0) # interpolation function
indexes = timeScalingEnv(
np.arange(outL)) # generate frame indexes for the output
ysfreq = sfreq[round(indexes[0]), :] # first output frame
mpl.use('Agg')
import matplotlib.pyplot as plt
import numpy as np
from smst.utils import audio
from smst.models import sine, stft
(fs, x) = audio.read_wav('../../../sounds/flute-A4.wav')
w = np.blackman(601)
N = 1024
H = 150
t = -80
minSineDur = .1
maxnSines = 150
mX, pX = stft.from_audio(x, w, N, H)
tfreq, tmag, tphase = sine.from_audio(x, fs, w, N, H, t, maxnSines, minSineDur)
plt.figure(1, figsize=(9.5, 5))
maxplotfreq = 5000.0
maxplotbin = int(N * maxplotfreq / fs)
numFrames = int(mX.shape[0])
frmTime = H * np.arange(numFrames) / float(fs)
binFreq = np.arange(maxplotbin + 1) * float(fs) / N
plt.pcolormesh(frmTime, binFreq, np.transpose(mX[:, :maxplotbin + 1]))
plt.autoscale(tight=True)
tracks = tfreq * np.less(tfreq, maxplotfreq)
tracks[tracks <= 0] = np.nan
plt.plot(frmTime, tracks, color='k', lw=1.5)
plt.autoscale(tight=True)
plt.title('mX + sinusoidal tracks (flute-A4.wav)')
def main(inputFile=demo_sound_path('bendir.wav'), window='hamming', M=2001, N=2048, t=-80, minSineDur=0.02,
maxnSines=150, freqDevOffset=10, freqDevSlope=0.001,
interactive=True, plotFile=False):
"""
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 = audio.read_wav(inputFile)
# compute analysis window
w = get_window(window, M)
# analyze the sound with the sinusoidal model
tfreq, tmag, tphase = sine.from_audio(x, fs, w, N, H, t, maxnSines, minSineDur, freqDevOffset, freqDevSlope)
# synthesize the output sound from the sinusoidal representation
y = sine.to_audio(tfreq, tmag, tphase, Ns, H, fs)
# output sound file name
outputFile = 'output_sounds/' + strip_file(inputFile) + '_sineModel.wav'
# write the synthesized sound obtained from the sinusoidal synthesis
audio.write_wav(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()
if interactive:
plt.show()
if plotFile:
plt.savefig('output_plots/%s_sine_model.png' % files.strip_file(inputFile))