def test_simple(self):
ofilename = join(TEST_DATA_DIR, 'test.wav')
# Open the test file for reading
a = sndfile(ofilename, 'read')
nframes = a.get_nframes()
buffsize = 1024
buffsize = min(nframes, buffsize)
# First, read some frames, go back, and compare buffers
buff = a.read_frames(buffsize)
a.seek(0)
buff2 = a.read_frames(buffsize)
assert_array_equal(buff, buff2)
a.close()
# Now, read some frames, go back, and compare buffers
# (check whence == 1 == SEEK_CUR)
a = sndfile(ofilename, 'read')
a.read_frames(buffsize)
buff = a.read_frames(buffsize)
a.seek(-buffsize, 1)
buff2 = a.read_frames(buffsize)
assert_array_equal(buff, buff2)
a.close()
# Now, read some frames, go back, and compare buffers
# (check whence == 2 == SEEK_END)
a = sndfile(ofilename, 'read')
buff = a.read_frames(nframes)
a.seek(-buffsize, 2)
buff2 = a.read_frames(buffsize)
assert_array_equal(buff[-buffsize:], buff2)
def test_basic_io(self):
""" Check open, close and basic read/write"""
# dirty !
ofilename = join(TEST_DATA_DIR, 'test.wav')
rfd, fd, cfilename = open_tmp_file('pysndfiletest.wav')
try:
nbuff = 22050
# Open the test file for reading
a = sndfile(ofilename, 'read')
nframes = a.get_nframes()
# Open the copy file for writing
format = audio_format('wav', 'pcm16')
b = sndfile(fd, 'write', format, a.get_channels(),
a.get_samplerate())
# Copy the data
for i in range(nframes / nbuff):
tmpa = a.read_frames(nbuff)
assert tmpa.dtype == np.float
b.write_frames(tmpa, nbuff)
nrem = nframes % nbuff
tmpa = a.read_frames(nrem)
assert tmpa.dtype == np.float
b.write_frames(tmpa, nrem)
a.close()
b.close()
finally:
close_tmp_file(rfd, cfilename)
def test_basic_io(self):
""" Check open, close and basic read/write"""
# dirty !
ofilename = join(TEST_DATA_DIR, 'test.wav')
rfd, fd, cfilename = open_tmp_file('pysndfiletest.wav')
try:
nbuff = 22050
# Open the test file for reading
a = sndfile(ofilename, 'read')
nframes = a.get_nframes()
# Open the copy file for writing
format = audio_format('wav', 'pcm16')
b = sndfile(fd, 'write', format, a.get_channels(),
a.get_samplerate())
# Copy the data
for i in range(nframes / nbuff):
tmpa = a.read_frames(nbuff)
assert tmpa.dtype == np.float
b.write_frames(tmpa, nbuff)
nrem = nframes % nbuff
tmpa = a.read_frames(nrem)
assert tmpa.dtype == np.float
b.write_frames(tmpa, nrem)
a.close()
b.close()
finally:
close_tmp_file(rfd, cfilename)
def test_rw(self):
"""Test read/write pointers for seek."""
ofilename = join(TEST_DATA_DIR, 'test.wav')
rfd, fd, cfilename = open_tmp_file('rwseektest.wav')
try:
ref = sndfile(ofilename, 'read')
test = sndfile(fd, 'rwrite', format = ref._format, channels =
ref.get_channels(), samplerate = ref.get_samplerate())
n = 1024
rbuff = ref.read_frames(n, dtype = np.int16)
test.write_frames(rbuff)
tbuff = test.read_frames(n, dtype = np.int16)
assert_array_equal(rbuff, tbuff)
# Test seeking both read and write pointers
test.seek(0, 0)
test.write_frames(rbuff)
tbuff = test.read_frames(n, dtype = np.int16)
assert_array_equal(rbuff, tbuff)
# Test seeking only read pointer
rbuff1 = rbuff.copy()
rbuff2 = rbuff1 * 2 + 1
rbuff2.clip(-30000, 30000)
test.seek(0, 0, 'r')
test.write_frames(rbuff2)
tbuff1 = test.read_frames(n, dtype = np.int16)
try:
tbuff2 = test.read_frames(n, dtype = np.int16)
except IOError, e:
msg = "write pointer was updated in read seek !"
msg += "\n(msg is %s)" % e
raise AssertionError(msg)
assert_array_equal(rbuff1, tbuff1)
assert_array_equal(rbuff2, tbuff2)
if np.all(rbuff2 == tbuff1):
raise AssertionError("write pointer was updated"\
" in read seek !")
# Test seeking only write pointer
rbuff3 = rbuff1 * 2 - 1
rbuff3.clip(-30000, 30000)
test.seek(0, 0, 'rw')
test.seek(n, 0, 'w')
test.write_frames(rbuff3)
tbuff1 = test.read_frames(n, np.int16)
try:
assert_array_equal(tbuff1, rbuff1)
except AssertionError:
raise AssertionError("read pointer was updated in write seek !")
try:
tbuff3 = test.read_frames(n, np.int16)
except IOError, e:
msg = "read pointer was updated in write seek !"
msg += "\n(msg is %s)" % e
raise AssertionError(msg)
def test_simple(self):
ofilename = join(TEST_DATA_DIR, 'test.wav')
# Open the test file for reading
a = sndfile(ofilename, 'read')
nframes = a.get_nframes()
buffsize = 1024
buffsize = min(nframes, buffsize)
# First, read some frames, go back, and compare buffers
buff = a.read_frames(buffsize)
a.seek(0)
buff2 = a.read_frames(buffsize)
assert_array_equal(buff, buff2)
a.close()
# Now, read some frames, go back, and compare buffers
# (check whence == 1 == SEEK_CUR)
a = sndfile(ofilename, 'read')
a.read_frames(buffsize)
buff = a.read_frames(buffsize)
a.seek(-buffsize, 1)
buff2 = a.read_frames(buffsize)
assert_array_equal(buff, buff2)
a.close()
# Now, read some frames, go back, and compare buffers
# (check whence == 2 == SEEK_END)
a = sndfile(ofilename, 'read')
buff = a.read_frames(nframes)
a.seek(-buffsize, 2)
buff2 = a.read_frames(buffsize)
assert_array_equal(buff[-buffsize:], buff2)
def test_int_io(self):
# TODO: check if neg or pos value is the highest in abs
rfd, fd, cfilename = open_tmp_file('pysndfiletest.wav')
try:
nb = 2**25
nbuff = 22050
fs = 22050
a = np.random.random_integers(-nb, nb, nbuff)
a = a.astype(np.int32)
# Open the file for writing
format = audio_format('wav', 'pcm32')
b = sndfile(fd, 'write', format, 1, fs)
b.write_frames(a, nbuff)
b.close()
b = sndfile(cfilename, 'read')
read_a = b.read_frames(nbuff, dtype=np.int32)
b.close()
assert_array_equal(a, read_a)
finally:
close_tmp_file(rfd, cfilename)
def test_int_io(self):
# TODO: check if neg or pos value is the highest in abs
rfd, fd, cfilename = open_tmp_file('pysndfiletest.wav')
try:
nb = 2 ** 25
nbuff = 22050
fs = 22050
a = np.random.random_integers(-nb, nb, nbuff)
a = a.astype(np.int32)
# Open the file for writing
format = audio_format('wav', 'pcm32')
b = sndfile(fd, 'write', format, 1, fs)
b.write_frames(a, nbuff)
b.close()
b = sndfile(cfilename, 'read')
read_a = b.read_frames(nbuff, dtype = np.int32)
b.close()
assert_array_equal(a, read_a)
finally:
close_tmp_file(rfd, cfilename)
def make_rawaudio_segment(self, seg_time=1):
'''
make audio segments each of which is length 1s, 2s, 3s, .... 10s.
'''
samplerate = 0
stack = np.array(())
with closing(sndfile(self.filename)) as f:
fs = f.get_samplerate() #number of samples in 1 second
samplerate = fs
num_samples = fs * seg_time # i th seconds total number samples
nframes = f.get_nframes()
j = 0
if num_samples < nframes:
while j < nframes:
if num_samples < (nframes - j):
samples = f.read_frames(num_samples)
if j == 0:
stack = np.column_stack(np.array(samples)).T
else:
stack = np.column_stack((stack, np.array(samples)))
j = j + num_samples
else:
samples = f.read_frames(nframes - j)
# stack = np.column_stack((stack,samples))
j = j + num_samples
else:
samples = f.read_frames(nframes)
stack = np.column_stack((samples))
return stack, samplerate
def create_png(input_filename, output_filename_w, output_filename_s, image_width, image_height, fft_size, f_max, f_min, wavefile, palette, channel):
print "processing file %s:\n\t" % input_file,
audio_file = audiolab.sndfile(input_filename, 'read')
samples_per_pixel = audio_file.get_nframes() / float(image_width)
nyquist_freq = (audio_file.get_samplerate() / 2) + 0.0
processor = AudioProcessor(audio_file, fft_size, channel, numpy.hanning)
if wavefile==1:
waveform = WaveformImage(image_width, image_height, palette)
spectrogram = SpectrogramImage(image_width, image_height, fft_size, f_max, f_min, nyquist_freq, palette)
for x in range(image_width):
if x % (image_width/10) == 0:
sys.stdout.write('.')
sys.stdout.flush()
seek_point = int(x * samples_per_pixel)
next_seek_point = int((x + 1) * samples_per_pixel)
(spectral_centroid, db_spectrum) = processor.spectral_centroid(seek_point)
if wavefile==1:
peaks = processor.peaks(seek_point, next_seek_point)
waveform.draw_peaks(x, peaks, spectral_centroid)
spectrogram.draw_spectrum(x, db_spectrum)
if wavefile==1:
waveform.save(output_filename_w)
spectrogram.save(output_filename_s)
print " done"
def main():
# Run some tests
# Combine same-size arrays w/various offsets
a = 1 * np.ones(3)
b = 2 * np.ones(3)
assert (crude_combine(a, b, 0, 0) == np.array([1.5, 1.5, 1.5])).all()
assert (crude_combine(a, b, 0, 1) == np.array([2, 1.5, 1.5, 1])).all()
assert (crude_combine(a, b, 1, 0) == np.array([1, 1.5, 1.5, 2])).all()
assert (crude_combine(a, b, 0, 2) == np.array([2, 2, 1.5, 1, 1])).all()
assert (crude_combine(a, b, 2, 0) == np.array([1, 1, 1.5, 2, 2])).all()
assert (crude_combine(a, b, 0, 3) == np.array([])).all()
assert (crude_combine(a, b, 3, 0) == np.array([])).all()
# Now try it out with some audio
import scikits.audiolab as audiolab
s0 = audiolab.sndfile("AfterTheBattle01.wav")
f0 = s0.read_frames(s0.get_nframes())
# Break into overlapping pieces
f1 = f0[:200]
f2 = f0[100:]
# Recombine
f3 = crude_combine(f1, f2, 100, 0)
assert (f3 == f0).all()
def main():
# open original signal
sound = audiolab.sndfile("AfterTheBattle01.wav")
signal = sound.read_frames(sound.get_nframes())
cutoff_set = np.logspace(np.log10(500), np.log10(2e4), 20)
offset_set = np.zeros(cutoff_set.size)
i = 0
for cutoff in cutoff_set:
print "Cutoff:", cutoff
filtered = butter_lowpass(signal, cutoff)
offset = find_offset(filtered, signal)
print "Offset:", offset
offset_set[i] = offset
i += 1
# attempt to plot
pyplot.subplot(2, 1, 1)
pyplot.plot(cutoff_set, offset_set)
pyplot.xscale('log')
pyplot.xlabel("Lowpass butterworth cutoff freq")
pyplot.ylabel("Frame offset from unfiltered signal")
#pyplot.yscale('log')
pyplot.show()
# Save cutoff_set and offset_set to file
pass
def create_png(input_filename, output_filename_w, output_filename_s, image_width, image_height, fft_size, f_max, f_min, wavefile, palette, channel):
print "processing file %s:\n\t" % input_file,
audio_file = audiolab.sndfile(input_filename, 'read')
samples_per_pixel = audio_file.get_nframes() / float(image_width)
nyquist_freq = (audio_file.get_samplerate() / 2) + 0.0
processor = AudioProcessor(audio_file, fft_size, channel, numpy.hanning)
if wavefile==1:
waveform = WaveformImage(image_width, image_height, palette)
spectrogram = SpectrogramImage(image_width, image_height, fft_size, f_max, f_min, nyquist_freq, palette)
for x in range(image_width):
if x % (image_width/10) == 0:
sys.stdout.write('.')
sys.stdout.flush()
seek_point = int(x * samples_per_pixel)
next_seek_point = int((x + 1) * samples_per_pixel)
(spectral_centroid, db_spectrum) = processor.spectral_centroid(seek_point)
if wavefile==1:
peaks = processor.peaks(seek_point, next_seek_point)
waveform.draw_peaks(x, peaks, spectral_centroid)
spectrogram.draw_spectrum(x, db_spectrum)
if wavefile==1:
waveform.save(output_filename_w)
spectrogram.save(output_filename_s)
print " done"
def test_nofile(self):
""" Check the failure when opening a non existing file."""
try:
f = sndfile("floupi.wav", "read")
raise AssertionError("call to non existing file should not succeed")
except IOError:
pass
except Exception, e:
raise AssertionError("opening non existing file should raise a IOError exception, got %s instead" % e.__class__)
def show(self, example):
sound = audiolab.sndfile(self.base + example.file)
frames = sound.read_frames(sound.get_nframes()) * 0.8
mfcc = features.mfcc(frames[example.start:example.stop:2], fs=41000)
print mfcc[0].shape
fig = plt.figure()
fig.set_size_inches(20, 20)
ax = fig.add_subplot(111)
ax.imshow(mfcc[0].transpose()[:, :100])
def show(self, example): sound = audiolab.sndfile(self.base + example.file) frames = sound.read_frames(sound.get_nframes()) * 0.8 mfcc = features.mfcc(frames[example.start: example.stop:2], fs=41000) print mfcc[0].shape fig = plt.figure() fig.set_size_inches(20, 20) ax = fig.add_subplot(111) ax.imshow(mfcc[0].transpose()[:, :100])
def test_float64(self):
"""Check float64 write/read works"""
# dirty !
ofilename = join(TEST_DATA_DIR, 'test.wav')
rfd, fd, cfilename = open_tmp_file('pysndfiletest.wav')
try:
nbuff = 22050
# Open the test file for reading
a = sndfile(ofilename, 'read')
nframes = a.get_nframes()
# Open the copy file for writing
format = audio_format('wav', 'float64')
b = sndfile(fd, 'write', format, a.get_channels(),
a.get_samplerate())
# Copy the data in the wav file
for i in range(nframes / nbuff):
tmpa = a.read_frames(nbuff, dtype = np.float64)
assert tmpa.dtype == np.float64
b.write_frames(tmpa, nbuff)
nrem = nframes % nbuff
tmpa = a.read_frames(nrem)
b.write_frames(tmpa, nrem)
a.close()
b.close()
# Now, reopen both files in for reading, and check data are
# the same
a = sndfile(ofilename, 'read')
b = sndfile(cfilename, 'read')
for i in range(nframes / nbuff):
tmpa = a.read_frames(nbuff, dtype = np.float64)
tmpb = b.read_frames(nbuff, dtype = np.float64)
assert_array_equal(tmpa, tmpb)
a.close()
b.close()
finally:
close_tmp_file(rfd, cfilename)
def test_float64(self):
"""Check float64 write/read works"""
# dirty !
ofilename = join(TEST_DATA_DIR, 'test.wav')
rfd, fd, cfilename = open_tmp_file('pysndfiletest.wav')
try:
nbuff = 22050
# Open the test file for reading
a = sndfile(ofilename, 'read')
nframes = a.get_nframes()
# Open the copy file for writing
format = audio_format('wav', 'float64')
b = sndfile(fd, 'write', format, a.get_channels(),
a.get_samplerate())
# Copy the data in the wav file
for i in range(nframes / nbuff):
tmpa = a.read_frames(nbuff, dtype=np.float64)
assert tmpa.dtype == np.float64
b.write_frames(tmpa, nbuff)
nrem = nframes % nbuff
tmpa = a.read_frames(nrem)
b.write_frames(tmpa, nrem)
a.close()
b.close()
# Now, reopen both files in for reading, and check data are
# the same
a = sndfile(ofilename, 'read')
b = sndfile(cfilename, 'read')
for i in range(nframes / nbuff):
tmpa = a.read_frames(nbuff, dtype=np.float64)
tmpb = b.read_frames(nbuff, dtype=np.float64)
assert_array_equal(tmpa, tmpb)
a.close()
b.close()
finally:
close_tmp_file(rfd, cfilename)
def show_specgram(file):
"""
Reads in wav and displays spectrogram
"""
sound = audiolab.sndfile(file, 'read')
""" get_nframes() returns number of frames """
sound_info = sound.read_frames(sound.get_nframes())
spectrogram = specgram(sound_info)
sound.close()
show()
return spectrogram
def create_png(input_filename, output_filename_w, image_width, image_height,
channels, fft_size, f_max, f_min):
print "processing file %s:\n\t" % input_filename
audio_file = audiolab.sndfile(input_filename, 'read')
samples_per_pixel = audio_file.get_nframes() / float(image_width)
nyquist_freq = (audio_file.get_samplerate() / 2) + 0.0
processor = AudioProcessor(audio_file, fft_size, numpy.hanning)
path_split = os.path.split(output_filename_w)
for channel in range(channels):
waveform = WaveformImage(image_width, image_height / channels)
for x in range(image_width):
if x % (image_width / 10) == 0:
sys.stdout.write('.')
sys.stdout.flush()
seek_point = int(x * samples_per_pixel)
next_seek_point = int((x + 1) * samples_per_pixel)
(spectral_centroid,
db_spectrum) = processor.spectral_centroid(seek_point, channel)
peaks = processor.peaks(seek_point, next_seek_point, channel)
waveform.draw_peaks(x, peaks, spectral_centroid)
# If we have only one channel, don't bother with renaming
if channels == 1:
waveform.save(output_filename_w)
else:
waveform.save(
os.path.join(path_split[0],
str(channel) + path_split[1]))
print " done"
if channels > 1:
combined = Image.new("RGBA", (image_width, image_height))
# Delete the segments
for channel in range(channels):
cur = Image.open(
os.path.join(path_split[0],
str(channel) + path_split[1]))
combined.paste(cur, (0, channel * (image_height / channels)))
os.remove(os.path.join(path_split[0],
str(channel) + path_split[1]))
combined.save(output_filename_w)
print " done"
def get_framer_audio(filename, size, hop):
from scikits import audiolab
loader = audiolab.sndfile(filename)
sr = loader.get_sampleRate()
nframes = loader.get_nframes()
nchannels = loader.get_channels()
framer = framer_audio(loader, size, hop)
return framer, sr, nframes, nchannels, loader
def test_nofile(self):
""" Check the failure when opening a non existing file."""
try:
f = sndfile("floupi.wav", "read")
raise AssertionError(
"call to non existing file should not succeed")
except IOError:
pass
except Exception, e:
raise AssertionError(
"opening non existing file should raise a IOError exception, got %s instead"
% e.__class__)
def processWav(filename, channel):
"""
filename: path to a wav file
Channel: 1 for left, 2 for right
Returns centroids, frequencies, volumes
"""
#open file
audio_file = audiolab.sndfile(filename, 'read')
#should be length of audiofile in seconds * 60. will fix this later
import contextlib
import wave
with contextlib.closing(wave.open(filename, 'r')) as f:
frames = f.getnframes()
rate = f.getframerate()
duration = frames / float(rate)
duration *= 30 #30 data points for every second of audio yay
duration = int(
duration) #can only return an integer number of frames so yeah
#print duration
#Not really samples per pixel but I'll let that slide
samples_per_pixel = audio_file.get_nframes() / float(duration)
#some rule says this frequency has to be half of the sample rate
nyquist_freq = (audio_file.get_samplerate() / 2) + 0.0
#fft_size stays 2048; smaller size == more efficient, fewer frequency samples
processor = AudioProcessor(audio_file, 2048, channel, numpy.hanning)
centroids = []
frequencies = []
volumes = []
for x in range(duration):
seek_point = int(x * samples_per_pixel)
next_seek_point = int((x + 1) * samples_per_pixel)
(spectral_centroid,
db_spectrum) = processor.spectral_centroid(seek_point)
peaks = processor.peaks(seek_point, next_seek_point)
centroids.append(spectral_centroid)
frequencies.append(db_spectrum)
volumes.append(peaks)
#print "Centroids:" + str(centroids)
#print "Frequencies:" + str(frequencies)
#print "Volumes:" + str(volumes)
#convert volumes[] from peaks to actual volumes
for i in range(len(volumes)):
volumes[i] = abs(volumes[i][0]) + abs(volumes[i][1])
#round frequencies to save resources
for i in range(len(frequencies)):
for j in range(len(frequencies[i])):
frequencies[i][j] = round(frequencies[i][j], 4)
return centroids, frequencies, volumes
def test_bigframes(self):
""" Try to seek really far"""
rawname = join(TEST_DATA_DIR, 'test.wav')
a = sndfile(rawname, 'read')
try:
try:
a.seek(2 ** 60)
raise Exception("Seek really succeded ! This should not happen")
except PyaudioIOError, e:
pass
finally:
a.close()
def show_specgram(speech):
sound = audiolab.sndfile(speech, 'read')
sound_info = sound.read_frames(sound.get_nframes())
#spectrogram = plt.specgram(sound_info)
mfcc = talkfeat.mfcc(sound_info)
#print mfcc
plt.imshow(mfcc[0].transpose())
plt.title('Spectrogram of %s' % sys.argv[1])
plt.show()
sound.close()
def parse_audio(self):
"""
"""
with closing(sndfile(self.filename)) as f:
# print("sampling rate = {} Hz \nlength = {} samples\nchannels = {}\nencoding={}\nendianness={}\n".format(f.get_samplerate(), f.get_nframes(), f.get_channels(),f.get_encoding(),f.get_endianness()))
sig = f.read_frames(f.get_nframes())
# self.plot_time_domain_signal(sig)
# plt.plot(sig)
# plt.show()
return sig, f.get_samplerate(), f.get_channels()
def test_bigframes(self):
""" Try to seek really far"""
rawname = join(TEST_DATA_DIR, 'test.wav')
a = sndfile(rawname, 'read')
try:
try:
a.seek(2**60)
raise Exception(
"Seek really succeded ! This should not happen")
except PyaudioIOError, e:
pass
finally:
a.close()
def processWav(filename, channel):
"""
filename: path to a wav file
Channel: 1 for left, 2 for right
Returns centroids, frequencies, volumes
"""
#open file
audio_file = audiolab.sndfile(filename, 'read')
#should be length of audiofile in seconds * 60. will fix this later
import contextlib
import wave
with contextlib.closing(wave.open(filename, 'r')) as f:
frames = f.getnframes()
rate = f.getframerate()
duration = frames / float(rate)
duration *= 30 #30 data points for every second of audio yay
duration = int(duration) #can only return an integer number of frames so yeah
#print duration
#Not really samples per pixel but I'll let that slide
samples_per_pixel = audio_file.get_nframes() / float(duration)
#some rule says this frequency has to be half of the sample rate
nyquist_freq = (audio_file.get_samplerate() / 2) + 0.0
#fft_size stays 2048; smaller size == more efficient, fewer frequency samples
processor = AudioProcessor(audio_file, 2048, channel, numpy.hanning)
centroids = []
frequencies = []
volumes = []
for x in range(duration):
seek_point = int(x * samples_per_pixel)
next_seek_point = int((x + 1) * samples_per_pixel)
(spectral_centroid, db_spectrum) = processor.spectral_centroid(seek_point)
peaks = processor.peaks(seek_point, next_seek_point)
centroids.append(spectral_centroid)
frequencies.append(db_spectrum)
volumes.append(peaks)
#print "Centroids:" + str(centroids)
#print "Frequencies:" + str(frequencies)
#print "Volumes:" + str(volumes)
#convert volumes[] from peaks to actual volumes
for i in range(len(volumes)):
volumes[i] = abs(volumes[i][0]) + abs(volumes[i][1])
#round frequencies to save resources
for i in range(len(frequencies)):
for j in range(len(frequencies[i])):
frequencies[i][j] = round(frequencies[i][j], 4)
return centroids, frequencies, volumes
def create_png(input_filename, output_filename_w, image_width, image_height, channels, fft_size, f_max, f_min):
print "processing file %s:\n\t" % input_filename
audio_file = audiolab.sndfile(input_filename, 'read')
samples_per_pixel = audio_file.get_nframes() / float(image_width)
nyquist_freq = (audio_file.get_samplerate() / 2) + 0.0
processor = AudioProcessor(audio_file, fft_size, numpy.hanning)
path_split = os.path.split(output_filename_w)
for channel in range(channels):
waveform = WaveformImage(image_width, image_height/channels)
for x in range(image_width):
if x % (image_width/10) == 0:
sys.stdout.write('.')
sys.stdout.flush()
seek_point = int(x * samples_per_pixel)
next_seek_point = int((x + 1) * samples_per_pixel)
(spectral_centroid, db_spectrum) = processor.spectral_centroid(seek_point, channel)
peaks = processor.peaks(seek_point, next_seek_point, channel)
waveform.draw_peaks(x, peaks, spectral_centroid)
# If we have only one channel, don't bother with renaming
if channels == 1:
waveform.save(output_filename_w)
else:
waveform.save(os.path.join(path_split[0], str(channel) +
path_split[1]))
print " done"
if channels > 1:
combined = Image.new("RGBA", (image_width, image_height))
# Delete the segments
for channel in range(channels):
cur = Image.open(os.path.join(path_split[0], str(channel) +
path_split[1]))
combined.paste(cur, (0, channel * (image_height/channels)))
os.remove(os.path.join(path_split[0], str(channel) +
path_split[1]))
combined.save(output_filename_w)
print " done"
def split_wav(wav_file, ref_file, is_save=False, dir_name="split"):
from scikits.audiolab import formatinfo as format
import scikits.audiolab as audiolab
import shutil
fr = audiolab.sndfile(wav_file, "read")
n_channels = fr.get_channels()
fmt = format("wav", fr.get_encoding())
fs = fr.get_samplerate()
if is_save:
shutil.rmtree(dir_name, ignore_errors=True)
shutil.os.mkdir(dir_name)
slices = dict()
with open(ref_file, "r") as ref_read:
for i, line in enumerate(ref_read.readlines()):
fields = line.strip().split(" ")
assert len(fields) == 3
begin = int(fields[0])
end = int(fields[1])
word = fields[2]
fr.seek(begin)
sli = fr.read_frames(end - begin + 1)
if not slices.has_key(word):
slices[word] = []
# if(len(sli) == 1):
# print word, line
slices[word].append(sli)
if is_save:
path = shutil.os.path.join(dir_name, str(i) + "-" + word + ".wav")
afile = audiolab.sndfile(path, "write", fmt, n_channels, fs)
afile.write_frames(sli, len(sli))
afile.close()
return slices
def test_float_frames(self):
""" Check nframes can be a float"""
rfd, fd, cfilename = open_tmp_file('pysndfiletest.wav')
try:
# Open the file for writing
format = audio_format('wav', 'pcm16')
a = sndfile(fd, 'rwrite', format, channels=1, samplerate=22050)
tmp = np.random.random_integers(-100, 100, 1000)
tmp = tmp.astype(np.short)
a.write_frames(tmp, tmp.size)
a.seek(0)
a.sync()
ctmp = a.read_frames(1e2, dtype=np.short)
a.close()
finally:
close_tmp_file(rfd, cfilename)
def play_rand(file, name):
sound = audiolab.sndfile(file)
limit = sound.get_nframes()
frames = sound.read_frames(sound.get_nframes()) * 0.8
if limit < size: return
for i in range(5):
start = random.randint(0, limit - size)
print("Is this a voice? [(y)es, (n)o, " +
"(s)kip/significant-portions-of-both/can't-tell, (r)eplay]")
audiolab.play(frames[start:start + size][:, 0])
while True:
input = raw_input()
if input == "r":
audiolab.play(frames[start:start + size][:, 0])
elif input in ["y", "n", "s"]:
break
print >> outfile, name, start, start + size, input
def play_rand(file, name):
sound = audiolab.sndfile(file)
limit = sound.get_nframes()
frames = sound.read_frames(sound.get_nframes()) * 0.8
if limit < size: return
for i in range(5):
start = random.randint(0, limit - size)
print("Is this a voice? [(y)es, (n)o, " +
"(s)kip/significant-portions-of-both/can't-tell, (r)eplay]")
audiolab.play(frames[start: start + size][:,0])
while True:
input = raw_input()
if input == "r":
audiolab.play(frames[start: start + size][:,0])
elif input in ["y", "n", "s"]:
break
print >> outfile, name, start, start + size, input
def test_mismatch(self):
# This test open a file for writing, but with bad args (channels and
# nframes inverted)
rfd, fd, cfilename = open_tmp_file('pysndfiletest.wav')
try:
# Open the file for writing
format = audio_format('wav', 'pcm16')
try:
b = sndfile(fd, 'write', \
format, channels = 22000, samplerate = 1)
raise Exception("Try to open a file with more than 256 "\
"channels, this should not succeed !")
except ValueError, e:
#print "Gave %d channels, error detected is \"%s\"" % (22000, e)
pass
finally:
close_tmp_file(rfd, cfilename)
def test_float_frames(self):
""" Check nframes can be a float"""
rfd, fd, cfilename = open_tmp_file('pysndfiletest.wav')
try:
# Open the file for writing
format = audio_format('wav', 'pcm16')
a = sndfile(fd, 'rwrite', format, channels = 1,
samplerate = 22050)
tmp = np.random.random_integers(-100, 100, 1000)
tmp = tmp.astype(np.short)
a.write_frames(tmp, tmp.size)
a.seek(0)
a.sync()
ctmp = a.read_frames(1e2, dtype = np.short)
a.close()
finally:
close_tmp_file(rfd, cfilename)
def test_mismatch(self):
# This test open a file for writing, but with bad args (channels and
# nframes inverted)
rfd, fd, cfilename = open_tmp_file('pysndfiletest.wav')
try:
# Open the file for writing
format = audio_format('wav', 'pcm16')
try:
b = sndfile(fd, 'write', \
format, channels = 22000, samplerate = 1)
raise Exception("Try to open a file with more than 256 "\
"channels, this should not succeed !")
except ValueError, e:
#print "Gave %d channels, error detected is \"%s\"" % (22000, e)
pass
finally:
close_tmp_file(rfd, cfilename)
def main():
# Run some tests
# Combine same-size arrays w/various offsets
a = 1 * numpy.ones(3)
b = 2 * numpy.ones(3)
assert (crude_combine(a, b, 0, 0) == numpy.array([1.5, 1.5, 1.5])).all()
assert (crude_combine(a, b, 0, 1) == numpy.array([2, 1.5, 1.5, 1])).all()
assert (crude_combine(a, b, 1, 0) == numpy.array([1, 1.5, 1.5, 2])).all()
assert (crude_combine(a, b, 0, 2) == numpy.array([2, 2, 1.5, 1, 1])).all()
assert (crude_combine(a, b, 2, 0) == numpy.array([1, 1, 1.5, 2, 2])).all()
assert (crude_combine(a, b, 0, 3) == numpy.array([])).all()
assert (crude_combine(a, b, 3, 0) == numpy.array([])).all()
# Let's try with one bigger than the other
a = 1 * numpy.ones(5)
b = 2 * numpy.ones(2)
assert (crude_combine(a, b, 0, 0) == numpy.array([1.5, 1.5, 1, 1, 1])).all()
assert (crude_combine(a, b, 0, 1) == numpy.array([2, 1.5, 1, 1, 1, 1])).all()
assert (crude_combine(a, b, 1, 0) == numpy.array([1, 1.5, 1.5, 1, 1])).all()
# Now try it out with some audio
import scikits.audiolab as audiolab
s0 = audiolab.sndfile("AfterTheBattle01.wav")
f0 = s0.read_frames(s0.get_nframes())
# Break into overlapping pieces
f1 = f0[:200]
f2 = f0[100:]
# Recombine
f3 = crude_combine(f1, f2, 100, 0)
assert (f3 == f0).all()
# Found a broken case
# Fixed, now it passes
f1 = f0
f2 = f0[f0.size - 160000 : f0.size - 160000 + 80000]
f3 = crude_combine(f2, f1, 0, f1.size - 160000)
assert (f3 == f0).all()
def the_soothing_sound_of_hash_collisions():
# it's a 3 minute song
SECONDS_PER_FRAME = 4
TREE_SIZE = 256
frames = []
with open('/dev/urandom', 'rb') as fp:
output = a.sndfile("merkle_chord.wav", format=a.formatinfo(),
mode='write', channels=1, samplerate=44100)
round_size = TREE_SIZE;
round_hashes = []
# we will seed this with 2 * TREE_SIZE random values that will be used
# by the algorithm to compute the leaves
for i in range(0, TREE_SIZE * 2):
this_seed = fp.read(32)
round_hashes.append(this_seed)
while (True):
round_hashes = compute_tree_level(round_hashes)
# we exhausted the tree by now
if not round_hashes:
break
frame_frequencies = compute_frequencies_for_tree_level(round_hashes)
print(frame_frequencies)
frame_pcm = compute_and_flatten_notes(frame_frequencies,
SECONDS_PER_FRAME)
output.write_frames(frame_pcm)
output.sync()
def play(self, example):
sound = audiolab.sndfile(self.base + example.file)
frames = sound.read_frames(sound.get_nframes()) * 0.8
audiolab.play(frames[example.start:example.stop][:, 0])
def create_png(input_filename, output_filename_w, output_filename_s,
image_width, image_height, fft_size, f_max, f_min, wavefile,
palette, channel):
"""
Given command line arguments this basically does everything.
WHAT I HAVE GATHERED:
db_spectrum has the frequencies of the sound file.
spectral_centroid tells us what the color of the sound is.
peaks tell us what the amplitude of the sound is.
Should be trivial to adapt this from image output to output
to our JavaScript visualizer now.
"""
print "processing file %s:\n\t" % input_file,
audio_file = audiolab.sndfile(
input_filename,
'read') #opens the wavfile; audio_file is an object now
samples_per_pixel = audio_file.get_nframes() / float(image_width)
nyquist_freq = (audio_file.get_samplerate() / 2) + 0.0
"""
Initializes AudioProcessor class, which does FFT analysis and spits
out amplitudes and frequencies to the SpectrogramImage and WaveformImage
classes below later. For a stereo wav file, this selects a single channel
to analyze. We might want to analyze both channels to give more input to
the visualizer,though.
"""
processor = AudioProcessor(audio_file, fft_size, channel, numpy.hanning)
if wavefile == 1:
waveform = WaveformImage(image_width, image_height, palette)
spectrogram = SpectrogramImage(image_width, image_height, fft_size, f_max,
f_min, nyquist_freq, palette)
for x in range(image_width):
#shows progress
if x % (image_width / 10) == 0:
sys.stdout.write('.')
sys.stdout.flush()
seek_point = int(x * samples_per_pixel)
next_seek_point = int((x + 1) * samples_per_pixel)
(spectral_centroid,
db_spectrum) = processor.spectral_centroid(seek_point)
#let's have a look at the spectral centroid and the db_spectrum
#print "Spectral Centroid:" + str(spectral_centroid)
#print "DB Spectrum:" + str(db_spectrum)
if wavefile == 1:
#aha! The peaks and spectral centroid make up the waveform.
#Since the spectral centroid indicates timbre (often referred to as color),
#it's probably what colors the waveform.
peaks = processor.peaks(seek_point, next_seek_point)
#let's have a look at these peaks
#print "Peaks:" + str(peaks)
waveform.draw_peaks(x, peaks, spectral_centroid)
spectrogram.draw_spectrum(x, db_spectrum)
if wavefile == 1:
waveform.save(output_filename_w)
spectrogram.save(output_filename_s)
print " done"
#!/usr/bin/env python from numpy import * from scipy import * # after numpy to import scipy's fft from scikits import audiolab filename = 'audio/lv1.aif' fft_window_size = 512 window_size = 2 * fft_window_size audiofile = audiolab.sndfile(filename) samples = audiofile.read_frames(audiofile.get_nframes()) # pad to right size so we can reshape it easily extras = repeat(0, window_size - len(samples) % window_size) samples = append(samples, extras) fft_shape = (samples.size / window_size, window_size) fft_windows = map(fft, samples.reshape(fft_shape)) print "got %d fft windows" % len(fft_windows)
def test_rw(self):
"""Test read/write pointers for seek."""
ofilename = join(TEST_DATA_DIR, 'test.wav')
rfd, fd, cfilename = open_tmp_file('rwseektest.wav')
try:
ref = sndfile(ofilename, 'read')
test = sndfile(fd,
'rwrite',
format=ref._format,
channels=ref.get_channels(),
samplerate=ref.get_samplerate())
n = 1024
rbuff = ref.read_frames(n, dtype=np.int16)
test.write_frames(rbuff)
tbuff = test.read_frames(n, dtype=np.int16)
assert_array_equal(rbuff, tbuff)
# Test seeking both read and write pointers
test.seek(0, 0)
test.write_frames(rbuff)
tbuff = test.read_frames(n, dtype=np.int16)
assert_array_equal(rbuff, tbuff)
# Test seeking only read pointer
rbuff1 = rbuff.copy()
rbuff2 = rbuff1 * 2 + 1
rbuff2.clip(-30000, 30000)
test.seek(0, 0, 'r')
test.write_frames(rbuff2)
tbuff1 = test.read_frames(n, dtype=np.int16)
try:
tbuff2 = test.read_frames(n, dtype=np.int16)
except IOError, e:
msg = "write pointer was updated in read seek !"
msg += "\n(msg is %s)" % e
raise AssertionError(msg)
assert_array_equal(rbuff1, tbuff1)
assert_array_equal(rbuff2, tbuff2)
if np.all(rbuff2 == tbuff1):
raise AssertionError("write pointer was updated"\
" in read seek !")
# Test seeking only write pointer
rbuff3 = rbuff1 * 2 - 1
rbuff3.clip(-30000, 30000)
test.seek(0, 0, 'rw')
test.seek(n, 0, 'w')
test.write_frames(rbuff3)
tbuff1 = test.read_frames(n, np.int16)
try:
assert_array_equal(tbuff1, rbuff1)
except AssertionError:
raise AssertionError(
"read pointer was updated in write seek !")
try:
tbuff3 = test.read_frames(n, np.int16)
except IOError, e:
msg = "read pointer was updated in write seek !"
msg += "\n(msg is %s)" % e
raise AssertionError(msg)
import numpy as N import scikits.audiolab as audiolab filename = 'test.wav' a = audiolab.sndfile(filename, 'read') tmp = a.read_frames(1e4) float_tmp = a.read_frames(1e4, dtype = N.float32) import pylab as P P.plot(tmp[:])
# This example plts 2 wav files, one above the other
import numpy
import scikits.audiolab as audiolab
import pylab
s1 = audiolab.sndfile("AfterTheBattle01.wav")
s2 = audiolab.sndfile("AfterTheBattle02.wav")
f1 = s1.read_frames(s1.get_nframes())
f2 = s2.read_frames(s2.get_nframes())
# subplot(nrows, ncols, plot_number)
# f1 will be plotted above f2
pylab.subplot(2, 1, 1)
pylab.plot(f1)
pylab.subplot(2, 1, 2)
pylab.plot(f2)
pylab.show()
def create_png(input_filename, output_filename_w, output_filename_s, image_width, image_height, fft_size, f_max, f_min, wavefile, palette, channel):
"""
Given command line arguments this basically does everything.
WHAT I HAVE GATHERED:
db_spectrum has the frequencies of the sound file.
spectral_centroid tells us what the color of the sound is.
peaks tell us what the amplitude of the sound is.
Should be trivial to adapt this from image output to output
to our JavaScript visualizer now.
"""
print "processing file %s:\n\t" % input_file,
audio_file = audiolab.sndfile(input_filename, 'read') #opens the wavfile; audio_file is an object now
samples_per_pixel = audio_file.get_nframes() / float(image_width)
nyquist_freq = (audio_file.get_samplerate() / 2) + 0.0
"""
Initializes AudioProcessor class, which does FFT analysis and spits
out amplitudes and frequencies to the SpectrogramImage and WaveformImage
classes below later. For a stereo wav file, this selects a single channel
to analyze. We might want to analyze both channels to give more input to
the visualizer,though.
"""
processor = AudioProcessor(audio_file, fft_size, channel, numpy.hanning)
if wavefile==1:
waveform = WaveformImage(image_width, image_height, palette)
spectrogram = SpectrogramImage(image_width, image_height, fft_size, f_max, f_min, nyquist_freq, palette)
for x in range(image_width):
#shows progress
if x % (image_width/10) == 0:
sys.stdout.write('.')
sys.stdout.flush()
seek_point = int(x * samples_per_pixel)
next_seek_point = int((x + 1) * samples_per_pixel)
(spectral_centroid, db_spectrum) = processor.spectral_centroid(seek_point)
#let's have a look at the spectral centroid and the db_spectrum
#print "Spectral Centroid:" + str(spectral_centroid)
#print "DB Spectrum:" + str(db_spectrum)
if wavefile==1:
#aha! The peaks and spectral centroid make up the waveform.
#Since the spectral centroid indicates timbre (often referred to as color),
#it's probably what colors the waveform.
peaks = processor.peaks(seek_point, next_seek_point)
#let's have a look at these peaks
#print "Peaks:" + str(peaks)
waveform.draw_peaks(x, peaks, spectral_centroid)
spectrogram.draw_spectrum(x, db_spectrum)
if wavefile==1:
waveform.save(output_filename_w)
spectrogram.save(output_filename_s)
print " done"
def frames_from_wav(filename):
sound = audiolab.sndfile(filename)
return sound.read_frames(sound.get_nframes())
#!python
import numpy as np
import scikits.audiolab as audiolab
import pylab as plt
import mel
from scipy.fftpack import dct
sound = audiolab.sndfile('prueba.wav')
data = sound.read_frames(sound.get_samplerate())
data = data[:, 0]
mfccs = []
for i in range(len(data) / 512 - 1):
win = data[512 * i:512 * i + 1024]
s = np.fft.rfft(win * ham, 512)
p = (s.real**2 + s.imag**2) / len(win)
m = np.log(np.dot(p, mel.MELfilterbank_speech).clip(1e-5, np.inf))
d = dct(m)
mfcc = d[1:13]
mfccs.append(mfcc)
plt.imshow(np.array(mfccs).T)
plt.show()
import scikits.audiolab as audiolab filename = 'test.wav' a = audiolab.sndfile(filename, 'read') print a
def test_basic_io_fd(self):
""" Check open from fd works"""
ofilename = join(TEST_DATA_DIR, 'test.wav')
fd = os.open(ofilename, os.O_RDONLY)
hdl = sndfile(fd, 'read')
hdl.close()
import scikits.audiolab as audiolab
a = audiolab.sndfile('test.wav', 'read')
data = a.read_frames(1000)
a.close()
def test_basic_io_fd(self):
""" Check open from fd works"""
ofilename = join(TEST_DATA_DIR, 'test.wav')
fd = os.open(ofilename, os.O_RDONLY)
hdl = sndfile(fd, 'read')
hdl.close()
import scikits.audiolab as al
import numpy as np
fps = 30.0
nband = 256
fft_size = 2**11
bandsize = fft_size/(2*nband)
w = al.sndfile('loop.wav', 'read')
sr = w.get_samplerate()
ns = w.get_nframes()
s = w.read_frames(ns)
binsize = int(sr/fps)
nbins = int(ns/binsize)
window = 2*binsize
tmp = np.zeros(window+len(s), dtype=np.float32)
tmp[binsize:-binsize] = s[:,0] #might want to avg channels?
s = tmp
ffts = np.zeros((nbins, nband), dtype=np.float32)
print nbins
for i in xrange(nbins):
tmp = s[i*binsize:i*binsize+window]
clip = (window-fft_size)/2
tmp = tmp[clip:-clip] * np.hanning(fft_size)
tmp2 = np.zeros(nband, dtype=np.float32)
fft_tmp = np.fft.fft(tmp)[:fft_size/2].real**2
for j in xrange(nband):
tmp2[j] = np.log(np.average(fft_tmp[j*bandsize:(j+1)*bandsize]))
ffts[i] = tmp2
bpm = 120.0
from tempfile import mkstemp
from os import remove
import numpy as N
from scikits.audiolab import formatinfo as format
import scikits.audiolab as audiolab
# Create a temp file in the system temporary dir, and always remove
# it at the end
cd, filename = mkstemp('tmptest.wav')
try:
fmt = format('wav', 'pcm24')
nchannels = 2
fs = 44100
afile = audiolab.sndfile(filename, 'write', fmt, nchannels, fs)
# Create a stereo white noise, with Gaussian distribution
tmp = 0.1 * N.random.randn(1000, nchannels)
# Write the first 500 frames of the signal
# Note that the write_frames method uses tmp's numpy dtype to determine how
# to write to the file; sndfile also converts the data on the fly if necessary
afile.write_frames(tmp, 500)
afile.close()
# Let's check that the written file has the expected meta data
afile = audiolab.sndfile(filename, 'read')
assert(afile.get_samplerate() == fs)
assert(afile.get_channels() == nchannels)
def test_raw(self):
rawname = join(TEST_DATA_DIR, 'test.raw')
format = audio_format('raw', 'pcm16', 'little')
a = sndfile(rawname, 'read', format, 1, 11025)
assert a.get_nframes() == 11290
a.close()
from tempfile import mkstemp
from os import remove
import numpy as N
from scikits.audiolab import formatinfo as format
import scikits.audiolab as audiolab
# Create a temp file in the system temporary dir, and always remove
# it at the end
cd, filename = mkstemp('tmptest.wav')
try:
fmt = format('wav', 'pcm24')
nchannels = 2
fs = 44100
afile = audiolab.sndfile(filename, 'write', fmt, nchannels, fs)
# Create a stereo white noise, with Gaussian distribution
tmp = 0.1 * N.random.randn(1000, nchannels)
# Write the first 500 frames of the signal
# Note that the write_frames method uses tmp's numpy dtype to determine how
# to write to the file; sndfile also converts the data on the fly if necessary
afile.write_frames(tmp, 500)
afile.close()
# Let's check that the written file has the expected meta data
afile = audiolab.sndfile(filename, 'read')
assert (afile.get_samplerate() == fs)
assert (afile.get_channels() == nchannels)
import matplotlib.pyplot as plt
import scikits.audiolab as audiolab
sound = audiolab.sndfile('Violin_for_spectogram.ogg', 'read')
y = sound.read_frames(sound.get_nframes())
Pxx, freqs, bins, im = plt.specgram(y, NFFT=512, Fs=44100)
plt.xlim(0, len(y) / 44100.0)
plt.ylim(0, 22050.0)
plt.colorbar(im).set_label(u'Intensidad (dB')
plt.xlabel(u'Tiempo(s')
plt.ylabel(u'Frecuencia(Hz)')
def test_raw(self):
rawname = join(TEST_DATA_DIR, 'test.raw')
format = audio_format('raw', 'pcm16', 'little')
a = sndfile(rawname, 'read', format, 1, 11025)
assert a.get_nframes() == 11290
a.close()
