def setup(self, channels=None, samplerate=None, blocksize=None,
totalframes=None):
super(AubioEncoder, self).setup(
channels, samplerate, blocksize, totalframes)
print ('setting up aubio with with', samplerate, blocksize)
self.sink = aubio.sink(self.filename, samplerate, channels)
def apply_filter(path):
from aubio import source, sink, digital_filter
from os.path import basename, splitext
# open input file, get its samplerate
s = source(path)
samplerate = s.samplerate
# create an A-weighting filter
f = digital_filter(7)
f.set_a_weighting(samplerate)
# alternatively, apply another filter
# create output file
o = sink("filtered_" + splitext(basename(path))[0] + ".wav", samplerate)
total_frames = 0
while True:
samples, read = s()
filtered_samples = f(samples)
o(filtered_samples, read)
total_frames += read
if read < s.hop_size: break
duration = total_frames / float(samplerate)
print ("read {:s}".format(s.uri))
print ("applied A-weighting filtered ({:d} Hz)".format(samplerate))
print ("wrote {:s} ({:.2f} s)".format(o.uri, duration))
def vocoder(file_id):
in_file = '/tmp/don_robot/output/' + file_id + ".mp3"
out_file = '/tmp/don_robot/output/' + file_id + "-voc.mp3"
samplerate = 44100
f = source(in_file, samplerate, 256)
g = sink(out_file, samplerate)
total_frames, read = 0, 256
win_s = 512 # fft size
hop_s = win_s // 2 # hop size
pv = pvoc(win_s, hop_s) # phase vocoder
while read:
samples, read = f()
spectrum = pv(samples) # compute spectrum
spectrum.norm *= .5 # reduce amplitude a bit .8
spectrum.phas[:] = 0. # zero phase
new_samples = pv.rdo(spectrum) # compute modified samples
g(new_samples, read) # write to output
total_frames += read
format_str = "read {:d} samples from {:s}, written to {:s}"
print(format_str.format(total_frames, f.uri, g.uri))
return out_file
def test_read_with(self):
samplerate = 44100
sink_path = get_tmp_sink_path()
vec = fvec(128)
with sink(sink_path, samplerate) as g:
for _ in range(10):
g(vec, 128)
def test_close_file_twice(self):
samplerate = 44100
sink_path = get_tmp_sink_path()
g = sink(sink_path, samplerate)
g.close()
g.close()
del_tmp_sink_path(sink_path)
def test_read_and_write_multi(self):
if not len(list_of_sounds):
self.skipTest('add some sound files in \'python/tests/sounds\'')
for path in list_of_sounds:
for samplerate, hop_size in zip([0, 44100, 8000, 32000],
[512, 1024, 64, 256]):
f = source(path, samplerate, hop_size)
if samplerate == 0: samplerate = f.samplerate
sink_path = get_tmp_sink_path()
g = sink(sink_path, samplerate, channels=f.channels)
total_frames = 0
while True:
vec, read = f.do_multi()
g.do_multi(vec, read)
total_frames += read
if read < f.hop_size: break
if 0:
print("read",
"%.2fs" % (total_frames / float(f.samplerate)),
end=' ')
print("(", total_frames, "frames", "in", end=' ')
print(f.channels, "channels", "in", end=' ')
print(total_frames / f.hop_size,
"blocks",
"at",
"%dHz" % f.samplerate,
")",
end=' ')
print("from", f.uri, end=' ')
print("to", g.uri, end=' ')
print("in", g.channels, "channels")
del_tmp_sink_path(sink_path)
def __init__(self, win_size=512, samplerate=44100):
# total number of audio frames read
self._total_frames = 0
# list of beats, in seconds since start
self._beats = []
# recording window and hop size
self._win_size = win_size
self._hop_size = win_size // 2 # hop size
# audio sample rate
self._samplerate = samplerate
# tempo finder algorithm instance
self._tempo_alg = tempo("default", self._win_size, self._hop_size, self._samplerate)
logging.info("Tempo silence=%d" % self._tempo_alg.get_silence())
self._tempo_alg.set_silence(-40)
self.on_pause = False
# pitch algorithm instance
# self.pitch_alg = pitch("default", self.win_size, self.hop_size, self.samplerate)
self._mono_vec = numpy.array([], dtype=numpy.float32)
self._tempo_found_callback = TempoFinder.default_tempo_found_callback
self._starting_millis = time() * 1000.
self._stream = sounddevice.InputStream(
channels=1, samplerate=float(self._samplerate), dtype='float32',
latency='low', callback=self.audio_callback)
if WRITE_WAV:
self._out_file = sink(samplerate=int(self._samplerate))
def test_close_file_twice(self):
samplerate = 44100
sink_path = get_tmp_sink_path()
g = sink(sink_path, samplerate)
g.close()
g.close()
del_tmp_sink_path(sink_path)
def apply_filter(path, target):
# open input file, get its samplerate
s = aubio.source(path)
samplerate = s.samplerate
# create an A-weighting filter
f = aubio.digital_filter(7)
f.set_a_weighting(samplerate)
# create output file
o = aubio.sink(target, samplerate)
total_frames = 0
while True:
# read from source
samples, read = s()
# filter samples
filtered_samples = f(samples)
# write to sink
o(filtered_samples, read)
# count frames read
total_frames += read
# end of file reached
if read < s.hop_size: break
# print some info
duration = total_frames / float(samplerate)
input_str = "input: {:s} ({:.2f} s, {:d} Hz)"
output_str = "output: {:s}, A-weighting filtered ({:d} frames total)"
print(input_str.format(s.uri, duration, samplerate))
print(output_str.format(o.uri, total_frames))
def apply_filter(path, target):
# open input file, get its samplerate
s = aubio.source(path)
samplerate = s.samplerate
# create an A-weighting filter
f = aubio.digital_filter(7)
f.set_a_weighting(samplerate)
# create output file
o = aubio.sink(target, samplerate)
total_frames = 0
while True:
# read from source
samples, read = s()
# filter samples
filtered_samples = f(samples)
# write to sink
o(filtered_samples, read)
# count frames read
total_frames += read
# end of file reached
if read < s.hop_size:
break
# print some info
duration = total_frames / float(samplerate)
input_str = "input: {:s} ({:.2f} s, {:d} Hz)"
output_str = "output: {:s}, A-weighting filtered ({:d} frames total)"
print(input_str.format(s.uri, duration, samplerate))
print(output_str.format(o.uri, total_frames))
def _handle_detected_problem(self, matcher_name: str) -> None:
now = datetime.datetime.now()
datestamp = now.strftime("%Y-%m-%d")
timestamp = now.strftime("%H:%M:%S")
path = pathlib.Path(
f"{tempfile.gettempdir()}/{self._name}_{datestamp}_{timestamp}.mp3"
)
# aubio only supports writing wav files, so use pydub to convert it to mp3
with tempfile.NamedTemporaryFile(suffix=".wav") as f:
with aubio.sink(f.name, self._source.samplerate) as output:
for match_samples in self._window_samples:
output(match_samples, self._source.hop_size)
wav_file = pydub.AudioSegment.from_wav(f)
wav_file.export(
path,
format="mp3",
tags={
"artist":
"Stream Monitor",
"title":
f"Problematic audio from stream {self._name} on {datestamp} at {timestamp}",
},
)
try:
self._problem_callback(self._name, matcher_name, path)
finally:
path.unlink()
def vocoder(file_id):
in_file = '/tmp/don_robot/output/' + file_id + ".mp3"
out_file = '/tmp/don_robot/output/' + file_id + "-voc.mp3"
samplerate = 44100
f = source(in_file, samplerate, 256)
g = sink(out_file, samplerate)
total_frames, read = 0, 256
win_s = 512 # fft size
hop_s = win_s // 2 # hop size
pv = pvoc(win_s, hop_s) # phase vocoder
while read:
samples, read = f()
spectrum = pv(samples) # compute spectrum
spectrum.norm *= .5 # reduce amplitude a bit .8
spectrum.phas[:] = 0. # zero phase
new_samples = pv.rdo(spectrum) # compute modified samples
g(new_samples, read) # write to output
total_frames += read
format_str = "read {:d} samples from {:s}, written to {:s}"
print(format_str.format(total_frames, f.uri, g.uri))
return out_file
def applyFilter(path, target):
# open input file, get its samplerate
s = source(path)
samplerate = s.samplerate
# create an A-weighting filter
f = digital_filter(7)
f.set_a_weighting(samplerate)
# create output file
o = sink(target, samplerate)
total_frames = 0
while True:
# read from source
samples, read = s()
# filter samples
filtered_samples = f(samples)
# write to sink
o(filtered_samples, read)
# count frames read
total_frames += read
# end of file reached
if read < s.hop_size:
break
def test_read_with(self):
samplerate = 44100
sink_path = get_tmp_sink_path()
vec = fvec(128)
with sink(sink_path, samplerate) as g:
for _ in range(10):
g(vec, 128)
def apply_filter(self):
# open input file, get its samplerate
s = aubio.source(self.path)
samplerate = s.samplerate
# create an A-weighting filter
f = aubio.digital_filter(7)
f.set_a_weighting(samplerate)
# create output file and replace original uploaded by user. Append the word filter to end of filename
new_file_path = self.path[:-4]
new_file_path += "_filter.wav"
o = aubio.sink(new_file_path, samplerate)
total_frames = 0
while True:
# read from source
samples, read = s()
# filter samples
filtered_samples = f(samples)
# write to sink
o(filtered_samples, read)
# count frames read
total_frames += read
# end of file reached
if read < s.hop_size:
break
def apply_filter(path):
from aubio import source, sink, digital_filter
from os.path import basename, splitext
# open input file, get its samplerate
s = source(path)
samplerate = s.samplerate
# create an A-weighting filter
f = digital_filter(7)
f.set_a_weighting(samplerate)
# alternatively, apply another filter
# create output file
o = sink("filtered_" + splitext(basename(path))[0] + ".wav", samplerate)
total_frames = 0
while True:
samples, read = s()
filtered_samples = f(samples)
o(filtered_samples, read)
total_frames += read
if read < s.hop_size: break
duration = total_frames / float(samplerate)
print("read {:s}".format(s.uri))
print("applied A-weighting filtered ({:d} Hz)".format(samplerate))
print("wrote {:s} ({:.2f} s)".format(o.uri, duration))
def test_many_sinks(self):
for i in range(100):
g = sink('/tmp/f.wav', 0)
write = 256
for n in range(200):
vec = fvec(write)
g(vec, write)
del g
def test_read_and_write_multi(self, hop_size, samplerate, path):
try:
f = source(path, samplerate, hop_size)
except RuntimeError as e:
self.skipTest('failed opening with hop_s = {:d}, samplerate = {:d} ({:s})'.format(hop_size, samplerate, str(e)))
if samplerate == 0: samplerate = f.samplerate
sink_path = get_tmp_sink_path()
g = sink(sink_path, samplerate, channels = f.channels)
total_frames = 0
while True:
vec, read = f.do_multi()
g.do_multi(vec, read)
total_frames += read
if read < f.hop_size: break
del_tmp_sink_path(sink_path)
def test_many_sinks(self):
from tempfile import mkdtemp
import os.path
import shutil
tmpdir = mkdtemp()
sink_list = []
for i in range(many_files):
path = os.path.join(tmpdir, 'f-' + str(i) + '.wav')
g = sink(path, 0)
sink_list.append(g)
write = 32
for _ in range(200):
vec = fvec(write)
g(vec, write)
g.close()
shutil.rmtree(tmpdir)
def test_many_sinks(self):
from tempfile import mkdtemp
import os.path
import shutil
tmpdir = mkdtemp()
sink_list = []
for i in range(many_files):
path = os.path.join(tmpdir, 'f-' + str(i) + '.wav')
g = sink(path, 0)
sink_list.append(g)
write = 32
for _ in range(200):
vec = fvec(write)
g(vec, write)
g.close()
shutil.rmtree(tmpdir)
def test_read(self):
for path in list_of_sounds:
for samplerate, hop_size in zip([0, 44100, 8000, 32000], [512, 1024, 64, 256]):
f = source(path, samplerate, hop_size)
if samplerate == 0: samplerate = f.samplerate
g = sink('/tmp/f.wav', samplerate)
total_frames = 0
while True:
vec, read = f()
g(vec, read)
total_frames += read
if read < f.hop_size: break
print "read", "%.2fs" % (total_frames / float(f.samplerate) ),
print "(", total_frames, "frames", "in",
print total_frames / f.hop_size, "blocks", "at", "%dHz" % f.samplerate, ")",
print "from", f.uri,
print "to", g.uri
def __call__(self, pipe):
for pool in pipe:
framesize = 2048
buff = pool["buffer"]
sink = aubio.sink(self.outfile, 0, self.mediafile.channels)
out_samples = numpy.array_split(
buff, int(float(self.mediafile.duration) / int(framesize)),
axis=1)
for frame in out_samples:
amount = frame[0].shape[0]
sink.do_multi(frame, amount)
sink.close()
del sink
yield {"out": self.outfile}
def apply_filter(path, params={}):
from aubio import source, sink, digital_filter
from os.path import basename, splitex, splitextt
s = source(path)
f = digital_filter(7)
f.set_a_weighting(s.samplerate)
#f = digital_filter(3)
#f.set_biquad(...)
o = sink("filtered_" + splitext(basename(path))[0] + ".wav")
# Total number of frames read
total_frames = 0
while True:
samples, read = s()
filtered_samples = f(samples)
o(samples, read)
total_frames += read
if read < s.hop_size: break
print "filtered", s.uri, "to", o.uri, "using an A-weighting filter"
def __call__(self, pipe):
for pool in pipe:
framesize = 2048
buff = pool["buffer"]
sink = aubio.sink(self.outfile, 0, self.mediafile.channels)
out_samples = numpy.array_split(
buff,
int(float(self.mediafile.duration) / int(framesize)),
axis=1)
for frame in out_samples:
amount = frame[0].shape[0]
sink.do_multi(frame, amount)
sink.close()
del sink
yield {"out": self.outfile}
def apply_filter(path, params = {}):
from aubio import source, sink, digital_filter
from os.path import basename, splitex, splitextt
s = source(path)
f = digital_filter(7)
f.set_a_weighting(s.samplerate)
#f = digital_filter(3)
#f.set_biquad(...)
o = sink("filtered_" + splitext(basename(path))[0] + ".wav")
# Total number of frames read
total_frames = 0
while True:
samples, read = s()
filtered_samples = f(samples)
o(samples, read)
total_frames += read
if read < s.hop_size: break
print "filtered", s.uri, "to", o.uri, "using an A-weighting filter"
def record_sink(sink_path):
"""Record an audio file using pysoundcard."""
from aubio import sink
from pysoundcard import Stream
hop_size = 256
duration = 5 # in seconds
s = Stream(block_length=hop_size)
g = sink(sink_path, samplerate=s.sample_rate)
s.start()
total_frames = 0
while total_frames < duration * s.sample_rate:
vec = s.read(hop_size)
# mix down to mono
mono_vec = vec.sum(-1) / float(s.input_channels)
g(mono_vec, hop_size)
total_frames += hop_size
s.stop()
def record_sink(sink_path):
"""Record an audio file using pysoundcard."""
from aubio import sink
from pysoundcard import Stream
hop_size = 256
duration = 5 # in seconds
s = Stream(block_length = hop_size)
g = sink(sink_path, samplerate = s.sample_rate)
s.start()
total_frames = 0
while total_frames < duration * s.sample_rate:
vec = s.read(hop_size)
# mix down to mono
mono_vec = vec.sum(-1) / float(s.input_channels)
g(mono_vec, hop_size)
total_frames += hop_size
s.stop()
def audio_filter(self):
"""
Taken from https://github.com/aubio/aubio/tree/master/python/demos
This was used for testing high and low pass filters which clean up the unwanted low and high frequencies.
This was done in an attempt to remove unwanted noise from the microphone signal. It may be useful in the
future
"""
# open input file, get its samplerate
s = aubio.source(self.filename)
samplerate = s.samplerate
target = "./audio/filteredAudio/filter.wav"
# A-weighting filter is 7, C-weighting is 5
# See https://en.wikipedia.org/wiki/A-weighting for a brief overview
f = aubio.digital_filter(7)
f.set_a_weighting(samplerate)
# create output file
o = aubio.sink(target, samplerate)
total_frames = 0
while True:
# read from source
samples, read = s()
# filter samples
filtered_samples = f(samples)
# write to sink
o(filtered_samples, read)
# count frames read
total_frames += read
# end of file reached
if read < s.hop_size:
break
# print some info
duration = total_frames / float(samplerate)
input_str = "input: {:s} ({:.2f} s, {:d} Hz)"
output_str = "output: {:s}, A-weighting filtered ({:d} frames total)"
print(input_str.format(s.uri, duration, samplerate))
print(output_str.format(o.uri, total_frames))
def check_write_and_read(self, samplerate, channels, hop_size, blocks,
write_samples):
expected_mono = np.sum(write_samples, axis=0) / write_samples.shape[0]
out = get_tmp_sink_path()
snk = aubio.sink(out, samplerate, channels=channels)
for i in range(blocks):
snk.do_multi(write_samples, hop_size)
# close the sink before reading from it
snk.close()
src = aubio.source(out, samplerate, hop_size)
for i in range(blocks):
read_samples, read = src.do_multi()
assert_equal(read_samples, write_samples)
assert_equal(read, hop_size)
src.seek(0)
for i in range(blocks):
read_samples, read = src()
assert_equal(read, hop_size)
assert_equal(read_samples, expected_mono)
def test_read_and_write_multi(self, hop_size, samplerate, path):
orig_samplerate = parse_file_samplerate(soundfile)
try:
if orig_samplerate is not None and orig_samplerate < samplerate:
# upsampling should emit a warning
with assert_warns(UserWarning):
f = source(soundfile, samplerate, hop_size)
else:
f = source(soundfile, samplerate, hop_size)
except RuntimeError as e:
err_msg = '{:s} (hop_s = {:d}, samplerate = {:d})'
skipTest(err_msg.format(str(e), hop_size, samplerate))
if samplerate == 0: samplerate = f.samplerate
sink_path = get_tmp_sink_path()
g = sink(sink_path, samplerate, channels = f.channels)
total_frames = 0
while True:
vec, read = f.do_multi()
g.do_multi(vec, read)
total_frames += read
if read < f.hop_size: break
del_tmp_sink_path(sink_path)
def test_read_and_write_multi(self, hop_size, samplerate, path):
orig_samplerate = parse_file_samplerate(soundfile)
try:
if orig_samplerate is not None and orig_samplerate < samplerate:
# upsampling should emit a warning
with assert_warns(UserWarning):
f = source(soundfile, samplerate, hop_size)
else:
f = source(soundfile, samplerate, hop_size)
except RuntimeError as e:
err_msg = '{:s} (hop_s = {:d}, samplerate = {:d})'
skipTest(err_msg.format(str(e), hop_size, samplerate))
if samplerate == 0: samplerate = f.samplerate
sink_path = get_tmp_sink_path()
g = sink(sink_path, samplerate, channels=f.channels)
total_frames = 0
while True:
vec, read = f.do_multi()
g.do_multi(vec, read)
total_frames += read
if read < f.hop_size: break
del_tmp_sink_path(sink_path)
def Robotize(inputfilename, outputfilename):
samplerate = 44100
f = source(inputfilename, samplerate, 256)
g = sink(outputfilename, samplerate)
total_frames, read = 0, 256
win_s = 512 # fft size
hop_s = win_s // 2 # hop size
pv = pvoc(win_s, hop_s) # phase vocoder
while read:
samples, read = f()
spectrum = pv(samples) # compute spectrum
# spectrum.norm *= .8 # reduce amplitude a bit
spectrum.phas[:] = 0. # zero phase
new_samples = pv.rdo(spectrum) # compute modified samples
g(new_samples, read) # write to output
total_frames += read
format_str = "read {:d} samples from {:s}, written to {:s}"
print(format_str.format(total_frames, f.uri, g.uri))
def test_many_sinks_not_closed(self):
from tempfile import mkdtemp
import os.path
import shutil
tmpdir = mkdtemp()
sink_list = []
try:
for i in range(many_files):
path = os.path.join(tmpdir, 'f-' + str(i) + '.wav')
g = sink(path, 0)
sink_list.append(g)
write = 256
for n in range(200):
vec = fvec(write)
g(vec, write)
except Exception:
pass
else:
self.fail("does not fail on too many files open")
for g in sink_list:
g.close()
shutil.rmtree(tmpdir)
def record_sink(sink_path):
"""Record an audio file using pysoundcard."""
from aubio import sink
from pysoundcard import Stream
hop_size = 256
duration = 5 # in seconds
s = Stream(blocksize = hop_size, channels = 1)
g = sink(sink_path, samplerate = int(s.samplerate))
s.start()
total_frames = 0
try:
while total_frames < duration * s.samplerate:
vec = s.read(hop_size)
# mix down to mono
mono_vec = vec.sum(-1) / float(s.channels[0])
g(mono_vec, hop_size)
total_frames += hop_size
except KeyboardInterrupt:
duration = total_frames / float(s.samplerate)
print("stopped after %.2f seconds" % duration)
s.stop()
def record_sink(sink_path):
"""Record an audio file using pysoundcard."""
from aubio import sink
from pysoundcard import Stream
hop_size = 256
duration = 5 # in seconds
s = Stream(blocksize=hop_size, channels=1)
g = sink(sink_path, samplerate=int(s.samplerate))
s.start()
total_frames = 0
try:
while total_frames < duration * s.samplerate:
vec = s.read(hop_size)
# mix down to mono
mono_vec = vec.sum(-1) / float(s.channels[0])
g(mono_vec, hop_size)
total_frames += hop_size
except KeyboardInterrupt:
duration = total_frames / float(s.samplerate)
print("stopped after %.2f seconds" % duration)
s.stop()
def test_wrong_samplerate(self):
with self.assertRaises(RuntimeError):
sink(get_tmp_sink_path(), -1)
def test_wrong_filename(self):
with self.assertRaises(RuntimeError):
sink('')
def test_wrong_channels_too_large(self):
with assert_raises(RuntimeError):
sink(get_tmp_sink_path(), 44100, 202020)
def slice_source_at_stamps(source_file, timestamps, timestamps_end=None,
output_dir=None, samplerate=0, hopsize=256):
""" slice a sound file at given timestamps """
if timestamps is None or len(timestamps) == 0:
raise ValueError("no timestamps given")
if timestamps[0] != 0:
timestamps = [0] + timestamps
if timestamps_end is not None:
timestamps_end = [timestamps[1] - 1] + timestamps_end
if timestamps_end is not None:
if len(timestamps_end) != len(timestamps):
raise ValueError("len(timestamps_end) != len(timestamps)")
else:
timestamps_end = [t - 1 for t in timestamps[1:]] + [_max_timestamp]
regions = list(zip(timestamps, timestamps_end))
#print regions
source_base_name, _ = os.path.splitext(os.path.basename(source_file))
if output_dir is not None:
if not os.path.isdir(output_dir):
os.makedirs(output_dir)
source_base_name = os.path.join(output_dir, source_base_name)
def new_sink_name(source_base_name, timestamp, samplerate):
""" create a sink based on a timestamp in samples, converted in seconds """
timestamp_seconds = timestamp / float(samplerate)
return source_base_name + "_%011.6f" % timestamp_seconds + '.wav'
# open source file
_source = source(source_file, samplerate, hopsize)
samplerate = _source.samplerate
total_frames = 0
slices = []
while True:
# get hopsize new samples from source
vec, read = _source.do_multi()
# if the total number of frames read will exceed the next region start
if len(regions) and total_frames + read >= regions[0][0]:
#print "getting", regions[0], "at", total_frames
# get next region
start_stamp, end_stamp = regions.pop(0)
# create a name for the sink
new_sink_path = new_sink_name(source_base_name, start_stamp, samplerate)
# create its sink
_sink = sink(new_sink_path, samplerate, _source.channels)
# create a dictionary containing all this
new_slice = {'start_stamp': start_stamp, 'end_stamp': end_stamp, 'sink': _sink}
# append the dictionary to the current list of slices
slices.append(new_slice)
for current_slice in slices:
start_stamp = current_slice['start_stamp']
end_stamp = current_slice['end_stamp']
_sink = current_slice['sink']
# sample index to start writing from new source vector
start = max(start_stamp - total_frames, 0)
# number of samples yet to written be until end of region
remaining = end_stamp - total_frames + 1
#print current_slice, remaining, start
# not enough frames remaining, time to split
if remaining < read:
if remaining > start:
# write remaining samples from current region
_sink.do_multi(vec[:, start:remaining], remaining - start)
#print "closing region", "remaining", remaining
# close this file
_sink.close()
elif read > start:
# write all the samples
_sink.do_multi(vec[:, start:read], read - start)
total_frames += read
if read < hopsize:
break
def slice_source_at_stamps(source_file, timestamps, timestamps_end=None,
output_dir=None, samplerate=0, hopsize=256,
create_first=False):
"""Slice a sound file at given timestamps.
This function reads `source_file` and creates slices, new smaller
files each starting at `t` in `timestamps`, a list of integer
corresponding to time locations in `source_file`, in samples.
If `timestamps_end` is unspecified, the slices will end at
`timestamps_end[n] = timestamps[n+1]-1`, or the end of file.
Otherwise, `timestamps_end` should be a list with the same length
as `timestamps` containing the locations of the end of each slice.
If `output_dir` is unspecified, the new slices will be written in
the current directory. If `output_dir` is a string, new slices
will be written in `output_dir`, after creating the directory if
required.
The default `samplerate` is 0, meaning the original sampling rate
of `source_file` will be used. When using a sampling rate
different to the one of the original files, `timestamps` and
`timestamps_end` should be expressed in the re-sampled signal.
The `hopsize` parameter simply tells :class:`source` to use this
hopsize and does not change the output slices.
If `create_first` is True and `timestamps` does not start with `0`, the
first slice from `0` to `timestamps[0] - 1` will be automatically added.
Parameters
----------
source_file : str
path of the resource to slice
timestamps : :obj:`list` of :obj:`int`
time stamps at which to slice, in samples
timestamps_end : :obj:`list` of :obj:`int` (optional)
time stamps at which to end the slices
output_dir : str (optional)
output directory to write the slices to
samplerate : int (optional)
samplerate to read the file at
hopsize : int (optional)
number of samples read from source per iteration
create_first : bool (optional)
always create the slice at the start of the file
Examples
--------
Create two slices: the first slice starts at the beginning of the
input file `loop.wav` and lasts exactly one second, starting at
sample `0` and ending at sample `44099`; the second slice starts
at sample `44100` and lasts until the end of the input file:
>>> aubio.slice_source_at_stamps('loop.wav', [0, 44100])
Create one slice, from 1 second to 2 seconds:
>>> aubio.slice_source_at_stamps('loop.wav', [44100], [44100 * 2 - 1])
Notes
-----
Slices may be overlapping. If `timestamps_end` is `1` element
shorter than `timestamps`, the last slice will end at the end of
the file.
"""
if not timestamps:
raise ValueError("no timestamps given")
if timestamps[0] != 0 and create_first:
timestamps = [0] + timestamps
if timestamps_end is not None:
timestamps_end = [timestamps[1] - 1] + timestamps_end
if timestamps_end is not None:
if len(timestamps_end) == len(timestamps) - 1:
timestamps_end = timestamps_end + [_max_timestamp]
elif len(timestamps_end) != len(timestamps):
raise ValueError("len(timestamps_end) != len(timestamps)")
else:
timestamps_end = [t - 1 for t in timestamps[1:]] + [_max_timestamp]
regions = list(zip(timestamps, timestamps_end))
source_base_name, _ = os.path.splitext(os.path.basename(source_file))
if output_dir is not None:
if not os.path.isdir(output_dir):
os.makedirs(output_dir)
source_base_name = os.path.join(output_dir, source_base_name)
def _new_sink_name(source_base_name, timestamp, samplerate):
# create name based on a timestamp in samples, converted in seconds
timestamp_seconds = timestamp / float(samplerate)
return source_base_name + "_%011.6f" % timestamp_seconds + '.wav'
# open source file
_source = source(source_file, samplerate, hopsize)
samplerate = _source.samplerate
total_frames = 0
slices = []
while True:
# get hopsize new samples from source
vec, read = _source.do_multi()
# if the total number of frames read will exceed the next region start
while regions and total_frames + read >= regions[0][0]:
# get next region
start_stamp, end_stamp = regions.pop(0)
# create a name for the sink
new_sink_path = _new_sink_name(source_base_name, start_stamp,
samplerate)
# create its sink
_sink = sink(new_sink_path, samplerate, _source.channels)
# create a dictionary containing all this
new_slice = {'start_stamp': start_stamp, 'end_stamp': end_stamp,
'sink': _sink}
# append the dictionary to the current list of slices
slices.append(new_slice)
for current_slice in slices:
start_stamp = current_slice['start_stamp']
end_stamp = current_slice['end_stamp']
_sink = current_slice['sink']
# sample index to start writing from new source vector
start = max(start_stamp - total_frames, 0)
# number of samples yet to written be until end of region
remaining = end_stamp - total_frames + 1
# not enough frames remaining, time to split
if remaining < read:
if remaining > start:
# write remaining samples from current region
_sink.do_multi(vec[:, start:remaining], remaining - start)
# close this file
_sink.close()
elif read > start:
# write all the samples
_sink.do_multi(vec[:, start:read], read - start)
total_frames += read
# remove old slices
slices = list(filter(lambda s: s['end_stamp'] > total_frames,
slices))
if read < hopsize:
break
# create python/tests/sounds if needed
output_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
output_dir = os.path.join(output_dir, 'tests', 'sounds')
if not os.path.isdir(output_dir):
os.makedirs(output_dir)
filenames = ['44100Hz_1f_silence.wav',
'22050Hz_5s_brownnoise.wav',
'32000Hz_127f_sine440.wav',
]
samplerates = [44100, 22050, 32000]
durations = [1, 5*22050, 127]
for fname, samplerate, duration in zip(filenames, samplerates, durations):
output_name = os.path.join(output_dir, fname)
g = sink(output_name, samplerate)
total_frames = 0
while total_frames < duration:
write = min(hop_size, duration - total_frames)
if 'brownnoise' in fname:
vec = np.random.rand(write).astype(float_type) * 2. - 1.
elif 'sine' in fname:
freq = 440
t = np.arange(write).astype(float_type) + total_frames
vec = np.sin(2. * np.pi * freq * t / float(samplerate))
else:
# silence
vec = fvec(write)
g(vec, write)
total_frames += write
outstr = "wrote {:2f}s".format(total_frames / float(samplerate))
def test_wrong_samplerate(self):
with assert_raises(RuntimeError):
sink(get_tmp_sink_path(), -1)
if len(sys.argv) < 3:
print "usage: %s <inputfile> <duration>" % sys.argv[0]
sys.exit(1)
source_file = sys.argv[1]
duration = float(sys.argv[2])
source_base_name, source_ext = os.path.splitext(os.path.basename(source_file))
hopsize = 256
slice_n, total_frames_written, read = 0, 0, hopsize
def new_sink_name(source_base_name, slice_n, duration=duration):
return source_base_name + "_%02.3f" % (slice_n * duration) + ".wav"
f = source(source_file, 0, hopsize)
samplerate = f.samplerate
g = sink(new_sink_name(source_base_name, slice_n), samplerate)
# print "new slice:", slice_n, 0, "+", 0, "=", 0
while read == hopsize:
vec, read = f()
start_of_next_region = int(duration * samplerate * (slice_n + 1))
remaining = start_of_next_region - total_frames_written
# number of samples remaining is less than what we got
if remaining <= read:
# write remaining samples from current region
g(vec[0:remaining], remaining)
# close this file
del g
# print "new slice", slice_n, total_frames_written, "+", remaining, "=", start_of_next_region
slice_n += 1
# create a new file for the new region
def test_wrong_samplerate_too_large(self):
with assert_raises(RuntimeError):
sink(get_tmp_sink_path(), 1536001, 2)
import sys
from aubio import source, sink, pvoc, tss
if __name__ == '__main__':
if len(sys.argv) < 2:
print('usage: %s <inputfile> <outputfile_transient> <outputfile_steady>' % sys.argv[0])
sys.exit(1)
samplerate = 44100
win_s = 1024 # fft size
hop_s = win_s // 4 # block size
threshold = 0.5
f = source(sys.argv[1], samplerate, hop_s)
g = sink(sys.argv[2], samplerate)
h = sink(sys.argv[3], samplerate)
pva = pvoc(win_s, hop_s) # a phase vocoder
pvb = pvoc(win_s, hop_s) # another phase vocoder
t = tss(win_s, hop_s) # transient steady state separation
t.set_threshold(threshold)
read = hop_s
while read:
samples, read = f() # read file
spec = pva(samples) # compute spectrum
trans_spec, stead_spec = t(spec) # transient steady-state separation
transients = pva.rdo(trans_spec) # overlap-add synthesis of transients
def test_wrong_channels(self):
with assert_raises(RuntimeError):
sink(get_tmp_sink_path(), 44100, -1)
def test_wrong_samplerate_too_large(self):
with self.assertRaises(RuntimeError):
sink(get_tmp_sink_path(), 1536001, 2)
def test_wrong_channels_too_large(self):
with self.assertRaises(RuntimeError):
sink(get_tmp_sink_path(), 44100, 202020)
def test_wrong_filename(self):
with assert_raises(RuntimeError):
sink('')
#! /usr/bin/env python
import sys
from aubio import source, sink, pvoc
if __name__ == '__main__':
if len(sys.argv) < 2:
print 'usage: %s <inputfile> <outputfile>' % sys.argv[0]
sys.exit(1)
samplerate = 44100
f = source(sys.argv[1], samplerate, 256)
g = sink(sys.argv[2], samplerate)
total_frames, read = 0, 256
win_s = 512 # fft size
hop_s = win_s / 2 # hop size
pv = pvoc(win_s, hop_s) # phase vocoder
while read:
samples, read = f()
spectrum = pv(samples) # compute spectrum
#spectrum.norm *= .8 # reduce amplitude a bit
spectrum.phas[:] = 0. # zero phase
new_samples = pv.rdo(spectrum) # compute modified samples
g(new_samples, read) # write to output
total_frames += read
print "wrote", total_frames, "from", f.uri, "to", g.uri
import os
import pandas as pd
import librosa
import glob
import matplotlib
import librosa.display
# Read and load input
input_wav = raw_input("Enter input file name: ")
output_wav = raw_input("Enter output file name: ")
# Manipulate audio to add robot effect
samplerate = 44100
f = source(input_wav, samplerate, 256)
g = sink(output_wav, samplerate)
total_frames, read = 0, 256
win_s = 512 # fft size
hop_s = win_s // 2 # hop size
pv = pvoc(win_s, hop_s) # phase vocoder
while read:
samples, read = f()
spectrum = pv(samples) # compute spectrum
#spectrum.norm *= .8 # reduce amplitude a bit
spectrum.phas[:] = 0. # zero phase
new_samples = pv.rdo(spectrum) # compute modified samples
g(new_samples, read) # write to output
total_frames += read
output_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
output_dir = os.path.join(output_dir, 'tests', 'sounds')
if not os.path.isdir(output_dir):
os.makedirs(output_dir)
filenames = [
'44100Hz_1f_silence.wav',
'22050Hz_5s_brownnoise.wav',
'32000Hz_127f_sine440.wav',
]
samplerates = [44100, 22050, 32000]
durations = [1, 5 * 22050, 127]
for fname, samplerate, duration in zip(filenames, samplerates, durations):
output_name = os.path.join(output_dir, fname)
g = sink(output_name, samplerate)
total_frames = 0
while total_frames < duration:
write = min(hop_size, duration - total_frames)
if 'brownnoise' in fname:
vec = np.random.rand(write).astype(float_type) * 2. - 1.
elif 'sine' in fname:
freq = 440
t = np.arange(write).astype(float_type) + total_frames
vec = np.sin(2. * np.pi * freq * t / float(samplerate))
else:
# silence
vec = fvec(write)
g(vec, write)
total_frames += write
outstr = "wrote {:2f}s".format(total_frames / float(samplerate))
# open stream
buffer_size = 1024
pyaudio_format = pyaudio.paFloat32
n_channels = 1
samplerate = 44100
stream = p.open(format=pyaudio_format,
channels=n_channels,
rate=samplerate,
input=True,
frames_per_buffer=buffer_size)
if len(sys.argv) > 1:
# record 5 seconds
output_filename = sys.argv[1]
record_duration = 5 # exit 1
outputsink = aubio.sink(sys.argv[1], samplerate)
total_frames = 0
else:
# run forever
outputsink = None
record_duration = None
# setup pitch
tolerance = 0.8
win_s = 4096 # fft size
hop_s = buffer_size # hop size
pitch_o = aubio.pitch("default", win_s, hop_s, samplerate)
pitch_o.set_unit("midi")
pitch_o.set_tolerance(tolerance)
print("*** starting recording")
def Stretch(inputfilename, outputfilename, stretchfactor, sample_rate=0):
win_s = 1024
hop_s = win_s // 8 # 87.5 % overlap
warmup = win_s // hop_s - 1
source_filename = inputfilename
output_filename = outputfilename
rate = float(stretchfactor)
samplerate = sample_rate
source_in = source(source_filename, samplerate, hop_s)
samplerate = source_in.samplerate
p = pvoc(win_s, hop_s)
# allocate memory to store norms and phases
n_blocks = source_in.duration // hop_s + 1
# adding an empty frame at end of spectrogram
norms = np.zeros((n_blocks + 1, win_s // 2 + 1), dtype=float_type)
phases = np.zeros((n_blocks + 1, win_s // 2 + 1), dtype=float_type)
block_read = 0
while True:
# read from source
samples, read = source_in()
# compute fftgrain
spec = p(samples)
# store current grain
norms[block_read] = spec.norm
phases[block_read] = spec.phas
# until end of file
if read < hop_s: break
# increment block counter
block_read += 1
# just to make sure
#source_in.close()
sink_out = sink(output_filename, samplerate)
# interpolated time steps (j = alpha * i)
steps = np.arange(0, n_blocks, rate, dtype=float_type)
# initial phase
phas_acc = phases[0]
# excepted phase advance in each bin
phi_advance = np.linspace(0, np.pi * hop_s,
win_s / 2 + 1).astype(float_type)
new_grain = cvec(win_s)
for (t, step) in enumerate(steps):
frac = 1. - np.mod(step, 1.0)
# get pair of frames
t_norms = norms[int(step):int(step + 2)]
t_phases = phases[int(step):int(step + 2)]
# compute interpolated frame
new_grain.norm = frac * t_norms[0] + (1. - frac) * t_norms[1]
new_grain.phas = phas_acc
#print t, step, new_grain.norm
#print t, step, phas_acc
# psola
samples = p.rdo(new_grain)
if t > warmup: # skip the first few frames to warm up phase vocoder
# write to sink
sink_out(samples, hop_s)
# calculate phase advance
dphas = t_phases[1] - t_phases[0] - phi_advance
# unwrap angle to [-pi; pi]
dphas = unwrap2pi(dphas)
# cumulate phase, to be used for next frame
phas_acc += phi_advance + dphas
for t in range(warmup + 1): # purge the last frames from the phase vocoder
new_grain.norm[:] = 0
new_grain.phas[:] = 0
samples = p.rdo(new_grain)
sink_out(samples, read if t == warmup else hop_s)
# just to make sure
#sink_out.close()
format_out = "read {:d} blocks from {:s} at {:d}Hz and rate {:f}, wrote {:d} blocks to {:s}"
print(
format_out.format(block_read, source_filename, samplerate, rate,
len(steps), output_filename))
period = float(samplerate) / pitch
# create a sine lookup table
tablelen = 1000
sinetable = arange(tablelen + 1, dtype = 'float32')
sinetable = 0.7 * sin(twopi * sinetable/tablelen)
sinetone = zeros((duration,), dtype = 'float32')
# compute sinetone at floating point period
for i in range(duration):
x = int((i % period) / float(period) * tablelen)
sinetone[i] = (sinetable[x] + sinetable[x+1]) / 2
# apply some envelope
float_ramp = arange(duration, dtype = 'float32')
sinetone *= exp( - e * float_ramp / duration / decay)
sinetone[:attack] *= exp( e * ( float_ramp[:attack] / attack - 1 ) )
if 0:
import matplotlib.pyplot as plt
plt.plot(sinetone)
plt.show()
my_sink = sink(sys.argv[1], samplerate)
total_frames = 0
while total_frames + blocksize < duration:
my_sink(sinetone[total_frames:total_frames+blocksize], blocksize)
total_frames += blocksize
my_sink(sinetone[total_frames:duration], duration - total_frames)
# read from source
samples, read = source_in()
# compute fftgrain
spec = p(samples)
# store current grain
norms[block_read] = spec.norm
phases[block_read] = spec.phas
# until end of file
if read < hop_s: break
# increment block counter
block_read += 1
# just to make sure
#source_in.close()
sink_out = sink(output_filename, samplerate)
# interpolated time steps (j = alpha * i)
steps = np.arange(0, n_blocks, rate, dtype = float_type)
# initial phase
phas_acc = phases[0]
# excepted phase advance in each bin
phi_advance = np.linspace(0, np.pi * hop_s, win_s / 2 + 1).astype (float_type)
new_grain = cvec(win_s)
for (t, step) in enumerate(steps):
frac = 1. - np.mod(step, 1.0)
# get pair of frames
t_norms = norms[int(step):int(step+2)]
import sys
from aubio import source, sink
if __name__ == '__main__':
if len(sys.argv) < 3:
print('usage: %s <inputfile> <outputfile> [samplerate] [hop_size]' % sys.argv[0])
sys.exit(1)
if len(sys.argv) > 3: samplerate = int(sys.argv[3])
else: samplerate = 0
if len(sys.argv) > 4: hop_size = int(sys.argv[4])
else: hop_size = 256
f = source(sys.argv[1], samplerate, hop_size)
if samplerate == 0: samplerate = f.samplerate
g = sink(sys.argv[2], samplerate, f.channels)
total_frames, read = 0, hop_size
while read:
vec, read = f.do_multi()
g.do_multi(vec, read)
total_frames += read
outstr = "wrote %.2fs" % (total_frames / float(samplerate))
outstr += " (%d frames in" % total_frames
outstr += " %d blocks" % (total_frames // f.hop_size)
outstr += " of %d channels" % f.channels
outstr += " at %dHz)" % f.samplerate
outstr += " from " + f.uri
outstr += " to " + g.uri
print(outstr)