def overlayAudio(offset, interval):
onsetMarker = AudioOnsetsMarker(onsets=1.0 * beats[offset::interval])
audioMarked = onsetMarker(audio)
writer = MonoWriter(filename='test.wav')
beginIndex = 0.25 * np.size(audioMarked)
endIndex = 0.35 * np.size(audioMarked)
writer(audioMarked[beginIndex:endIndex]) #Only write fragment
def _flush_save_audio_buffer(self, queue):
while True:
filename, audio, tracklist = queue.get()
if not (filename is None):
logger.debug('Saving {} to disk'.format(filename))
writer = MonoWriter(filename=filename, bitrate=320,format='mp3')
writer(np.array(audio,dtype='single'))
# Save tracklist
with open(self.save_dir_tracklist,'a+') as csvfile:
writer = csv.writer(csvfile)
for line in tracklist:
writer.writerow([line])
else:
logger.debug('Stopping audio saving thread!')
return
def thresholdAudio(audioPath='testDownload/', t=-30, fs=44100):
'''
Run thresholdAudio for trim all de mp3 audio files in a audioPath/../.. where the first .. is related
with the queryText and the second .. is related with the location of freesound track.
It was thougt to use after soundDownload.py from sms-tool package at freesound source.
:param audioPath: path where sounds where download (possible path used for soundDownload.py). Default: testDownload/
:param t: threshold (in dB) to trim audiofiles related to max value of file. Default: -30
:param fs: fs for the output sound. Default: 44100
:return: print(Done!!)
'''
thTimes = 10 ** (t/20) #threshold: dB to times
instrument = ls(audioPath) #read the queryText path inside the given path
audioTrack = [ls(str(key)) for key in instrument] #read the different folder inside each queryText path
a, b = np.shape(audioTrack) #size of the matrix
finalArray = [os.path.join(str(audioTrack[i][j]), arch.name)
for i in np.arange(a) for j in np.arange(b) for arch in
Path(str(audioTrack[i][j])).iterdir()
if arch.name.endswith('.mp3')] #array for each track
for key in finalArray:
track = MonoLoader(filename=key, sampleRate=fs)() #read audio and transform into mono
maximo = np.max(abs(track)) #set the abs maximum
i = 0
j = -1
while abs(track[i]) < maximo * thTimes: #find the first significant value
i += 1
while abs(track[j]) < maximo * thTimes: #find the last significant value
j -= 1
shortTrack = track[i:j] #build the trimed track
MonoWriter(filename=key + 'computed.wav')(shortTrack) #write the file at same location of given
print('Done!!')
def convert_to_wav(audiofile, samplerate=44100):
logger.debug('{0}: converting to WAV'.format(audiofile))
# Convert to WAV using Essentia so that timbral models always read WAV file
output_filename = '/tmp/{0}-converted.wav'.format(str(uuid.uuid4()))
audio = MonoLoader(filename=audiofile, sampleRate=samplerate)()
MonoWriter(filename=output_filename, format='wav', sampleRate=samplerate)(audio)
return output_filename
def get(inputDir, descExt):
exception = {}
output = {}
for path, dname, fnames in os.walk(
inputDir
): # dname directories names in current directory, fnames file names in current directory.
for fname in fnames:
if descExt in fname.lower():
new_pid = os.fork()
if new_pid == 0: # Si new_pid == 0 > forked process.
try: # uso un try..except..finally para asegurarme de _siempre_ terminar el proceso
file_name = path + "/" + fname
[Sound, Fs, nChannels, md5, bit_rate,
codec] = AudioLoader(filename=file_name)() # lo cargo
Sound = Sound[:,
0] # El algoritmo siempre tira dos canales
print file_name
impulse, FsImpulse = LoadImpulse.get_impulse()
impulse = impulse.astype('float32',
casting='same_kind')
if Fs != FsImpulse:
Rs = Resample(inputSampleRate=FsImpulse,
outputSampleRate=Fs)
impulse = Rs(impulse)
final = np.convolve(Sound, impulse)
if descExt == '.aif': descExt = '.aiff'
mw = MonoWriter(filename=path + '/R1_' + fname,
sampleRate=Fs,
format=descExt.split('.')[1])
mw(final)
print 'Done!'
except Exception:
exception[fname] = [
'oops'
] # De esta forma puedo fijarme si hubo alguna excepcion
# pass
finally:
os._exit(0) # Cierro el fork y vuelvo al proceso padre
else:
child = new_pid
os.waitpid(
child, 0
) # evito crear procesos paralelos, lo que limita la performance de mi programa
return exception
def VCTK(model_config):
print("Preprocessing VCTK dataset")
VCTK_path = model_config["raw_data_path"] + "/VCTK"
VCTK_preprocessed_path = model_config[
"preprocessed_data_path"] + "/VCTK_8k_DBE"
clean_dirs = ["/clean_trainset_wav", "/clean_testset_wav"]
noisy_dirs = ["/noisy_trainset_wav", "/noisy_testset_wav"]
# copy clean dirs
for clean_dir in clean_dirs:
shutil.copytree(VCTK_path + clean_dir,
VCTK_preprocessed_path + clean_dir)
# create dirs
for noisy_dir in noisy_dirs:
noisy8k_dir = VCTK_preprocessed_path + noisy_dir.replace(
"noisy", "noisy8k")
os.makedirs(noisy8k_dir)
# preprocessing
for root, dirs, files in os.walk(VCTK_path + noisy_dir):
for file in files:
if file.endswith('.wav'):
# read audio
file_name = os.path.join(root, file)
noisy = MonoLoader(filename=file_name, sampleRate=44100)()
noisy8k = MonoLoader(filename=file_name, sampleRate=8000)()
# resample audio
noisy8k_resampled = Resample(
inputSampleRate=8000, outputSampleRate=44100)(noisy8k)
# lengths
len_noisy = len(noisy)
len_noisy8k_resampled = len(noisy8k_resampled)
# trimming/appending
len_diff = len_noisy8k_resampled - len_noisy
if len_diff > 0:
noisy8k_resampled = noisy8k_resampled[:len_noisy]
elif len_diff < 0:
noisy8k_resampled = np.pad(noisy8k_resampled,
(0, abs(len_diff)),
'constant',
constant_values=(0, 0))
# write audio
output_name = noisy8k_dir + "/" + file.split(
".")[0] + "_8k.wav"
MonoWriter(filename=output_name,
sampleRate=44100)(noisy8k_resampled)
def process_data_and_extract_profiles(
segment_id: int, song_file
) -> Tuple[int, Tuple, Tuple, Tuple, Tuple, Tuple, Tuple, Tuple]:
"""Extracts BPM, timbre and moods from a segmented songfile
Parameters
----------
segment_id
id of the segment of the audiofile corresponds to
song_file
the audio from a segmented song
Returns
-------
Tuple[int, Tuple, Tuple, Tuple, Tuple, Tuple, Tuple, Tuple]
A tuple of the segment_id and tuples describing BPM,
timbre and all moods
"""
# creating the temporary files
temp_song = tempfile.NamedTemporaryFile(delete=True, suffix='.wav')
temp_classifier = tempfile.NamedTemporaryFile(delete=True)
MonoWriter(filename=temp_song.name)(song_file)
make_low_level_data_file(temp_song.name, temp_classifier.name)
# closing and effectively deleting the song tempfile
temp_song.close()
bpm, beats_confidence = get_song_bpm(song_file)
bpm_tuple = (bpm, beats_confidence)
(timbre, mood_relaxed, mood_party, mood_aggressive, mood_happy,
mood_sad) = get_classifier_data(temp_classifier.name)
# closing and effectively deleting the classifier tempfile
temp_classifier.close()
return (segment_id, bpm_tuple, timbre, mood_relaxed, mood_party,
mood_aggressive, mood_happy, mood_sad)
def _flush_save_audio_buffer(self, queue):
while True:
filename, audio, tracklist = queue.get()
if not (filename is None):
logger.debug('Saving {} to disk, length {}'.format(
filename, len(audio)))
if self.stereo:
writer = AudioWriter(filename=filename, format='wav')
else:
writer = MonoWriter(filename=filename, format='wav')
writer(np.array(audio.T, dtype='single'))
# Save tracklist
logger.debug('Saving tracklist')
with open(self.save_dir_tracklist, 'a+') as csvfile:
writer = csv.writer(csvfile)
for line in tracklist:
writer.writerow([line])
else:
logger.debug('Stopping audio saving thread!')
return
# BPM_floor = np.floor(BPM) # valid_BPMs = [BPM_floor - 1, BPM_floor, BPM_floor+1, BPM_floor+2] # form bpm in valid_BPMs: # sPerBeat = (60./BPM) # beatsInSong = (len(audio)/44100.) / sPerBeat # perfectBeats = sPerBeat * range(int(beatsInSong) - 1) # One beat shorter than the original file for phase aligning # np.correlate(beats, perfectBeats, mode = 'valid') # Overlay the audio file with onsets onsetMarker = AudioOnsetsMarker(onsets=beats) audioMarked = onsetMarker(audio / 2.) # Stretch the result #from librosa.effects import time_stretch #audioMarked = time_stretch(audioMarked, 175./172.) # Output the marked file writer = MonoWriter(filename='test.wav') writer(audioMarked[:]) #Only write fragment # Play the result from subprocess import call call(["mplayer", 'test.wav']) # Display the waveform #print 'Displaying waveform' #plt.plot(audio[::4410]) # 44.1 kHz, plotted every 4410th sample => 10 samples per second #for b in beats: # plt.axvline(x=b*10) #plt.show() # unnecessary if you started "ipython --pylab"
def get_fool(self, songid, stretch_rate):
audio = np.load(os.path.join(self.npy_path, songid)+'.npy')
t_audio = librosa.effects.time_stretch(audio, stretch_rate)
MonoWriter(filename=os.path.join(self.fool_audio_path, songid)+'.wav', format='wav', sampleRate=self.fs)(t_audio)
np.save(os.path.join(self.fool_npy_path, songid)+'.npy', t_audio)
def DAPS(model_config):
print("Preprocessing DAPS dataset")
DAPS_path = model_config["raw_data_path"] + "/DAPS/"
DAPS_preprocessed_path = model_config[
"preprocessed_data_path"] + "/DAPS_8k_DBE/"
# create processed dir
if not os.path.exists(DAPS_preprocessed_path):
os.makedirs(DAPS_preprocessed_path)
devices_and_rooms = [
'ipad_bedroom1', 'ipad_confroom1', 'ipad_confroom2',
'ipad_livingroom1', 'ipad_office1', 'ipad_office2',
'ipadflat_confroom1', 'ipadflat_office1', 'iphone_bedroom1',
'iphone_livingroom1'
] #'ipad_balcony1' and 'iphone_balcony1' excluded
duration = 5
target = "clean"
# Train samples
for i in np.arange(9):
subject = str(i + 1)
scripts = [str(i % 5 + 1),
str((i + 1) % 5 + 1)] # 2 scripts per subject
for script in scripts:
# Female
target_file = DAPS_path + target + "/f" + subject + "_script" + script + "_" + target + ".wav"
target_audio = MonoLoader(filename=target_file, sampleRate=44100)()
starts = np.arange(0, len(target_audio), 44100 * duration)
for j, start in enumerate(starts):
# output dir
output_dir = DAPS_preprocessed_path + target
if not os.path.exists(output_dir):
os.makedirs(output_dir)
if len(target_audio[start:]) > (
44100 *
duration): # check if there is a complete excerpt
audio_to_write = target_audio[starts[j]:starts[j + 1]]
output_name = output_dir + "/f" + subject + "_script" + script + "_" + target + "_" + str(
j) + ".wav"
MonoWriter(filename=output_name,
sampleRate=44100)(audio_to_write)
for device_and_room in devices_and_rooms:
device_path = DAPS_path + device_and_room
device_file = device_path + "/f" + subject + "_script" + script + "_" + device_and_room + ".wav"
device_audio = MonoLoader(filename=device_file,
sampleRate=44100)()
starts = np.arange(0, len(device_audio), 44100 * duration)
for j, start in enumerate(starts):
# output dir
output_dir = DAPS_preprocessed_path + device_and_room
if not os.path.exists(output_dir):
os.makedirs(output_dir)
if len(device_audio[start:]) > (44100 * duration):
audio_to_write = device_audio[starts[j]:starts[j + 1]]
audio_44k_to_8k = Resample(
inputSampleRate=44100,
outputSampleRate=8000)(audio_to_write)
audio_8k_to_44k = Resample(
inputSampleRate=8000,
outputSampleRate=44100)(audio_44k_to_8k)
len_audio_44k = len(audio_to_write)
len_audio_44k_resampled = len(audio_8k_to_44k)
# trimming/appending
len_diff = len_audio_44k_resampled - len_audio_44k
if len_diff > 0:
audio_8k_to_44k = audio_8k_to_44k[:len_audio_44k]
elif len_diff < 0:
audio_8k_to_44k = np.pad(audio_8k_to_44k,
(0, abs(len_diff)),
'constant',
constant_values=(0, 0))
output_name = output_dir + "/f" + subject + "_script" + script + "_" + device_and_room + "_" + str(
j) + ".wav"
MonoWriter(filename=output_name,
sampleRate=44100)(audio_8k_to_44k)
# Male
target_file = DAPS_path + target + "/m" + subject + "_script" + script + "_" + target + ".wav"
target_audio = MonoLoader(filename=target_file, sampleRate=44100)()
starts = np.arange(0, len(target_audio), 44100 * duration)
for j, start in enumerate(starts):
# output dir
output_dir = DAPS_preprocessed_path + target
if not os.path.exists(output_dir):
os.makedirs(output_dir)
if len(target_audio[start:]) > (
44100 *
duration): # check if there is a complete excerpt
audio_to_write = target_audio[starts[j]:starts[j + 1]]
output_name = output_dir + "/m" + subject + "_script" + script + "_" + target + "_" + str(
j) + ".wav"
MonoWriter(filename=output_name,
sampleRate=44100)(audio_to_write)
for device_and_room in devices_and_rooms:
device_path = DAPS_path + device_and_room
device_file = device_path + "/m" + subject + "_script" + script + "_" + device_and_room + ".wav"
device_audio = MonoLoader(filename=device_file,
sampleRate=44100)()
starts = np.arange(0, len(device_audio), 44100 * duration)
for j, start in enumerate(starts):
# output dir
output_dir = DAPS_preprocessed_path + device_and_room
if not os.path.exists(output_dir):
os.makedirs(output_dir)
if len(device_audio[start:]) > (44100 * duration):
audio_to_write = device_audio[starts[j]:starts[j + 1]]
audio_44k_to_8k = Resample(
inputSampleRate=44100,
outputSampleRate=8000)(audio_to_write)
audio_8k_to_44k = Resample(
inputSampleRate=8000,
outputSampleRate=44100)(audio_44k_to_8k)
len_audio_44k = len(audio_to_write)
len_audio_44k_resampled = len(audio_8k_to_44k)
# trimming/appending
len_diff = len_audio_44k_resampled - len_audio_44k
if len_diff > 0:
audio_8k_to_44k = audio_8k_to_44k[:len_audio_44k]
elif len_diff < 0:
audio_8k_to_44k = np.pad(audio_8k_to_44k,
(0, abs(len_diff)),
'constant',
constant_values=(0, 0))
output_name = output_dir + "/m" + subject + "_script" + script + "_" + device_and_room + "_" + str(
j) + ".wav"
MonoWriter(filename=output_name,
sampleRate=44100)(audio_8k_to_44k)
# Test samples
# Female
target_file = DAPS_path + target + "/f10_script5_" + target + ".wav"
target_audio = MonoLoader(filename=target_file, sampleRate=44100)()
output_name = DAPS_preprocessed_path + target + "/f10_script5_" + target + ".wav"
MonoWriter(filename=output_name, sampleRate=44100)(target_audio)
for device_and_room in devices_and_rooms:
device_path = DAPS_path + device_and_room
device_file = device_path + "/f10_script5_" + device_and_room + ".wav"
device_audio = MonoLoader(filename=device_file, sampleRate=8000)()
audio_8k_to_44k = Resample(inputSampleRate=8000,
outputSampleRate=44100)(device_audio)
len_audio_44k = len(target_audio)
len_audio_44k_resampled = len(audio_8k_to_44k)
# trimming/appending
len_diff = len_audio_44k_resampled - len_audio_44k
if len_diff > 0:
audio_8k_to_44k = audio_8k_to_44k[:len_audio_44k]
elif len_diff < 0:
audio_8k_to_44k = np.pad(audio_8k_to_44k, (0, abs(len_diff)),
'constant',
constant_values=(0, 0))
output_name = DAPS_preprocessed_path + device_and_room + "/f10_script5_" + device_and_room + ".wav"
MonoWriter(filename=output_name, sampleRate=44100)(audio_8k_to_44k)
# Male
target_file = DAPS_path + target + "/m10_script5_" + target + ".wav"
target_audio = MonoLoader(filename=target_file, sampleRate=44100)()
output_name = DAPS_preprocessed_path + target + "/m10_script5_" + target + ".wav"
MonoWriter(filename=output_name, sampleRate=44100)(target_audio)
for device_and_room in devices_and_rooms:
device_path = DAPS_path + device_and_room
device_file = device_path + "/m10_script5_" + device_and_room + ".wav"
device_audio = MonoLoader(filename=device_file, sampleRate=8000)()
audio_8k_to_44k = Resample(inputSampleRate=8000,
outputSampleRate=44100)(device_audio)
len_audio_44k = len(target_audio)
len_audio_44k_resampled = len(audio_8k_to_44k)
# trimming/appending
len_diff = len_audio_44k_resampled - len_audio_44k
if len_diff > 0:
audio_8k_to_44k = audio_8k_to_44k[:len_audio_44k]
elif len_diff < 0:
audio_8k_to_44k = np.pad(audio_8k_to_44k, (0, abs(len_diff)),
'constant',
constant_values=(0, 0))
output_name = DAPS_preprocessed_path + device_and_room + "/m10_script5_" + device_and_room + ".wav"
MonoWriter(filename=output_name, sampleRate=44100)(audio_8k_to_44k)
def save_audio(audio, filename, file_format='wav', bit_rate=320):
from essentia.standard import MonoWriter
MonoWriter(filename=filename, bitrate=bit_rate, format=file_format)(audio)