def __init__(self, filename, filehash = True):
AudioFile.__init__(self, filename, filehash)
self.type = 'MPEG-3 audio stream'
self.audio = mp3.MP3(self.file)
self.__load_tags()
self.__load_info()
del self.audio
def __init__(self, filename, filehash = True):
AudioFile.__init__(self, filename, filehash)
self.type = 'MPEG-4 audio stream'
self.aac = mp4.MP4(filename)
self.__load_tags()
self.__load_info()
del self.aac
def check(self):
answer = self.answerEdit.text()
if self.audiofile.check_answer(answer):
self.correct_nb += 1
self.correct_nbLabel.setText(str(self.correct_nb))
self.audiofile = AudioFile()
else:
self.wrong_nb += 1
self.wrong_nbLabel.setText(str(self.wrong_nb))
_stats.add_stats(
answer,
self.audiofile.get_pinyin(),
self.audiofile.get_id(),
)
self.answerEdit.clear()
def getListFromPath(self, parent, index=0):
logger.debug( "Getting contents from: %s" % parent )
try:
if parent == 0: parent = "/"
dirlist = os.listdir(parent)
dirlist.sort()
except Exception, e: logger.error( "at Browser.getListFromPath(): " + str(e) )
list = []
for x in dirlist:
name, ext = os.path.splitext(x)
if x != get_ascii_filename(x):
continue
real_path = os.path.join(parent, uniconvert2(x))
if os.path.isfile(real_path):
if ext.lower() == '.ogg' or ext.lower() == '.mp3':
# read vorbis info
self.af = AudioFile(real_path)
time = int(self.af.getLength())
length = " (" + str(time) + ")"
list.append( {'location':real_path, 'id':'idd', 'img':'imgg', 'name': x + length, 'time': time, 'seconds':time } )
else:
list.append( {'location':real_path, 'id':'idd', 'img':'imgg', 'name': "/" + x } )
return list
def biteval(filename):
audio = AudioFile(filename)
cnt = [0] * audio.bitDepth
audio = np.array([2**(audio.bitDepth - 1) * v for v in audio])
for v in audio[:1000]:
bin = convert_to_bin_array(int(v), len(cnt))
cnt = [i + j for i, j in zip(cnt, bin)]
print(cnt)
inp = input("Does this file have BIT DEPTH problems? [y/n]")
return u.str2bool(inp)
def searchFood(term,
limit=5,
sort=2,
radius_filter=8000): #sort = 2 is for highested rated
params = {
'term': term,
'limit': limit,
'sort': sort,
'radius_filter': radius_filter,
'lang': 'fr'
}
result = client.search('Pittsburgh', **params)
businesses = result.businesses
names = ""
for i in range(0, limit):
if i == limit - 1:
names += "and " + businesses[i].name
else:
names += businesses[i].name + ", "
text_to_speech = TextToSpeechV1(
username='******',
password='******',
x_watson_learning_opt_out=True) # Optional flag
with open(join(dirname(__file__), 'resources/recommendations.wav'),
'wb') as audio_file:
audio_file.write(
text_to_speech.synthesize('Some good' + term + 'places are, ' +
names,
accept='audio/wav',
voice="en-US_AllisonVoice"))
audio = AudioFile('resources/recommendations.wav')
audio.play()
audio.close()
def __init__(self, fn, userInfo, mode='publish'):
self.fn = fn
self.userInfo = userInfo
self.mode = mode
self.warnIncomplete = 1
tags = ''
self.config = LoadConfig.LoadConfig()
# open ogg file
self.af = AudioFile(fn)
self.metadata = self.af.read_comments()
if self.mode is 'publish':
newFilePath = self.fn
logger.debug5("ImportOGG() self.fn: " + self.fn)
logger.debug5("ImportOGG() newFilePath: " + newFilePath)
if newFilePath:
if self.import2db(newFilePath):
self.setImported()
elif self.mode is 'edit':
if self.import2db(self.fn):
logger.debug1("Updated database from file: %s" % self.fn)
def augment(input_dir: Path, output_dir: Path):
seed(SEED)
if not input_dir.exists:
raise Exception('Input directory does not exist.')
if not output_dir.exists:
print("Making output directory {}".format(output_dir))
output_dir.mkdir(parents=True)
filelist = set([x for x in input_dir.glob('*.wav')])
print("{} input files found".format(len(filelist)))
while len(filelist) > 1:
print("{} files remaining...".format(len(filelist)))
f1 = choice(tuple(filelist))
filelist.remove(f1)
f2 = choice(tuple(filelist))
filelist.remove(f2)
# load audio files and apply a random amount of:
# gain reduction in steps of -3 db from [-15 db, 0 db]
# varispeed between [0.9, 1.1]
reduction = [0, -3, -6, -12, -15]
f1 = AudioFile(path=f1).varispeed(uniform(0.9,
1.1)).gain(choice(reduction))
f2 = AudioFile(path=f2).varispeed(uniform(0.9,
1.1)).gain(choice(reduction))
# mix two audio files - random amount of overlap from [0.5 to 1.5]
f1.mix(audio=f2, relative_start=uniform(0.5, 1.5))
# add a random amounts of silence [0 sec, 5 sec] before and after audio
f1.add_silence(sec_before=uniform(0, 5), sec_after=uniform(0, 5))
# save as new clean file
filename = f1.filename.split(".wav")[0] + "_" + f2.filename
f1.save(output_path=output_dir, filename=filename)
def save_spec(self, filepath, converted_dir, audio_path):
y, sr = librosa.load(filepath, sr=None, mono=False)
# Reshaping the Audio file (mono) to deal with all wav files similarly
if y.ndim == 1:
y = y.reshape(1, -1)
for ch in range(y.shape[0]):
length = int(len(y[ch]) / sr)
remainder = length % self.window_size
audio = AudioFile(self.base_dir, converted_dir)
acc = AccelerometerFile(self.base_dir, converted_dir, ch)
file_type = audio if y.shape[0] == 1 else acc
for i in range(0, length - remainder - self.window_size,
self.slide):
begin_time = i
end_time = i + self.window_size
s_db = get_spectrogram(begin_time, end_time, y[ch], sr)
# Extracting file identifier from the filepath
# Example: (i.e.'cc16_352a_14401s' from path "'/cache/rmishra/cc16_ML/cc16_352a/cc16_352a_14401s_acc.wav'")
# for saving spec and label files with begin and end timestamp
file = os.path.split(filepath)[-1].rsplit("_", maxsplit=1)[0]
# fetch the label txt file against the file identifier and create a label dataframe for calls between
# the start and end timestamp
call = fetch_files_with_numcalls(audio_path,
1).loc[file]['calls']
timesteps = s_db.shape[1]
timesteps_per_second = timesteps / self.window_size
df = create_label_dataframe(os.path.join(audio_path,
call), begin_time,
end_time, self.window_size,
timesteps_per_second)
# one hot encoding the label information for the audio data in a spec frame
label_matrix = create_label_matrix(df, timesteps)
if 1 in label_matrix[0][:8, :]:
print("Saving spectrogram: " + filepath + " " +
str(begin_time) + " to " + str(end_time))
file_type.save_spec_label(s_db, begin_time, end_time, file,
label_matrix)
def __init__(self, fn, userInfo, mode='publish'):
self.fn = fn
self.userInfo = userInfo
self.mode = mode
self.warnIncomplete = 1
tags = ''
self.config = LoadConfig.LoadConfig()
# open ogg file
self.af = AudioFile(fn)
self.metadata = self.af.read_comments()
if self.mode is 'publish':
newFilePath = self.fn
logger.debug5( "ImportOGG() self.fn: " + self.fn )
logger.debug5( "ImportOGG() newFilePath: " + newFilePath )
if newFilePath:
if self.import2db(newFilePath):
self.setImported()
elif self.mode is 'edit':
if self.import2db(self.fn):
logger.debug1( "Updated database from file: %s" % self.fn )
class Window(QWidget):
def __init__(self):
super().__init__()
self.initUI()
self.audiofile = AudioFile()
def initUI(self):
self.wrong_nb = 0
self.correct_nb = 0
grid = QGridLayout()
play = QPushButton('Play', self)
check = QPushButton('Check', self)
self.answerEdit = QLineEdit()
title = QLabel('answer')
correctLabel = QLabel('correct: ')
wrongLabel = QLabel('wrong: ')
self.wrong_nbLabel = QLabel(str(self.wrong_nb))
self.correct_nbLabel = QLabel(str(self.correct_nb))
self.setLayout(grid)
play.clicked.connect(self.play)
check.clicked.connect(self.check)
self.answerEdit.returnPressed.connect(self.check)
grid.addWidget(correctLabel, 0, 0)
grid.addWidget(self.correct_nbLabel, 0, 1)
grid.addWidget(wrongLabel, 0, 2)
grid.addWidget(self.wrong_nbLabel, 0, 3)
grid.addWidget(title, 1, 1)
grid.addWidget(self.answerEdit, 1, 2)
grid.addWidget(play, 2, 1)
grid.addWidget(check, 2, 2)
self.shortcut = QShortcut(QKeySequence(Qt.Key_Space), self)
self.shortcut.activated.connect(self.play)
self.answerEdit.setFocus()
print(self.__dict__)
self.show()
def play(self):
self.audiofile.play()
def check(self):
answer = self.answerEdit.text()
if self.audiofile.check_answer(answer):
self.correct_nb += 1
self.correct_nbLabel.setText(str(self.correct_nb))
self.audiofile = AudioFile()
else:
self.wrong_nb += 1
self.wrong_nbLabel.setText(str(self.wrong_nb))
_stats.add_stats(
answer,
self.audiofile.get_pinyin(),
self.audiofile.get_id(),
)
self.answerEdit.clear()
def extractor_var(origen, framelen=0.25, destino=''):
"""
Metodo extractor de features (caracteristicas o descriptores) de un array de tipo PCM
Se realizara mediante segmentacion por segundos del array de muestras de audio.
Por cada segundo tendremos una fila de un vector 1D de longitud el numero de parametros que extraigamos
Al final se almacena esta matriz de dimensiones NumSegundosxNumParametros en formato npy en una ruta designada
Un archivo por cada audio
O tambien,si no le damos un destino, genera un vector uniendo todos los audios y lo devuelve
:param s: ruta de los arrays originales para parametrizar
:type s: string
:param p: ruta donde se almacenaran los vectores parametrizados
:type p: string
Actualmente en config tenemos establecido este set de features:
Vector de 20 parametros:
lst_energy - hzcrr - centroid - spread - variance - rolloff - mean - crest - mfcc (8)
"""
if not os.path.isdir(origen):
if not os.path.isfile(origen):
print 'Directorio o nombre de archivos de origen no valido o sin extension (wav/mp3)'
sys.exit()
else:
origen = origen
else:
origen = os.path.join(origen, '*.wav')
if not glob.glob(origen):
print 'no hay archivos de formato wav en el directorio'
sys.exit()
vectortotal = []
primero = True
print 'Inicio del parametrizador. Extraccion segundo a segundo y con %i parametros' % NUM_PARAMETROS
for filename in (glob.glob(origen)):
print '\nVectorizando archivo: ', filename
t1 = time.time()
s = AudioFile.open(filename)
sf = s.frames(s.sampleRate * framelen)
num_ventanas = len(sf)
lenv = np.round(s.sampleRate * VENTANA)
vector = np.zeros([num_ventanas, NUM_PARAMETROS], dtype=np.float32)
for indf, frame in enumerate(sf):
print len(frame)
if len(frame) < s.sampleRate:
break
Espectro = frame.spectrum()
acumulado = 0
for param in zip(PARAMETROS, TIPO_PARAMETROS):
if not param[1]:
vector[indf,
acumulado] = getattr(Energy,
param[0])(frame,
windowSize=lenv,
solape=SOLAPE)
else:
vector[indf, acumulado] = Espectro.mean(
) # getattr(Espectro, param[0])()
acumulado = acumulado + 1
if MFCC > 0:
mfcc_features = mfcc(frame,
samplerate=s.sampleRate,
winlen=VENTANA,
numcep=MFCC)
mfcc_means = np.mean(mfcc_features, 0)
for i in range(0, MFCC):
vector[indf, acumulado] = mfcc_means[i]
acumulado = acumulado + 1
if DELTAS:
delta = np.zeros(MFCC)
dobledelta = np.zeros(MFCC)
for i in range(0, MFCC):
diferencias = np.diff(mfcc_features[:, i])
delta[i] = np.sum(diferencias)
dobledelta[i] = np.sum(np.diff(diferencias))
for i in range(0, MFCC):
vector[indf, acumulado] = delta[i]
acumulado = acumulado + 1
for i in range(0, MFCC):
vector[indf, acumulado] = dobledelta[i]
acumulado = acumulado + 1
if CHROMA > 0:
array_chroma = Espectro.chroma()
for i in range(0, CHROMA):
vector[indf, acumulado] = array_chroma[i]
acumulado = acumulado + 1
if FLUX > 0:
spectral_frames = frame.frames(lenv)
spectra = [f.spectrum_dct() for f in spectral_frames]
flujo = SpectralFlux.spectralFlux(spectra, rectify=True)
for i in range(0, FLUX):
vector[indf, acumulado] = flujo[i]
acumulado = acumulado + 1
print 'Tiempo de parametrizacion (minutos): '
print(time.time() - t1) / 60
archivo = os.path.split(filename)[1].split('.')[0]
ruta = os.path.join(destino, archivo)
if destino:
np.save(ruta, vector)
if primero:
vectortotal = np.array(vector)
else:
vectortotal = np.append(vectortotal, np.array(vector), axis=0)
primero = False
return vectortotal
def __init__(self):
super().__init__()
self.initUI()
self.audiofile = AudioFile()
class Browser:
'''
The browser object returns an array of items, typically to be used
by the scroller object from the Burn Station.
It should also provide a series of methods to browse back and forth
through the child/parent items.
'''
#--------------------------------------------------------------------
def __init__(self, level='', itemID=0):
self.level = level
self.itemID = itemID
self.j = Jamendo()
logger.debug("******** level: " + level)
#--------------------------------------------------------------------
def SetType(self, type):
self.level = type
#--------------------------------------------------------------------
def getList(self, parent=0, index=0):
if self.level == "path": return self.getListFromPath(parent, index)
elif self.level == "jamendo":
return self.getListFromJamendo(parent, index)
elif self.level == "labels":
return self.getListFromDB(parent, index)
else:
return self.getListFromDB(parent, index)
#--------------------------------------------------------------------
def getListFromJamendo(self, parent, index=0):
return self.j.search_artist_by_name(search="")
#--------------------------------------------------------------------
def getListFromPath(self, parent, index=0):
logger.debug("Getting contents from: %s" % parent)
try:
if parent == 0: parent = "/"
dirlist = os.listdir(parent)
dirlist.sort()
except Exception, e:
logger.error("at Browser.getListFromPath(): " + str(e))
list = []
for x in dirlist:
name, ext = os.path.splitext(x)
if x != get_ascii_filename(x):
continue
real_path = os.path.join(parent, uniconvert2(x))
if os.path.isfile(real_path):
if ext.lower() == '.ogg' or ext.lower() == '.mp3':
# read vorbis info
self.af = AudioFile(real_path)
time = int(self.af.getLength())
length = " (" + str(time) + ")"
list.append({
'location': real_path,
'id': 'idd',
'img': 'imgg',
'name': x + length,
'time': time,
'seconds': time
})
else:
list.append({
'location': real_path,
'id': 'idd',
'img': 'imgg',
'name': "/" + x
})
return list
class ImportOGG:
def __init__(self, fn, userInfo, mode='publish'):
self.fn = fn
self.userInfo = userInfo
self.mode = mode
self.warnIncomplete = 1
tags = ''
self.config = LoadConfig.LoadConfig()
# open ogg file
self.af = AudioFile(fn)
self.metadata = self.af.read_comments()
if self.mode is 'publish':
newFilePath = self.fn
logger.debug5("ImportOGG() self.fn: " + self.fn)
logger.debug5("ImportOGG() newFilePath: " + newFilePath)
if newFilePath:
if self.import2db(newFilePath):
self.setImported()
elif self.mode is 'edit':
if self.import2db(self.fn):
logger.debug1("Updated database from file: %s" % self.fn)
def isImported(self, path):
metadata = {}
path = removePathComponent(self.fn, "/usr/local/media/")
name, ext = os.path.splitext(path)
path = name + ".ogg"
db = DB.connect()
cursor = db.cursor()
sql = "SELECT tr.id, tr.name, ar.name, al.name, la.name, li.name"
sql += " FROM netjuke_tracks tr, netjuke_artists ar, netjuke_albums al, netjuke_labels la, licenses li"
sql += " WHERE location = '%s'" % MySQLdb.escape_string(path)
sql += " AND tr.ar_id=ar.id AND tr.al_id=al.id AND tr.la_id=la.id AND tr.license=li.id"
logger.debug99("Using SQL query to check if file is imported: %s" %
sql)
cursor.execute(sql)
result = cursor.fetchall()
record = None
for record in result:
metadata['ID'] = record[0]
metadata['title'] = record[1]
metadata['artist'] = unicode(record[2], 'utf8')
metadata['album'] = unicode(record[3], 'utf8')
metadata['label'] = unicode(record[4], 'utf8')
metadata['license'] = record[5]
logger.debug5("Import.isImported(%s) rowcount = %i" %
(path, cursor.rowcount))
return (cursor.rowcount, metadata)
def setImported(self):
'''
Should flag or delete from mp3tmp when file was imported, and delete the source file.
'''
sourceFile = removePathComponent(self.fn, self.userInfo['spoolDir'])
name, ext = os.path.splitext(sourceFile)
sourceFile = name + ".mp3"
db = DB.connect()
sql = "SELECT location FROM mp3tmp WHERE location LIKE '%s'" % MySQLdb.escape_string(
sourceFile)
cursor = db.cursor()
cursor.execute(sql)
result = cursor.fetchall()
for record in result:
sql = "UPDATE mp3tmp SET imported = 1 WHERE location = '%s'" % MySQLdb.escape_string(
record[0])
db.set_character_set('utf8')
cursor = db.cursor()
cursor.execute(sql)
result = cursor.fetchall()
"""
def MoveImported(self, file, target, spoolDir):
return moveExactTree(file, target, spoolDir, 1)
def CopyImported(self, file, target, spoolDir):
logger.debug5( "---------------------------" )
logger.debug5( "in ImportOGG.CopyImported: " )
logger.debug5( "file: " + file )
logger.debug5( "target: " + target )
logger.debug5( "---------------------------" )
target = copyExactTree(file, target, spoolDir, 1)
return target
"""
def import2db(self, newFilePath):
self.af.SetFile(newFilePath)
self.metadata = self.af.read_comments()
logger.debug99("at import2db, metadata is: %s" % self.metadata)
location = location2db(newFilePath, self.userInfo['home'])
print "****************** LOCATION: %s" % location
if self.mode is 'publish':
(isImported, metadata) = self.isImported(location)
elif self.mode is 'edit':
isImported = 1
logger.debug99("------------------------------------")
logger.debug99("in ImportOGG.import2db():")
logger.debug99("user home = " + self.userInfo['home'])
logger.debug99("is imported = " + str(isImported))
logger.debug99("newFilePath = " + newFilePath)
logger.debug99("location = " + location)
logger.debug99("------------------------------------")
if (self.mode is 'publish') and isImported:
logger.debug1("File already imported.. skipping: %s" % location)
return
artistID = getID('artists', self.metadata['artist'])
albumID = getID('albums', self.metadata['album'])
labelID = getID('labels', self.metadata['label'])
licenseID = getID('licenses', self.metadata['license'])
if artistID == 1 or albumID == 1 or labelID == 1 or licenseID == 1:
logger.warn("Incomplete tags for file!")
self.warnIncomplete = 0
filename = os.path.basename(newFilePath)
if self.metadata['title'] == '':
title = MySQLdb.escape_string(filename)
else:
title = MySQLdb.escape_string(
uniconvert2(self.metadata['title']).encode('utf8'))
size = self.af.getSize()
time = int(self.af.getLength())
track_number = self.af.getTag('track_number')
if track_number == '': track_number = self.af.getTag('tracknumber')
if track_number == '': track_number = 0
year = self.af.getTag('year')
if year == '': year = 0
bitrate = self.af.getBitrate()
sample_rate = self.af.getSamplerate()
kind = 'OGG/Vorbis'
location = MySQLdb.escape_string(uniconvert2(location).encode('utf8'))
comments = MySQLdb.escape_string(
uniconvert2(self.metadata['comment']).encode('utf8'))
if self.mode is 'publish':
sql = "INSERT INTO"
elif self.mode is 'edit':
sql = "UPDATE"
trackID = getIDbyLocation(location)
sql += " netjuke_tracks SET ar_id = '" + str(
artistID
) + "', al_id = '" + str(albumID) + "', ge_id = '1', la_id = '" + str(
labelID
) + "', name = '" + title + "', size = '" + str(
size
) + "', time = '" + str(time) + "', track_number = '" + str(
track_number
) + "', year = '" + str(year) + "', date = now(), bit_rate = '" + str(
bitrate
) + "', sample_rate = '" + str(
sample_rate
) + "', kind = '" + kind + "', location = '" + str(
location
) + "', comments = '" + comments + "', mtime = now(), license = '" + str(
licenseID) + "', lg_id = '1', enabled = 2"
if self.mode is 'edit': sql += " WHERE id = %s" % str(trackID)
logger.debug99("Using SQL query: " + sql)
db = DB.connect()
cursor = db.cursor()
cursor.execute(sql)
fileID = db.insert_id()
self.DoFileUploaderRelation(fileID, self.userInfo['ID'])
# FIXME: do some checks, and return False if this fails
print "=" * 77
return True
def DoFileUploaderRelation(self, fileID, userID):
sql = "INSERT IGNORE INTO file_uploader SET file = '%i', uploader = '%i'" % (
fileID, userID)
logger.debug99("Using SQL query: " + sql)
db = DB.connect()
cursor = db.cursor()
cursor.execute(sql)
def __init__(self, file):
AudioFile.__init__(self, file)
def processSpeech(name):
'''
Asks name what they want to do and processes
the speech into text.
@param name string of name from face recognition
@return extracted keyword from speech
'''
if "Cho" in name:
name = "Edward"
elif "Wu" in name:
name = "Johnny"
elif "Deng" in name:
name = "Michelle"
elif "Liu" in name:
name = "Vincent"
# Initial greeting
tts('Hello ' + name + '. \
What can I do for you today?', 'resources/greeting.wav')
audio = AudioFile('resources/greeting.wav')
audio.play()
audio.close()
# Wait for response
print('waiting for response')
recorder = Recorder('request.wav')
recorder.record_to_file()
print('transcribing')
result = stt('request.wav')
transcript = result['results'][0]['alternatives'][0]['transcript']
# Eating Section
if "eat" in transcript:
keyPhrase = 'I want to eat '
if keyPhrase in transcript:
yelp_search.searchFood(transcript[len(keyPhrase)::])
return processSpeech(name)
else:
print('Did not say key phrase')
return processSpeech(name)
# Budget Section
elif "budget" in transcript:
watson_budget("Edward", "Friday")
return processSpeech(name)
elif "time" in transcript:
if (datetime.datetime.time(datetime.datetime.now()) > datetime.time(4, 20, 0, 0) and
datetime.datetime.time(datetime.datetime.now()) < datetime.time(4, 20, 59, 0)):
tts("ay. lmao", 'resources/420.wav')
audio = AudioFile('resources/420.wav')
audio.play()
audio.close()
else:
tts("The current time is " +
datetime.datetime.now().strftime("%I:%M%p on %B %d, %Y"),
'resources/time.wav')
audio = AudioFile('resources/time.wav')
audio.play()
audio.close()
# Default Case for not getting any proper key words
else:
return processSpeech(name)
def preprocess(filename, timidity, latency, truncate, pad=1, get_raw=False):
"""
Preprocess an audio file ands its MIDI counterpart. Computes transforms and labels.
:param filename: audio filename
:param timidity: set to True if the files was rendered with timidity
:param latency: in seconds
:param truncate: in seconds (0 for no truncation)
:param pad: in seconds, will be added at the start and end before spectral transforms
:param get_raw: set to True to return raw computed spectrograms (e.g. for visualization)
:return:
"""
filename_midi = filename.rsplit('.')[0] + '.mid'
dname = filename.replace('/', '_').replace('\\', '_')
# Load files
ipad = int(pad * 44100)
audio_pad = (ipad, ipad
) # add one blank second at the beginning and at the end
if truncate > 0:
audio = AudioFile(filename,
truncate=int(truncate * 44100),
pad=audio_pad)
else:
audio = AudioFile(filename, pad=audio_pad)
mid = MidiFile(filename_midi)
step = 0.02 # seconds
latency = int(round(latency / step, 0))
# Compute spectrograms
spectrograms = ComputeSpectrograms(audio, step=step)
# Compute filtered spectrograms
melgrams = ComputeMelLayers(spectrograms, step, audio.Fs, latency)
# Build the input tensor
cnn_window = 15
tensor_mel = BuildTensor(melgrams[:, 2], cnn_window)
# Compute CQT
FreqAxisLog, time, cqgram = ComputeCqt(audio,
200.,
4000.,
step,
latency,
r=3)
tensor_cqt = BuildTensor([
cqgram,
], cnn_window)
# Global data length
max_len = min(tensor_mel.shape[0], tensor_cqt.shape[0])
# Compute output labels
notes = mid.getNotes(timidity)
notes_onset = np.array(notes)[:, 0] # get only the note timing
notes_value = np.array(notes, dtype=np.int)[:, 1] # get only the key value
onset_labels = np.zeros(max_len)
onset_caracs = np.zeros((max_len, 5))
onset_caracs[:, 2] = np.arange(max_len)
note_low = 21 # lowest midi note on a keyboard
note_high = 108 # highest midi note on a keyboard
notes_labels = np.zeros((max_len, note_high - note_low + 1))
notes_caracs = np.zeros((max_len, note_high - note_low + 1))
for i in range(len(notes_onset)):
t_win = int(np.floor(
(notes_onset[i] + audio_pad[0] / audio.Fs) / step))
if t_win >= len(onset_labels):
break
if t_win >= 0:
onset_labels[t_win] = 1
onset_caracs[t_win][0] += 1 # nb_notes
onset_caracs[t_win][1] = max(onset_caracs[t_win][1],
notes[i][2]) # max volume
if t_win + 1 < len(onset_labels):
onset_caracs[t_win + 1:, 2] -= onset_caracs[t_win + 1][
2] # nb of blank windows since the last onset
n = notes_value[i] - note_low
notes_labels[t_win][n] = 1
notes_caracs[t_win][n] = notes[i][2] # volume
counter = 0
for i in range(len(onset_labels) - 1, -1, -1):
onset_caracs[i][3] = counter
if onset_labels[i] == 1:
counter = 0
else:
counter += 1
onset_caracs[:, 4] = np.minimum(onset_caracs[:, 2], onset_caracs[:, 3])
# Extract useful CQT
select = [i for i in range(max_len) if onset_labels[i] > 0]
tensor_cqt_select = np.take(tensor_cqt, select, axis=0)
notes_labels_select = np.take(notes_labels, select, axis=0)
notes_caracs_select = np.take(notes_caracs, select, axis=0)
if not get_raw:
return (tensor_mel[:max_len, ...], tensor_cqt_select, onset_labels,
onset_caracs, notes_labels_select, notes_caracs_select, dname)
else:
return (melgrams, tensor_mel, onset_labels, cqgram, tensor_cqt, time,
FreqAxisLog, max_len, step)
def extractor_audio(buffer, fs):
'''
:param buffer: string list del audio a vectorizar
:return: una fila del vector
'''
s = AudioFile.open_frombuffer(buffer)
sf = s.frames(fs)
num_ventanas = len(sf)
lenv = np.round(s.sampleRate*VENTANA)
vector = np.zeros([num_ventanas, NUM_PARAMETROS], dtype=np.float32)
for indf, frame in enumerate(sf):
Espectro = frame.spectrum()
acumulado=0
for param in zip(PARAMETROS, TIPO_PARAMETROS):
if not param[1]:
vector[indf,acumulado] = getattr(Energy, param[0])(frame, windowSize= lenv, solape=SOLAPE)
else:
vector[indf,acumulado] = Espectro.mean()# getattr(Espectro, param[0])()
acumulado = acumulado + 1
if MFCC > 0:
mfcc_features = mfcc(s, samplerate=s.sampleRate,winlen=VENTANA, numcep=MFCC)
mfcc_means = np.mean(mfcc_features, 0)
for i in range(0, MFCC):
vector[indf, acumulado] = mfcc_means[i]
acumulado = acumulado + 1
if DELTAS:
delta=np.zeros(MFCC)
dobledelta=np.zeros(MFCC)
for i in range(0, MFCC):
diferencias=np.diff(mfcc_features[:, i])
delta[i] = np.sum(diferencias)
dobledelta[i] = np.sum(np.diff(diferencias))
for i in range(0, MFCC):
vector[indf,acumulado] = delta[i]
acumulado = acumulado + 1
for i in range(0, MFCC):
vector[indf,acumulado] = dobledelta[i]
acumulado = acumulado + 1
if CHROMA > 0:
array_chroma=Espectro.chroma()
for i in range(0,CHROMA):
vector[indf,acumulado]=array_chroma[i]
acumulado = acumulado + 1
if FLUX > 0:
spectral_frames=s.frames(lenv)
spectra = [f.spectrum_dct() for f in spectral_frames]
flujo=SpectralFlux.spectralFlux(spectra, rectify=True)
for i in range(0,FLUX):
vector[indf, acumulado]=flujo[i]
acumulado = acumulado + 1
return vector
class ImportOGG:
def __init__(self, fn, userInfo, mode='publish'):
self.fn = fn
self.userInfo = userInfo
self.mode = mode
self.warnIncomplete = 1
tags = ''
self.config = LoadConfig.LoadConfig()
# open ogg file
self.af = AudioFile(fn)
self.metadata = self.af.read_comments()
if self.mode is 'publish':
newFilePath = self.fn
logger.debug5( "ImportOGG() self.fn: " + self.fn )
logger.debug5( "ImportOGG() newFilePath: " + newFilePath )
if newFilePath:
if self.import2db(newFilePath):
self.setImported()
elif self.mode is 'edit':
if self.import2db(self.fn):
logger.debug1( "Updated database from file: %s" % self.fn )
def isImported(self, path):
metadata = {}
path = removePathComponent(self.fn, "/usr/local/media/")
name, ext = os.path.splitext(path)
path = name + ".ogg"
db = DB.connect()
cursor = db.cursor()
sql = "SELECT tr.id, tr.name, ar.name, al.name, la.name, li.name"
sql += " FROM netjuke_tracks tr, netjuke_artists ar, netjuke_albums al, netjuke_labels la, licenses li"
sql += " WHERE location = '%s'" % MySQLdb.escape_string(path)
sql += " AND tr.ar_id=ar.id AND tr.al_id=al.id AND tr.la_id=la.id AND tr.license=li.id"
logger.debug99( "Using SQL query to check if file is imported: %s" % sql )
cursor.execute(sql)
result = cursor.fetchall()
record = None
for record in result:
metadata['ID'] = record[0]
metadata['title'] = record[1]
metadata['artist'] = unicode(record[2], 'utf8')
metadata['album'] = unicode(record[3], 'utf8')
metadata['label'] = unicode(record[4], 'utf8')
metadata['license'] = record[5]
logger.debug5( "Import.isImported(%s) rowcount = %i" % (path, cursor.rowcount) )
return (cursor.rowcount, metadata)
def setImported(self):
'''
Should flag or delete from mp3tmp when file was imported, and delete the source file.
'''
sourceFile = removePathComponent(self.fn, self.userInfo['spoolDir'])
name, ext = os.path.splitext(sourceFile)
sourceFile = name + ".mp3"
db = DB.connect()
sql = "SELECT location FROM mp3tmp WHERE location LIKE '%s'" % MySQLdb.escape_string(sourceFile)
cursor = db.cursor()
cursor.execute(sql)
result = cursor.fetchall()
for record in result:
sql = "UPDATE mp3tmp SET imported = 1 WHERE location = '%s'" % MySQLdb.escape_string(record[0])
db.set_character_set('utf8')
cursor = db.cursor()
cursor.execute(sql)
result = cursor.fetchall()
"""
def MoveImported(self, file, target, spoolDir):
return moveExactTree(file, target, spoolDir, 1)
def CopyImported(self, file, target, spoolDir):
logger.debug5( "---------------------------" )
logger.debug5( "in ImportOGG.CopyImported: " )
logger.debug5( "file: " + file )
logger.debug5( "target: " + target )
logger.debug5( "---------------------------" )
target = copyExactTree(file, target, spoolDir, 1)
return target
"""
def import2db(self, newFilePath):
self.af.SetFile(newFilePath)
self.metadata = self.af.read_comments()
logger.debug99("at import2db, metadata is: %s" % self.metadata)
location = location2db(newFilePath, self.userInfo['home'])
print "****************** LOCATION: %s" % location
if self.mode is 'publish': (isImported, metadata) = self.isImported(location)
elif self.mode is 'edit': isImported = 1
logger.debug99( "------------------------------------" )
logger.debug99( "in ImportOGG.import2db():" )
logger.debug99( "user home = " + self.userInfo['home'] )
logger.debug99( "is imported = " + str(isImported) )
logger.debug99( "newFilePath = " + newFilePath )
logger.debug99( "location = " + location )
logger.debug99( "------------------------------------" )
if (self.mode is 'publish') and isImported:
logger.debug1( "File already imported.. skipping: %s" % location )
return
artistID = getID('artists', self.metadata['artist'])
albumID = getID('albums', self.metadata['album'])
labelID = getID('labels', self.metadata['label'])
licenseID = getID('licenses', self.metadata['license'])
if artistID == 1 or albumID == 1 or labelID == 1 or licenseID == 1:
logger.warn("Incomplete tags for file!")
self.warnIncomplete = 0
filename = os.path.basename(newFilePath)
if self.metadata['title'] == '': title = MySQLdb.escape_string(filename)
else: title = MySQLdb.escape_string(uniconvert2(self.metadata['title']).encode('utf8'))
size = self.af.getSize()
time = int( self.af.getLength() )
track_number = self.af.getTag('track_number')
if track_number == '': track_number = self.af.getTag('tracknumber')
if track_number == '': track_number = 0
year = self.af.getTag('year')
if year == '': year = 0
bitrate = self.af.getBitrate()
sample_rate = self.af.getSamplerate()
kind = 'OGG/Vorbis'
location = MySQLdb.escape_string(uniconvert2(location).encode('utf8'))
comments = MySQLdb.escape_string(uniconvert2(self.metadata['comment']).encode('utf8'))
if self.mode is 'publish':
sql = "INSERT INTO"
elif self.mode is 'edit':
sql = "UPDATE"
trackID = getIDbyLocation(location)
sql += " netjuke_tracks SET ar_id = '"+ str(artistID) +"', al_id = '"+ str(albumID) +"', ge_id = '1', la_id = '"+ str(labelID) +"', name = '"+ title +"', size = '"+ str(size) +"', time = '"+ str(time) +"', track_number = '"+ str(track_number) +"', year = '"+ str(year) +"', date = now(), bit_rate = '"+ str(bitrate) +"', sample_rate = '"+ str(sample_rate) +"', kind = '"+ kind +"', location = '"+ str(location) +"', comments = '"+ comments +"', mtime = now(), license = '"+ str(licenseID) +"', lg_id = '1', enabled = 2"
if self.mode is 'edit': sql += " WHERE id = %s" % str(trackID)
logger.debug99( "Using SQL query: " + sql )
db = DB.connect()
cursor = db.cursor()
cursor.execute(sql)
fileID = db.insert_id()
self.DoFileUploaderRelation(fileID, self.userInfo['ID'])
# FIXME: do some checks, and return False if this fails
print "="*77
return True
def DoFileUploaderRelation(self, fileID, userID):
sql = "INSERT IGNORE INTO file_uploader SET file = '%i', uploader = '%i'" % (fileID, userID)
logger.debug99( "Using SQL query: " + sql )
db = DB.connect()
cursor = db.cursor()
cursor.execute(sql)
def extractor_audio(buffer, fs):
'''
:param buffer: string list del audio a vectorizar
:return: una fila del vector
'''
s = AudioFile.open_frombuffer(buffer)
sf = s.frames(fs)
num_ventanas = len(sf)
lenv = np.round(s.sampleRate * VENTANA)
vector = np.zeros([num_ventanas, NUM_PARAMETROS], dtype=np.float32)
for indf, frame in enumerate(sf):
Espectro = frame.spectrum()
acumulado = 0
for param in zip(PARAMETROS, TIPO_PARAMETROS):
if not param[1]:
vector[indf, acumulado] = getattr(Energy,
param[0])(frame,
windowSize=lenv,
solape=SOLAPE)
else:
vector[indf, acumulado] = Espectro.mean(
) # getattr(Espectro, param[0])()
acumulado = acumulado + 1
if MFCC > 0:
mfcc_features = mfcc(s,
samplerate=s.sampleRate,
winlen=VENTANA,
numcep=MFCC)
mfcc_means = np.mean(mfcc_features, 0)
for i in range(0, MFCC):
vector[indf, acumulado] = mfcc_means[i]
acumulado = acumulado + 1
if DELTAS:
delta = np.zeros(MFCC)
dobledelta = np.zeros(MFCC)
for i in range(0, MFCC):
diferencias = np.diff(mfcc_features[:, i])
delta[i] = np.sum(diferencias)
dobledelta[i] = np.sum(np.diff(diferencias))
for i in range(0, MFCC):
vector[indf, acumulado] = delta[i]
acumulado = acumulado + 1
for i in range(0, MFCC):
vector[indf, acumulado] = dobledelta[i]
acumulado = acumulado + 1
if CHROMA > 0:
array_chroma = Espectro.chroma()
for i in range(0, CHROMA):
vector[indf, acumulado] = array_chroma[i]
acumulado = acumulado + 1
if FLUX > 0:
spectral_frames = s.frames(lenv)
spectra = [f.spectrum_dct() for f in spectral_frames]
flujo = SpectralFlux.spectralFlux(spectra, rectify=True)
for i in range(0, FLUX):
vector[indf, acumulado] = flujo[i]
acumulado = acumulado + 1
return vector
def transcribe_to_midi(filename, onset_model, note_model, output):
"""
Transcribes and audio file to midi and renders it to wav if timidity is found in the path
:param filename: path of audio file to transcribe
:param onset_model: onset detection model to use (hdf5 filename)
:param note_model: key identification model to use (hdf5 filename)
:param output: output filename (without extension)
:return: None
"""
step = 0.02
_vprint(f'load audio {filename}...')
audio = AudioFile(filename, pad=(44100, 44100))
_vprint('computing spectrograms...')
spectrograms = preprocess.ComputeSpectrograms(audio, step=step)
_vprint('computing (mel) filtered spectrograms...')
melgrams = preprocess.ComputeMelLayers(spectrograms,
step,
audio.Fs,
latency=0)
cnn_window = 15
tensor_mel = preprocess.BuildTensor(melgrams[:, 2], cnn_window)
_vprint('onset detection...')
model = keras.models.load_model(onset_model)
preds_onset = 1. * (model.predict(tensor_mel) >= 0.2)
nb_notes = np.sum(preds_onset)
_vprint(f'{nb_notes} onsets detected')
_vprint('computing CQT...')
# TODO: compute only useful ones
FreqAxisLog, time, cqgram = preprocess.ComputeCqt(audio,
200.,
4000.,
step,
latency=0,
r=3)
tensor_cqt = preprocess.BuildTensor([
cqgram,
], cnn_window)
max_len = min(tensor_mel.shape[0], tensor_cqt.shape[0])
select = [i for i in range(max_len) if preds_onset[i] > 0]
tensor_cqt_select = np.take(tensor_cqt, select, axis=0)
_vprint('key identification...')
model = keras.models.load_model(note_model)
preds_notes = 1. * (model.predict(tensor_cqt_select) >= 0.5)
_vprint(f'{np.sum(preds_notes)} keys identified')
_vprint('midi writing...')
mid = mido.MidiFile(ticks_per_beat=500)
track = mido.MidiTrack()
mid.tracks.append(track)
track.append(mido.Message('program_change', program=1, time=0))
i = 0
t = 0 # time in seconds
t_last = 0
notes_on = np.zeros(88)
for w in preds_onset:
if w[0]:
if np.sum(preds_notes[i]) > 0:
delta_t = int((t - t_last) / 0.001) # delta_time in midi ticks
for n in notes_on.nonzero()[0]:
midi_note = int(n + 21)
track.append(
mido.Message('note_off',
note=midi_note,
velocity=0,
time=delta_t))
notes_on[n] = 0
delta_t = 0
for n in preds_notes[i].nonzero()[0]:
midi_note = int(n + 21)
track.append(
mido.Message('note_on',
note=midi_note,
velocity=64,
time=delta_t))
notes_on[n] = 1
delta_t = 0
t_last = t
i += 1
t += step
mid.save(f'{output}.mid')
timidity = find_exec(['timidity', 'timidity.exe'],
additional_dirs=[_TIMIDITY_DIR])
if timidity is not None:
_vprint('timidity found, rendering to audio: ' + timidity)
subprocess.run([
timidity, f'{output}.mid', '-Ow', '-o', f'{output}.wav',
'--output-mono'
])
class Browser:
'''
The browser object returns an array of items, typically to be used
by the scroller object from the Burn Station.
It should also provide a series of methods to browse back and forth
through the child/parent items.
'''
#--------------------------------------------------------------------
def __init__(self, level='', itemID=0):
self.level = level
self.itemID = itemID
self.j = Jamendo()
logger.debug( "******** level: " + level)
#--------------------------------------------------------------------
def SetType(self, type):
self.level = type
#--------------------------------------------------------------------
def getList(self, parent=0, index=0):
if self.level == "path": return self.getListFromPath(parent, index)
elif self.level == "jamendo": return self.getListFromJamendo(parent, index)
elif self.level == "labels": return self.getListFromDB(parent, index)
else: return self.getListFromDB(parent, index)
#--------------------------------------------------------------------
def getListFromJamendo(self, parent, index=0):
return self.j.search_artist_by_name(search="")
#--------------------------------------------------------------------
def getListFromPath(self, parent, index=0):
logger.debug( "Getting contents from: %s" % parent )
try:
if parent == 0: parent = "/"
dirlist = os.listdir(parent)
dirlist.sort()
except Exception, e: logger.error( "at Browser.getListFromPath(): " + str(e) )
list = []
for x in dirlist:
name, ext = os.path.splitext(x)
if x != get_ascii_filename(x):
continue
real_path = os.path.join(parent, uniconvert2(x))
if os.path.isfile(real_path):
if ext.lower() == '.ogg' or ext.lower() == '.mp3':
# read vorbis info
self.af = AudioFile(real_path)
time = int(self.af.getLength())
length = " (" + str(time) + ")"
list.append( {'location':real_path, 'id':'idd', 'img':'imgg', 'name': x + length, 'time': time, 'seconds':time } )
else:
list.append( {'location':real_path, 'id':'idd', 'img':'imgg', 'name': "/" + x } )
return list
#--------------------------------------------------------------------
def getListFromDB(self, itemID=0, index=0):
#print "AT GET LIST FROM DB:::::::::::::::" + self.level + "::" + str(itemID) + '::' + str(index) + "::"
'''Get the list of items to be displayed'''
if self.level == 'labels':
logger.debug5( "GETTING LABELS" )
items = Database.GetLabels()
self.level = 'artists'
elif self.level == 'artists' and itemID>0:
logger.debug5( "GETTING ARTISTS" )
items = Database.GetArtists(itemID)
self.labelID = itemID
self.labelIndex = index
self.level = 'albums'
elif self.level == 'albums':
logger.debug5( "GETTING ALBUMS" )
items = Database.GetAlbums(self.labelID, itemID)
self.level = 'tracks'
self.artistID = itemID
self.artistIndex = index
elif self.level == 'tracks':
logger.debug5( "GETTING TRACKS" )
items = Database.GetTracks(self.labelID, self.artistID, itemID)
self.albumID = itemID
self.albumIndex = index
self.level = 'tracks_list'
elif self.level == 'tracks_list':
raise Exception, "Trying to open track means PLAY IT !"
elif self.level == 'playlist' or self.level == 'playlist_tracks':
logger.debug5( "GETTING DB PLAYLIST TRACKS" )
if itemID == 0 and self.itemID > 0 : itemID = self.itemID
items = Database.GetPlaylist(itemID)
self.level = 'playlist_tracks'
elif self.level == 'fs_playlist':
logger.debug5( "GETTING FILESYSTEM PLAYLIST TRACKS" )
if itemID == 0 and self.itemID > 0 : itemID = self.itemID
items = Database.GetFilesystemPlaylist(itemID)
self.level = 'fs_playlist'
try: return items
except: return [ { 'id':"a", 'name':"a", 'img':"a", 'info':"a" } ]
#--------------------------------------------------------------------
def descend(self, parent, selected):
self.parentIndex = selected
if self.level == 'jamendo':
return self.j.descend(parent, selected)
elif self.level == 'path':
# browsing filesystem
logger.debug( "Descend to: " + parent )
if os.path.isdir(parent):
self.currentDir = parent
return self.getList(parent, selected)
else:
self.Play()
elif self.level != 'path' and self.level != '' and self.level != 'playlist_tracks':
# browsing database
return self.getList(parent, selected)
else:
# no childs, impossible to descend
raise Exception, "Descend() impossible. No child items found!"
#--------------------------------------------------------------------
def Back(self):
print "BACK FROM TRACKS, self.level = " + self.level
if self.level == 'path':
dir = self.currentDir+'/..'
logger.debug(" ********** currentDir: "+dir)
items = self.getList(dir, self.parentIndex)
self.currentDir = dir
return { 'list':items, 'index':self.parentIndex }
elif self.level == 'tracks_list':
items = Database.GetAlbums(self.labelID, self.artistID)
self.level = 'tracks'
return { 'list':items, 'index':self.albumIndex }
elif self.level == 'tracks':
items = Database.GetArtists(self.labelID)
self.level = 'albums'
return { 'list':items, 'index':self.artistIndex }
elif self.level == 'albums':
items = Database.GetLabels()
self.level = 'artists'
return { 'list':items, 'index':self.labelIndex }
else: return
def augment(input_dir: Path, noise_dir: Path, output_dir: Path):
seed(SEED)
if not input_dir.exists:
raise Exception('Input directory does not exist.')
if not noise_dir.exists:
raise Exception('Noise directory does not exist.')
if not output_dir.exists:
print("Making output directory {}".format(output_dir))
output_dir.mkdir(parents=True)
filelist = [x for x in input_dir.glob('*.wav')]
print("{} input files found".format(len(filelist)))
noiselist = set([x for x in noise_dir.glob('*.wav')])
print("{} noise files found".format(len(noiselist)))
print("Loading noise files into memory...")
noise_files = [AudioFile(path=x) for x in noiselist]
print("Done loading noise files.")
irs = [x for x in Path('./IMreverbs-realistic').glob('*.wav')]
while len(filelist) > 0:
print("{} files remaining...".format(len(filelist)))
f1 = filelist.pop()
noise = choice(noise_files)
# load audio files and apply a random amount of processing to noisy file:
# gain reduction in steps of -6 db
# varispeed between [0.9, 1.1]
# start position of audio in noise file (trimming start of file)
gain = [6, 3, 1.5, 0, -1.5, -3, -6]
f1 = AudioFile(path=f1)
noise = noise.copy() \
.trim_start(relative_start=uniform(0.0, 0.5)) \
.trim_to_n_samples(n=f1.length)
# noise = noise.varispeed(uniform(0.9, 1.1))
noise = noise.gain(choice(gain)).clip()
# add dynamic lpf to simulate speaker turning away
filter_start = random()
filter_end = random()
if filter_start < filter_end:
f1.dynamic_lpf(cutoff=uniform(1000, 8000),
order=randint(0, 3),
relative_start=filter_start,
relative_end=filter_end,
exponential=random())
# add noise to audio
f1.mix(noise, maintain_length=True)
# choose random impulse response and add reverb to noisy audio
ir = AudioFile(path=choice(irs))
f1.conv_reverb(ir, wet_db=uniform(-70, -30), predelay=uniform(0, 50))
# filtering
f1.lpf(uniform(5000, 8000))
f1.hpf(uniform(0, 250))
# clipping
clipping = [0.0, 1.0, 2.0, 3.0]
f1.clip(choice(clipping))
# save
f1.save(output_path=output_dir)
def extractor_var(origen, framelen = 0.25, destino=''):
"""
Metodo extractor de features (caracteristicas o descriptores) de un array de tipo PCM
Se realizara mediante segmentacion por segundos del array de muestras de audio.
Por cada segundo tendremos una fila de un vector 1D de longitud el numero de parametros que extraigamos
Al final se almacena esta matriz de dimensiones NumSegundosxNumParametros en formato npy en una ruta designada
Un archivo por cada audio
O tambien,si no le damos un destino, genera un vector uniendo todos los audios y lo devuelve
:param s: ruta de los arrays originales para parametrizar
:type s: string
:param p: ruta donde se almacenaran los vectores parametrizados
:type p: string
Actualmente en config tenemos establecido este set de features:
Vector de 20 parametros:
lst_energy - hzcrr - centroid - spread - variance - rolloff - mean - crest - mfcc (8)
"""
if not os.path.isdir(origen):
if not os.path.isfile(origen):
print 'Directorio o nombre de archivos de origen no valido o sin extension (wav/mp3)'
sys.exit()
else:
origen = origen
else:
origen = os.path.join(origen, '*.wav')
if not glob.glob(origen):
print 'no hay archivos de formato wav en el directorio'
sys.exit()
vectortotal = []
primero = True
print 'Inicio del parametrizador. Extraccion segundo a segundo y con %i parametros' % NUM_PARAMETROS
for filename in (glob.glob(origen)):
print '\nVectorizando archivo: ', filename
t1 = time.time()
s = AudioFile.open(filename)
sf = s.frames(s.sampleRate*framelen)
num_ventanas = len(sf)
lenv = np.round(s.sampleRate*VENTANA)
vector = np.zeros([num_ventanas, NUM_PARAMETROS], dtype=np.float32)
for indf, frame in enumerate(sf):
print len(frame)
if len(frame)<s.sampleRate:
break
Espectro = frame.spectrum()
acumulado=0
for param in zip(PARAMETROS, TIPO_PARAMETROS):
if not param[1]:
vector[indf, acumulado] = getattr(Energy, param[0])(frame, windowSize= lenv, solape=SOLAPE)
else:
vector[indf, acumulado] = Espectro.mean()# getattr(Espectro, param[0])()
acumulado = acumulado + 1
if MFCC > 0:
mfcc_features = mfcc(frame, samplerate=s.sampleRate,winlen=VENTANA, numcep=MFCC)
mfcc_means = np.mean(mfcc_features, 0)
for i in range(0, MFCC):
vector[indf, acumulado] = mfcc_means[i]
acumulado = acumulado + 1
if DELTAS:
delta=np.zeros(MFCC)
dobledelta=np.zeros(MFCC)
for i in range(0, MFCC):
diferencias=np.diff(mfcc_features[:, i])
delta[i] = np.sum(diferencias)
dobledelta[i] = np.sum(np.diff(diferencias))
for i in range(0, MFCC):
vector[indf, acumulado] = delta[i]
acumulado = acumulado + 1
for i in range(0, MFCC):
vector[indf, acumulado] = dobledelta[i]
acumulado = acumulado + 1
if CHROMA > 0:
array_chroma=Espectro.chroma()
for i in range(0,CHROMA):
vector[indf, acumulado]=array_chroma[i]
acumulado = acumulado + 1
if FLUX > 0:
spectral_frames=frame.frames(lenv)
spectra = [f.spectrum_dct() for f in spectral_frames]
flujo=SpectralFlux.spectralFlux(spectra, rectify=True)
for i in range(0,FLUX):
vector[indf, acumulado]=flujo[i]
acumulado = acumulado + 1
print 'Tiempo de parametrizacion (minutos): '
print (time.time()-t1)/60
archivo = os.path.split(filename)[1].split('.')[0]
ruta = os.path.join(destino,archivo)
if destino:
np.save(ruta, vector)
if primero:
vectortotal = np.array(vector)
else:
vectortotal = np.append(vectortotal,np.array(vector), axis=0)
primero = False
return vectortotal