def get_recording(self, file_name):
'''
This reads the song's wave data into recording, which gets returned
INPUT: file_name = string, name of wav file
OUTPUT: recording = 2d numpy float array of song data in time domain;
sample_rate = float, sampling rate of song
'''
song_name = Globals.getSongDir() + "\\" + file_name
sample_rate, recording = wavfile.read(song_name)
return sample_rate, recording
def get_recording(self, file_name):
'''
This reads the song's wave data into recording, which gets returned
INPUT: file_name = string, name of wav file
OUTPUT: recording = 2d numpy float array of song data in time domain;
sample_rate = float, sampling rate of song
'''
song_name = Globals.getSongDir() + "\\" + file_name
sample_rate, recording = wavfile.read(song_name)
return sample_rate, recording
def write_pre_labels(self):
'''
This loads all of the song names in the song directory
and creates a csv called pre_labels where each song will be
manually labeled. the resulting file should be saved as labels.csv
INPUT: NONE
OUTPUT: NONE
'''
song_dir = Globals.getSongDir()
file_list = [f for f in listdir(song_dir) if isfile(join(song_dir, f))]
read_count = 0
song_names = []
for file_name in file_list:
if ".wav" not in file_name:
break
read_count += 1
song_names.append(file_name)
self.write_pre_labels_to_csv(song_names, Globals.getLabelsDir())
def convert_all(self):
'''
This function can be thought of as the main, where all
song wav data is converted. nothing is returned
INPUT: NONE
OUTPUT: NONE
'''
self.delete_files()
song_dir = Globals.getSongDir()
file_list = [f for f in listdir(song_dir) if isfile(join(song_dir, f))]
read_count = 0
incr = 1
'''Begin conversion of all songs'''
for file_name in file_list:
if read_count > self.files_read_count:
break
if ".wav" not in file_name:
break
song_position = 1
cycles = 0
'''get the .wave recording using numpy functionality'''
sample_rate, recording = self.get_recording(file_name)
'''
This while loop breaks only when the cycles condition
below is reached. the code allows for the song to be analyzed
in separate segments, but for now we just work with one segment
as noted by the cycles > 0 termination condition.
the best way to solve this problem is to analyze the song as
multiple segments, where each segment is individually listened
to by a human and classified manually. This adds
a level of diversification to modeling songs. I did not get
around to multiple song feature rows per song so here we
just output one feature row per song for simplicity
'''
while 1 == 1:
self.song_begin = float(song_position * 30)
self.song_end = float(self.song_begin + 90)
try:
'''split the song into .05 second segments'''
segments =\
self.get_segments_temporally(file_name, sample_rate, recording)
'''get all of the song's feature row data'''
song_data, column_labels = \
self.convert_one(file_name, self.segment_length, sample_rate, segments, recording)
print file_name
except Exception, e:
print e
break
'''
write one feature row for this part of the
song using write_data_to_csv
'''
'''
ZULU6FOX6TROT is nothing but a delimiter
to ease splitting later on
'''
self.write_data_to_csv(song_data,\
file_name.split(".wav")[0]+"ZULU6FOX6TROT" +str(int(self.song_begin))\
+ str(int(self.song_end)), self.labels[file_name], column_labels)
song_position += incr
cycles += 1
if cycles > 0:
break
read_count += 1
def convert_all(self):
'''
This function can be thought of as the main, where all
song wav data is converted. nothing is returned
INPUT: NONE
OUTPUT: NONE
'''
self.delete_files()
song_dir = Globals.getSongDir()
file_list = [f for f in listdir(song_dir) if isfile(join(song_dir, f))]
read_count = 0
incr = 1
'''Begin conversion of all songs'''
for file_name in file_list:
if read_count > self.files_read_count:
break
if ".wav" not in file_name:
break
song_position = 1
cycles = 0
'''get the .wave recording using numpy functionality'''
sample_rate, recording = self.get_recording(file_name)
'''
This while loop breaks only when the cycles condition
below is reached. the code allows for the song to be analyzed
in separate segments, but for now we just work with one segment
as noted by the cycles > 0 termination condition.
the best way to solve this problem is to analyze the song as
multiple segments, where each segment is individually listened
to by a human and classified manually. This adds
a level of diversification to modeling songs. I did not get
around to multiple song feature rows per song so here we
just output one feature row per song for simplicity
'''
while 1==1:
self.song_begin = float(song_position * 30)
self.song_end = float(self.song_begin + 90)
try:
'''split the song into .05 second segments'''
segments =\
self.get_segments_temporally(file_name, sample_rate, recording)
'''get all of the song's feature row data'''
song_data, column_labels = \
self.convert_one(file_name, self.segment_length, sample_rate, segments, recording)
print file_name
except Exception, e:
print e
break
'''
write one feature row for this part of the
song using write_data_to_csv
'''
'''
ZULU6FOX6TROT is nothing but a delimiter
to ease splitting later on
'''
self.write_data_to_csv(song_data,\
file_name.split(".wav")[0]+"ZULU6FOX6TROT" +str(int(self.song_begin))\
+ str(int(self.song_end)), self.labels[file_name], column_labels)
song_position += incr
cycles += 1
if cycles > 0:
break
read_count += 1
