def get_metadata(filename):
"""
Deprecated: Given the name of an audio file, this function returns the metadata (composer, genre and lyrics(if available))
:param filename:
:return:
"""
import pandas as pd
import os
base_path = platform_details.get_platform_path(
r"Users\theko\Documents\Dataset")
metadata_path = os.path.join(base_path, 'metadata.csv')
filename_new = filename.strip(base_path).replace("\\", "/")
df = pd.read_csv(metadata_path, sep='\n')
df[['Path', 'Artist', 'Raga', 'RagaId', 'fileId'
]] = df['Path,Artist,Raga,RagaId,fileId'].str.split(',', expand=True)
df = df.drop(columns=['Path,Artist,Raga,RagaId,fileId'])
df = df.set_index('Path')
df = pd.DataFrame.drop_duplicates(df)
# print(df)
try:
artist = df.loc[filename_new, 'Artist']
genre = df.loc[filename_new, 'Raga']
lyrics = ''
return artist, genre, lyrics
except:
print(filename, filename_new, base_path)
return 'unknown', 'Celtic', ''
def _creat_artist_ID_dictionary():
artist_ids = {}
artists = {}
artist_ids_file = platform_details.get_platform_path(
r"Users\theko\Documents\Dataset\v4_artist_ids.txt")
with open(artist_ids_file, 'r', encoding="utf-8") as f:
for line in f:
artist, artist_id = line.strip().split(';')
artist_ids[artist] = int(artist_id)
artists[int(artist_id)] = artist
return artist_ids, artists
def _create_Raga_ID_ditionary():
genre_ids = {}
genres = {}
genre_ids_file = platform_details.get_platform_path(
r"Users\theko\Documents\Dataset\v4_genre_ids.txt")
with open(genre_ids_file, 'r', encoding="utf-8") as f:
for line in f:
genre, genre_id = line.strip().split(';')
genre_ids[genre] = int(genre_id)
genres[int(genre_id)] = genre
return genre_ids, genres
def add_genre_to_dataset(genreID, genreFilesPath, copy=True):
genre_id, genre = _create_Raga_ID_ditionary()
files_list = librosa.util.find_files(
platform_details.get_platform_path(genreFilesPath))
artists = []
fileIDs = []
for i in range(len(files_list)):
meta = audio_metadata.load(files_list[i])
if meta['tags'].__contains__('artist'):
artist = meta['tags'].artist
else:
artist = ''
artists.append(artist)
fileIDs.append(
str(genreID).zfill(3) + str(0).zfill(3) + str(i + 1).zfill(3))
if genre.keys().__contains__(genreID):
gen = genre[genreID]
else:
try:
meta = audio_metadata.load(files_list[0])
gen = meta['tags'].genre
except:
raise Exception("Add genre name to the 'v4_genre_ids' first")
### assigning artist ID := 0 for now, will edit the code later
### TODO: add get_artist() and get_artist_id()
metadata_path = os.path.join(copy_base, 'metadata.csv')
if copy:
with open(metadata_path, 'a', encoding='utf8') as f:
for i in range(len(files_list)):
copy_dest = os.path.join(copy_base, fileIDs[i])
try:
os.mkdir(copy_dest)
except:
pass
new_path = shutil.copy2(files_list[i], copy_dest)
new_path = new_path.strip(copy_base)
f.write(
f"{new_path},{artists[i]},{gen},{gen.lower()},{fileIDs[i]}\n"
)
print(f"file {i + 1} of {len(files_list)} copied!!")
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import os
import shutil
import librosa
import platform_details
from numba import vectorize
import audio_metadata
#
copy_base = platform_details.get_platform_path('Users/theko/Documents/Dataset')
ALL_COMPOSITIONS = [
name for name in os.listdir(copy_base)
if os.path.isdir(os.path.join(copy_base, name))
]
# The following functions (`collate_raga_data`, `collate_and_add_celtic`, and `create_file_artist_genre_ids` ) are
# specific to the current dataset and must not be used
def collate_raga_data(base_dir, labels_json):
IDX = os.listdir(base_dir)
print(labels_json)
LABELS = pd.read_json(labels_json, orient='index')
AUDIO_DATA = pd.DataFrame()
local_artists = {'': []}
AUDIOS = []
Audios_artist = []
Audios_raga = []
Audios_ragaId = []
Audios_fileId = []
genre_id = 0
name = new_name[0]
else:
new_name = fname.replace('.opus', '')
name = new_name
name = name + '.mp3'
out = ffmpeg.input(folder + '/' + fname).output(filename=folder +
"mp3/" + name,
format='mp3')
out.run()
i = i + 1
if __name__ == '__main__':
# files = librosa.util.find_files(copy_base)
# fname = np.random.choice(files)
path = platform_details.get_platform_path("/vismaya/")
extract_audio_from_videos(path)
# print(fname)
# cut_file_1min_segments(fname)
# signal, sr = librosa.load(fname)
# three_mins = 3*get_one_min(sr)
# one_min = get_one_min(sr)
# signal = signal[-three_mins:-one_min]
# raga = get_metadata_windows(fname)
# print(raga)
# pcd_X, pcd_Y = calculate_pcd(signal, sr)
# plt.plot(pcd_X, pcd_Y, label=raga)
# plt.show()
# sample= (1, 2, 3)
# X = np.zeros(5)