Merge branch 'main' of https://github.com/Footsiefat/zspotify into Footsiefat-main

This commit is contained in:
yiannisha 2021-10-23 18:59:25 +03:00
commit a02b2f03d1
4 changed files with 254 additions and 196 deletions

3
.gitignore vendored
View File

@ -146,3 +146,6 @@ credentials.json
#Download Folder
ZSpotify\ Music/
ZSpotify\ Podcasts/
# Intellij
.idea

View File

@ -16,7 +16,7 @@ Binaries
- Python 3.8 or greater
- ffmpeg*
- Git
- Git**
Python packages:
@ -26,37 +26,52 @@ Python packages:
\*ffmpeg can be installed via apt for Debian-based distros or by downloading the binaries from [ffmpeg.org](https://ffmpeg.org) and placing them in your %PATH% in Windows.
\*\*Git can be installed via apt for Debian-based distros or by downloading the binaries from [git-scm.com](https://git-scm.com/download/win) for Windows.
```
Command line usage:
python zspotify.py Loads search prompt to find then download a specific track, album or playlist
python zspotify.py Loads search prompt to find then download a specific track, album or playlist
python zspotify.py <track/album/playlist/episode url> Downloads the track, album, playlist or podcast episode specified as a command line argument
python zspotify.py <artist url> Downloads all albums by specified artist
Extra command line options:
-p, --playlist Downloads a saved playlist from your account
-ls, --liked-songs Downloads all the liked songs from your account
Special hardcoded options:
Options that can be configured in zs_config.json:
ROOT_PATH Change this path if you don't like the default directory where ZSpotify saves the music
ROOT_PODCAST_PATH Change this path if you don't like the default directory where ZSpotify saves the podcasts
SKIP_EXISTING_FILES Set this to False if you want ZSpotify to overwrite files with the same name rather than skipping the song
SKIP_EXISTING_FILES Set this to false if you want ZSpotify to overwrite files with the same name rather than skipping the song
MUSIC_FORMAT Set this to "ogg" if you would rather that format audio over "mp3"
RAW_AUDIO_AS_IS Set this to True to only stream the audio to a file and do no re-encoding or post processing
FORCE_PREMIUM Set this to True if ZSpotify isn't automatically detecting that you are using a premium account
ANTI_BAN_WAIT_TIME Change this setting if the time waited between bulk downloads is too high or low
OVERRIDE_AUTO_WAIT Change this to True if you want to completely disable the wait between songs for faster downloads with the risk of instability
```
RAW_AUDIO_AS_IS Set this to true to only stream the audio to a file and do no re-encoding or post processing
FORCE_PREMIUM Set this to true if ZSpotify isn't automatically detecting that you are using a premium account
ANTI_BAN_WAIT_TIME Change this setting if the time waited between bulk downloads is too high or low
OVERRIDE_AUTO_WAIT Change this to true if you want to completely disable the wait between songs for faster downloads with the risk of instability
```
### Will my account get banned if I use this tool?
Currently no user has reported their account getting banned after using ZSpotify.
This isn't to say _you_ won't get banned as it is technically against Spotify's TOS.
**Use ZSpotify at your own risk**, the developers of ZSpotify are not responsible if your account gets banned.
### Contributing
Please be sure to lint your code with pylint before issuing a pull-request, thanks!
## **Changelog:**
**v2.1 (23 Oct 2021):**
- Moved configuration from hard-coded values to separate zs_config.json file.
- Add subfolders for each disc.
- Can now search and download all songs by artist.
- Show single progress bar for entire album.
- Added song number at start of track name in albums.
**v2.0 (22 Oct 2021):**
- Added progress bar for downloads.
- Added multi-select support for all results when searching.
- Added GPLv3 Licence.
- Changed welcome banner and removed unnecessary debug print statments.
- Changed welcome banner and removed unnecessary debug print statements.
**v1.9 (22 Oct 2021):**
- Added Gitea mirror for when the Spotify Glowies come to DMCA the shit out of this.

12
zs_config.json Normal file
View File

@ -0,0 +1,12 @@
{
"ROOT_PATH": "ZSpotify Music/",
"ROOT_PODCAST_PATH": "ZSpotify Podcasts/",
"SKIP_EXISTING_FILES": true,
"MUSIC_FORMAT": "mp3",
"RAW_AUDIO_AS_IS": false,
"FORCE_PREMIUM": false,
"ANTI_BAN_WAIT_TIME": 1,
"OVERRIDE_AUTO_WAIT": false,
"CHUNK_SIZE": 50000,
"SPLIT_ALBUM_DISCS": false
}

View File

@ -7,6 +7,7 @@ It's like youtube-dl, but for Spotify.
(Made by Deathmonger/Footsiefat - @doomslayer117:matrix.org)
"""
from getpass import getpass
import json
import os
import os.path
@ -14,36 +15,23 @@ import platform
import re
import sys
import time
from getpass import getpass
import music_tag
import requests
from librespot.audio.decoders import AudioQuality, VorbisOnlyAudioQuality
from librespot.core import Session
from librespot.metadata import TrackId, EpisodeId
import music_tag
from pydub import AudioSegment
import requests
from tqdm import tqdm
QUALITY = None
SESSION: Session = None
sanitize = ["\\", "/", ":", "*", "?", "'", "<", ">", '"']
SANITIZE = ["\\", "/", ":", "*", "?", "'", "<", ">", "\""]
# user-customizable variables that adjust the core functionality of ZSpotify
with open("zs_config.json", encoding="utf-8") as config_file:
ZS_CONFIG = json.load(config_file)
# Hardcoded variables that adjust the core functionality of ZSpotify
ROOT_PATH = "ZSpotify Music/"
ROOT_PODCAST_PATH = "ZSpotify Podcasts/"
SKIP_EXISTING_FILES = True
MUSIC_FORMAT = "mp3" # or "ogg"
RAW_AUDIO_AS_IS = False # set to True if you wish to just save the raw audio
FORCE_PREMIUM = False # set to True if not detecting your premium account automatically
# This is how many seconds ZSpotify waits between downloading tracks so spotify doesn't get out the ban hammer
ANTI_BAN_WAIT_TIME = 1
# Set this to True to not wait at all between tracks and just go balls to the wall
OVERRIDE_AUTO_WAIT = False
CHUNK_SIZE = 50000
# miscellaneous functions for general use
@ -59,14 +47,13 @@ def clear():
def wait(seconds: int = 3):
""" Pause for a set number of seconds """
for i in range(seconds)[::-1]:
print("\rWait for %d second(s)..." % (i + 1), end="")
print(f"\rWait for {i + 1} second(s)...", end="")
time.sleep(1)
def sanitize_data(value):
""" Returns given string with problematic removed """
global sanitize
for i in sanitize:
for i in SANITIZE:
value = value.replace(i, "")
return value.replace("|", "-")
@ -77,12 +64,11 @@ def split_input(selection):
if "-" in selection:
for number in range(int(selection.split("-")[0]), int(selection.split("-")[1]) + 1):
inputs.append(number)
return inputs
else:
selections = selection.split(",")
for i in selections:
inputs.append(i.strip())
return inputs
return inputs
def splash():
@ -99,7 +85,7 @@ def splash():
# two mains functions for logging in and doing client stuff
def login():
""" Authenticates with Spotify and saves credentials to a file """
global SESSION
global SESSION # pylint: disable=global-statement
if os.path.isfile("credentials.json"):
try:
@ -117,9 +103,9 @@ def login():
pass
def client():
def client(): # pylint: disable=too-many-branches,too-many-statements
""" Connects to spotify to perform query's and get songs to download """
global QUALITY, SESSION
global QUALITY # pylint: disable=global-statement
splash()
token = SESSION.tokens().get("user-read-email")
@ -131,20 +117,45 @@ def client():
print("[ DETECTED FREE ACCOUNT - USING HIGH QUALITY ]\n\n")
QUALITY = AudioQuality.HIGH
if len(sys.argv) > 1:
if sys.argv[1] == "-p" or sys.argv[1] == "--playlist":
download_from_user_playlist()
elif sys.argv[1] == "-ls" or sys.argv[1] == "--liked-songs":
for song in get_saved_tracks(token):
if not song['track']['name']:
print(
"### SKIPPING: SONG DOES NOT EXISTS ON SPOTIFY ANYMORE ###")
else:
download_track(song['track']['id'], "Liked Songs/")
print("\n")
while True:
if len(sys.argv) > 1:
if sys.argv[1] == "-p" or sys.argv[1] == "--playlist":
download_from_user_playlist()
elif sys.argv[1] == "-ls" or sys.argv[1] == "--liked-songs":
for song in get_saved_tracks(token):
if not song["track"]["name"]:
print(
"### SKIPPING: SONG DOES NOT EXISTS ON SPOTIFY ANYMORE ###")
else:
download_track(song["track"]["id"], "Liked Songs/")
print("\n")
else:
track_id_str, album_id_str, playlist_id_str, episode_id_str, show_id_str, artist_id_str = regex_input_for_urls(
sys.argv[1])
if track_id_str is not None:
download_track(track_id_str)
elif artist_id_str is not None:
download_artist_albums(artist_id_str)
elif album_id_str is not None:
download_album(album_id_str)
elif playlist_id_str is not None:
playlist_songs = get_playlist_songs(token, playlist_id_str)
name, _ = get_playlist_info(token, playlist_id_str)
for song in playlist_songs:
download_track(song["track"]["id"],
sanitize_data(name) + "/")
print("\n")
elif episode_id_str is not None:
download_episode(episode_id_str)
elif show_id_str is not None:
for episode in get_show_episodes(token, show_id_str):
download_episode(episode)
else:
track_id_str, album_id_str, playlist_id_str, episode_id_str, show_id_str = regex_input_for_urls(
sys.argv[1])
search_text = input("Enter search or URL: ")
track_id_str, album_id_str, playlist_id_str, episode_id_str, show_id_str, artist_id_str = regex_input_for_urls(
search_text)
if track_id_str is not None:
download_track(track_id_str)
@ -154,9 +165,9 @@ def client():
download_album(album_id_str)
elif playlist_id_str is not None:
playlist_songs = get_playlist_songs(token, playlist_id_str)
name, creator = get_playlist_info(token, playlist_id_str)
name, _ = get_playlist_info(token, playlist_id_str)
for song in playlist_songs:
download_track(song['track']['id'],
download_track(song["track"]["id"],
sanitize_data(name) + "/")
print("\n")
elif episode_id_str is not None:
@ -164,37 +175,13 @@ def client():
elif show_id_str is not None:
for episode in get_show_episodes(token, show_id_str):
download_episode(episode)
else:
search_text = input("Enter search or URL: ")
track_id_str, album_id_str, playlist_id_str, episode_id_str, show_id_str, artist_id_str = regex_input_for_urls(
search_text)
if track_id_str is not None:
download_track(track_id_str)
elif artist_id_str is not None:
download_artist_albums(artist_id_str)
elif album_id_str is not None:
download_album(album_id_str)
elif playlist_id_str is not None:
playlist_songs = get_playlist_songs(token, playlist_id_str)
name, creator = get_playlist_info(token, playlist_id_str)
for song in playlist_songs:
download_track(song['track']['id'],
sanitize_data(name) + "/")
print("\n")
elif episode_id_str is not None:
download_episode(episode_id_str)
elif show_id_str is not None:
for episode in get_show_episodes(token, show_id_str):
download_episode(episode)
else:
search(search_text)
# wait()
else:
search(search_text)
# wait()
def regex_input_for_urls(search_input):
def regex_input_for_urls(search_input): # pylint: disable=too-many-locals
""" Since many kinds of search may be passed at the command line, process them all here. """
track_uri_search = re.search(
r"^spotify:track:(?P<TrackID>[0-9a-zA-Z]{22})$", search_input)
track_url_search = re.search(
@ -237,7 +224,6 @@ def regex_input_for_urls(search_input):
search_input,
)
if track_uri_search is not None or track_url_search is not None:
track_id_str = (track_uri_search
if track_uri_search is not None else
@ -275,32 +261,31 @@ def regex_input_for_urls(search_input):
if artist_uri_search is not None or artist_url_search is not None:
artist_id_str = (artist_uri_search
if artist_uri_search is not None else
artist_url_search).group("ArtistID")
if artist_uri_search is not None else
artist_url_search).group("ArtistID")
else:
artist_id_str = None
return track_id_str, album_id_str, playlist_id_str, episode_id_str, show_id_str, artist_id_str
def get_episode_info(episode_id_str):
def get_episode_info(episode_id_str): # pylint: disable=missing-function-docstring
token = SESSION.tokens().get("user-read-email")
info = json.loads(requests.get("https://api.spotify.com/v1/episodes/" +
episode_id_str, headers={"Authorization": "Bearer %s" % token}).text)
info = json.loads(requests.get(f"https://api.spotify.com/v1/episodes/{episode_id_str}",
headers={"Authorization": f"Bearer {token}"}).text)
if "error" in info:
return None, None
else:
# print(info['images'][0]['url'])
return sanitize_data(info["show"]["name"]), sanitize_data(info["name"])
# print(info["images"][0]["url"])
return sanitize_data(info["show"]["name"]), sanitize_data(info["name"])
def get_show_episodes(access_token, show_id_str):
def get_show_episodes(access_token, show_id_str): # pylint: disable=missing-function-docstring
episodes = []
headers = {'Authorization': f'Bearer {access_token}'}
headers = {"Authorization": f"Bearer {access_token}"}
resp = requests.get(
f'https://api.spotify.com/v1/shows/{show_id_str}/episodes', headers=headers).json()
f"https://api.spotify.com/v1/shows/{show_id_str}/episodes", headers=headers).json()
for episode in resp["items"]:
episodes.append(episode["id"])
@ -308,9 +293,7 @@ def get_show_episodes(access_token, show_id_str):
return episodes
def download_episode(episode_id_str):
global ROOT_PODCAST_PATH, MUSIC_FORMAT
def download_episode(episode_id_str): # pylint: disable=missing-function-docstring
podcast_name, episode_name = get_episode_info(episode_id_str)
extra_paths = podcast_name + "/"
@ -326,28 +309,28 @@ def download_episode(episode_id_str):
# print("### DOWNLOADING '" + podcast_name + " - " +
# episode_name + "' - THIS MAY TAKE A WHILE ###")
if not os.path.isdir(ROOT_PODCAST_PATH + extra_paths):
os.makedirs(ROOT_PODCAST_PATH + extra_paths)
os.makedirs(ZS_CONFIG["ROOT_PODCAST_PATH"] +
extra_paths, exist_ok=True)
total_size = stream.input_stream.size
with open(ROOT_PODCAST_PATH + extra_paths + filename + ".wav", 'wb') as file, tqdm(
with open(ZS_CONFIG["ROOT_PODCAST_PATH"] + extra_paths + filename + ".wav", "wb") as file, tqdm(
desc=filename,
total=total_size,
unit='B',
unit="B",
unit_scale=True,
unit_divisor=1024
) as bar:
for _ in range(int(total_size / CHUNK_SIZE) + 1):
bar.update(file.write(
stream.input_stream.stream().read(CHUNK_SIZE)))
) as p_bar:
for _ in range(int(total_size / ZS_CONFIG["CHUNK_SIZE"]) + 1):
p_bar.update(file.write(
stream.input_stream.stream().read(ZS_CONFIG["CHUNK_SIZE"])))
# convert_audio_format(ROOT_PODCAST_PATH +
# convert_audio_format(ZS_CONFIG["ROOT_PODCAST_PATH"] +
# extra_paths + filename + ".wav")
# related functions that do stuff with the spotify API
def search(search_term):
def search(search_term): # pylint: disable=too-many-locals,too-many-branches
""" Searches Spotify's API for relevant data """
token = SESSION.tokens().get("user-read-email")
@ -380,7 +363,7 @@ def search(search_term):
if split == "-t" or split == "-type":
allowed_types = ["track", "playlist", "album"]
allowed_types = ["track", "playlist", "album", "artist"]
for i in range(index+1, len(splits)):
if splits[i][0] == "-":
break
@ -392,7 +375,7 @@ def search(search_term):
params["type"].add(splits[i])
if len(params["type"]) == 0:
params["type"] = {"track", "album", "playlist"}
params["type"] = {"track", "album", "playlist", "artist"}
# Clean search term
search_term_list = []
@ -412,7 +395,7 @@ def search(search_term):
"q": params["q"],
"type": ",".join(params["type"])
},
headers={"Authorization": "Bearer %s" % token},
headers={"Authorization": f"Bearer {token}"},
)
# print(resp.json())
@ -508,6 +491,30 @@ def search(search_term):
else:
total_playlists = 0
if "artist" in params["type"]:
artists = resp.json()["artists"]["items"]
if len(artists) > 0:
print("### ARTISTS ###")
for artist in artists:
# collect needed data
dic = {
"id" : artist["id"],
"name" : artist["name"],
"type" : "artist",
}
dics.append(dic)
print("{}, {}".format(
enum,
dic["name"],
))
enum += 1
total_artists = enum - total_tracks - total_albums - total_playlists - 1
print("\n")
else:
total_artists = 0
if total_tracks + total_albums + total_playlists == 0:
print("NO RESULTS FOUND - EXITING...")
else:
@ -525,6 +532,9 @@ def search(search_term):
# if request is for album
if dic["type"] == "album":
download_album(dic["id"])
# if request is for artist
if dic["type"] == "artist":
download_artist_albums(dic["id"])
# if request is for playlist
if dic["type"] == "playlist":
playlist_songs = get_playlist_songs(token, dic["id"])
@ -540,53 +550,52 @@ def get_song_info(song_id):
token = SESSION.tokens().get("user-read-email")
info = json.loads(requests.get("https://api.spotify.com/v1/tracks?ids=" + song_id +
'&market=from_token', headers={"Authorization": "Bearer %s" % token}).text)
"&market=from_token", headers={"Authorization": f"Bearer {token}"}).text)
artists = []
for data in info['tracks'][0]['artists']:
artists.append(sanitize_data(data['name']))
album_name = sanitize_data(info['tracks'][0]['album']["name"])
name = sanitize_data(info['tracks'][0]['name'])
image_url = info['tracks'][0]['album']['images'][0]['url']
release_year = info['tracks'][0]['album']['release_date'].split("-")[0]
disc_number = info['tracks'][0]['disc_number']
track_number = info['tracks'][0]['track_number']
scraped_song_id = info['tracks'][0]['id']
is_playable = info['tracks'][0]['is_playable']
for data in info["tracks"][0]["artists"]:
artists.append(sanitize_data(data["name"]))
album_name = sanitize_data(info["tracks"][0]["album"]["name"])
name = sanitize_data(info["tracks"][0]["name"])
image_url = info["tracks"][0]["album"]["images"][0]["url"]
release_year = info["tracks"][0]["album"]["release_date"].split("-")[0]
disc_number = info["tracks"][0]["disc_number"]
track_number = info["tracks"][0]["track_number"]
scraped_song_id = info["tracks"][0]["id"]
is_playable = info["tracks"][0]["is_playable"]
return artists, album_name, name, image_url, release_year, disc_number, track_number, scraped_song_id, is_playable
def check_premium():
""" If user has spotify premium return true """
global FORCE_PREMIUM
return bool((SESSION.get_user_attribute("type") == "premium") or FORCE_PREMIUM)
return bool((SESSION.get_user_attribute("type") == "premium") or ZS_CONFIG["FORCE_PREMIUM"])
# Functions directly related to modifying the downloaded audio and its metadata
def convert_audio_format(filename):
""" Converts raw audio into playable mp3 or ogg vorbis """
global MUSIC_FORMAT
# print("### CONVERTING TO " + MUSIC_FORMAT.upper() + " ###")
# print("### CONVERTING TO " + ZS_CONFIG["MUSIC_FORMAT"].upper() + " ###")
raw_audio = AudioSegment.from_file(filename, format="ogg",
frame_rate=44100, channels=2, sample_width=2)
if QUALITY == AudioQuality.VERY_HIGH:
bitrate = "320k"
else:
bitrate = "160k"
raw_audio.export(filename, format=MUSIC_FORMAT, bitrate=bitrate)
raw_audio.export(
filename, format=ZS_CONFIG["MUSIC_FORMAT"], bitrate=bitrate)
def set_audio_tags(filename, artists, name, album_name, release_year, disc_number, track_number):
def set_audio_tags(filename, artists, name, album_name, release_year, disc_number, track_number): # pylint: disable=too-many-arguments
""" sets music_tag metadata """
# print("### SETTING MUSIC TAGS ###")
tags = music_tag.load_file(filename)
tags['artist'] = conv_artist_format(artists)
tags['tracktitle'] = name
tags['album'] = album_name
tags['year'] = release_year
tags['discnumber'] = disc_number
tags['tracknumber'] = track_number
tags["artist"] = conv_artist_format(artists)
tags["tracktitle"] = name
tags["album"] = album_name
tags["year"] = release_year
tags["discnumber"] = disc_number
tags["tracknumber"] = track_number
tags.save()
@ -595,7 +604,7 @@ def set_music_thumbnail(filename, image_url):
# print("### SETTING THUMBNAIL ###")
img = requests.get(image_url).content
tags = music_tag.load_file(filename)
tags['artwork'] = img
tags["artwork"] = img
tags.save()
@ -615,14 +624,14 @@ def get_all_playlists(access_token):
offset = 0
while True:
headers = {'Authorization': f'Bearer {access_token}'}
params = {'limit': limit, 'offset': offset}
headers = {"Authorization": f"Bearer {access_token}"}
params = {"limit": limit, "offset": offset}
resp = requests.get("https://api.spotify.com/v1/me/playlists",
headers=headers, params=params).json()
offset += limit
playlists.extend(resp['items'])
playlists.extend(resp["items"])
if len(resp['items']) < limit:
if len(resp["items"]) < limit:
break
return playlists
@ -635,14 +644,14 @@ def get_playlist_songs(access_token, playlist_id):
limit = 100
while True:
headers = {'Authorization': f'Bearer {access_token}'}
params = {'limit': limit, 'offset': offset}
headers = {"Authorization": f"Bearer {access_token}"}
params = {"limit": limit, "offset": offset}
resp = requests.get(
f'https://api.spotify.com/v1/playlists/{playlist_id}/tracks', headers=headers, params=params).json()
f"https://api.spotify.com/v1/playlists/{playlist_id}/tracks", headers=headers, params=params).json()
offset += limit
songs.extend(resp['items'])
songs.extend(resp["items"])
if len(resp['items']) < limit:
if len(resp["items"]) < limit:
break
return songs
@ -650,10 +659,11 @@ def get_playlist_songs(access_token, playlist_id):
def get_playlist_info(access_token, playlist_id):
""" Returns information scraped from playlist """
headers = {'Authorization': f'Bearer {access_token}'}
headers = {"Authorization": f"Bearer {access_token}"}
resp = requests.get(
f'https://api.spotify.com/v1/playlists/{playlist_id}?fields=name,owner(display_name)&market=from_token', headers=headers).json()
return resp['name'].strip(), resp['owner']['display_name'].strip()
f"https://api.spotify.com/v1/playlists/{playlist_id}?fields=name,owner(display_name)&market=from_token",
headers=headers).json()
return resp["name"].strip(), resp["owner"]["display_name"].strip()
# Extra functions directly related to spotify albums
@ -664,14 +674,14 @@ def get_album_tracks(access_token, album_id):
limit = 50
while True:
headers = {'Authorization': f'Bearer {access_token}'}
params = {'limit': limit, 'offset': offset}
headers = {"Authorization": f"Bearer {access_token}"}
params = {"limit": limit, "offset": offset}
resp = requests.get(
f'https://api.spotify.com/v1/albums/{album_id}/tracks', headers=headers, params=params).json()
f"https://api.spotify.com/v1/albums/{album_id}/tracks", headers=headers, params=params).json()
offset += limit
songs.extend(resp['items'])
songs.extend(resp["items"])
if len(resp['items']) < limit:
if len(resp["items"]) < limit:
break
return songs
@ -679,21 +689,25 @@ def get_album_tracks(access_token, album_id):
def get_album_name(access_token, album_id):
""" Returns album name """
headers = {'Authorization': f'Bearer {access_token}'}
headers = {"Authorization": f"Bearer {access_token}"}
resp = requests.get(
f'https://api.spotify.com/v1/albums/{album_id}', headers=headers).json()
return resp['artists'][0]['name'], sanitize_data(resp['name'])
f"https://api.spotify.com/v1/albums/{album_id}", headers=headers).json()
return resp["artists"][0]["name"], sanitize_data(resp["name"])
# Extra functions directly related to spotify artists
def get_artist_albums(access_token, artist_id):
""" Returns artist's albums """
headers = {'Authorization': f'Bearer {access_token}'}
headers = {"Authorization": f"Bearer {access_token}"}
resp = requests.get(
f'https://api.spotify.com/v1/artists/{artist_id}/albums', headers=headers).json()
f"https://api.spotify.com/v1/artists/{artist_id}/albums", headers=headers).json()
# Return a list each album's id
return [resp['items'][i]['id'] for i in range(len(resp['items']))]
return [resp["items"][i]["id"] for i in range(len(resp["items"]))]
# Extra functions directly related to our saved tracks
def get_saved_tracks(access_token):
""" Returns user's saved tracks """
songs = []
@ -701,39 +715,47 @@ def get_saved_tracks(access_token):
limit = 50
while True:
headers = {'Authorization': f'Bearer {access_token}'}
params = {'limit': limit, 'offset': offset}
resp = requests.get('https://api.spotify.com/v1/me/tracks',
headers = {"Authorization": f"Bearer {access_token}"}
params = {"limit": limit, "offset": offset}
resp = requests.get("https://api.spotify.com/v1/me/tracks",
headers=headers, params=params).json()
offset += limit
songs.extend(resp['items'])
songs.extend(resp["items"])
if len(resp['items']) < limit:
if len(resp["items"]) < limit:
break
return songs
# Functions directly related to downloading stuff
def download_track(track_id_str: str, extra_paths=""):
def download_track(track_id_str: str, extra_paths="", prefix=False, prefix_value="", disable_progressbar=False): # pylint: disable=too-many-locals,too-many-branches
""" Downloads raw song audio from Spotify """
global ROOT_PATH, SKIP_EXISTING_FILES, MUSIC_FORMAT, RAW_AUDIO_AS_IS, ANTI_BAN_WAIT_TIME, OVERRIDE_AUTO_WAIT
try:
artists, album_name, name, image_url, release_year, disc_number, track_number, scraped_song_id, is_playable = get_song_info(
track_id_str)
song_name = artists[0] + " - " + name
filename = ROOT_PATH + extra_paths + song_name + '.' + MUSIC_FORMAT
except Exception as e:
if prefix:
song_name = f"{prefix_value.zfill(2)} - {song_name}" if prefix_value.isdigit(
) else f"{prefix_value} - {song_name}"
if ZS_CONFIG["SPLIT_ALBUM_DISCS"]:
filename = os.path.join(ZS_CONFIG["ROOT_PATH"], extra_paths, "Disc " + str(
disc_number) + "/" + song_name + "." + ZS_CONFIG["MUSIC_FORMAT"])
else:
filename = os.path.join(ZS_CONFIG["ROOT_PATH"], extra_paths,
song_name + "." + ZS_CONFIG["MUSIC_FORMAT"])
except Exception as err: # pylint: disable=broad-except,unused-variable
print("### SKIPPING SONG - FAILED TO QUERY METADATA ###")
# print(e)
# print(err)
else:
try:
if not is_playable:
print("### SKIPPING:", song_name,
"(SONG IS UNAVAILABLE) ###")
else:
if os.path.isfile(filename) and os.path.getsize(filename) and SKIP_EXISTING_FILES:
if os.path.isfile(filename) and os.path.getsize(filename) and ZS_CONFIG["SKIP_EXISTING_FILES"]:
print("### SKIPPING:", song_name,
"(SONG ALREADY EXISTS) ###")
else:
@ -747,30 +769,35 @@ def download_track(track_id_str: str, extra_paths=""):
track_id, VorbisOnlyAudioQuality(QUALITY), False, None)
# print("### DOWNLOADING RAW AUDIO ###")
if not os.path.isdir(ROOT_PATH + extra_paths):
os.makedirs(ROOT_PATH + extra_paths)
if ZS_CONFIG["SPLIT_ALBUM_DISCS"]:
os.makedirs(
ZS_CONFIG["ROOT_PATH"] + extra_paths + "/Disc " + str(disc_number) + "/", exist_ok=True)
else:
os.makedirs(ZS_CONFIG["ROOT_PATH"] +
extra_paths, exist_ok=True)
total_size = stream.input_stream.size
with open(filename, 'wb') as file, tqdm(
with open(filename, "wb") as file, tqdm(
desc=song_name,
total=total_size,
unit='B',
unit="B",
unit_scale=True,
unit_divisor=1024
) as bar:
for _ in range(int(total_size / CHUNK_SIZE) + 1):
bar.update(file.write(
stream.input_stream.stream().read(CHUNK_SIZE)))
unit_divisor=1024,
disable=disable_progressbar
) as p_bar:
for _ in range(int(total_size / ZS_CONFIG["CHUNK_SIZE"]) + 1):
p_bar.update(file.write(
stream.input_stream.stream().read(ZS_CONFIG["CHUNK_SIZE"])))
if not RAW_AUDIO_AS_IS:
if not ZS_CONFIG["RAW_AUDIO_AS_IS"]:
convert_audio_format(filename)
set_audio_tags(filename, artists, name, album_name,
release_year, disc_number, track_number)
set_music_thumbnail(filename, image_url)
if not OVERRIDE_AUTO_WAIT:
time.sleep(ANTI_BAN_WAIT_TIME)
except:
if not ZS_CONFIG["OVERRIDE_AUTO_WAIT"]:
time.sleep(ZS_CONFIG["ANTI_BAN_WAIT_TIME"])
except Exception: # pylint: disable=broad-except
print("### SKIPPING:", song_name,
"(GENERAL DOWNLOAD ERROR) ###")
if os.path.exists(filename):
@ -782,9 +809,10 @@ def download_album(album):
token = SESSION.tokens().get("user-read-email")
artist, album_name = get_album_name(token, album)
tracks = get_album_tracks(token, album)
for track in tracks:
download_track(track['id'], artist + " - " + album_name + "/")
print("\n")
for num, track in tqdm(enumerate(tracks, start=1), unit_scale=True, unit="Song", total=len(tracks)):
download_track(track["id"], f"{artist}/{album_name}",
prefix=True, prefix_value=str(num), disable_progressbar=True)
def download_artist_albums(artist):
""" Downloads albums of an artist """
@ -793,17 +821,18 @@ def download_artist_albums(artist):
for album_id in albums:
download_album(album_id)
def download_playlist(playlists, playlist_choice):
"""Downloads all the songs from a playlist"""
token = SESSION.tokens().get("user-read-email")
playlist_songs = get_playlist_songs(
token, playlists[int(playlist_choice) - 1]['id'])
token, playlists[int(playlist_choice) - 1]["id"])
for song in playlist_songs:
if song['track']['id'] is not None:
download_track(song['track']['id'], sanitize_data(
playlists[int(playlist_choice) - 1]['name'].strip()) + "/")
if song["track"]["id"] is not None:
download_track(song["track"]["id"], sanitize_data(
playlists[int(playlist_choice) - 1]["name"].strip()) + "/")
print("\n")
@ -814,7 +843,7 @@ def download_from_user_playlist():
count = 1
for playlist in playlists:
print(str(count) + ": " + playlist['name'].strip())
print(str(count) + ": " + playlist["name"].strip())
count += 1
print("\n> SELECT A PLAYLIST BY ID")
@ -827,7 +856,7 @@ def download_from_user_playlist():
download_playlist(playlists, playlist_choices[0])
else:
start = int(playlist_choices[0])
end = int(playlist_choices[1])+1
end = int(playlist_choices[1]) + 1
print(f"Downloading from {start} to {end}...")
@ -840,10 +869,9 @@ def download_from_user_playlist():
# Core functions here
def check_raw():
global RAW_AUDIO_AS_IS, MUSIC_FORMAT
if RAW_AUDIO_AS_IS:
MUSIC_FORMAT = "wav"
def check_raw(): # pylint: disable=missing-function-docstring
if ZS_CONFIG["RAW_AUDIO_AS_IS"]:
ZS_CONFIG["MUSIC_FORMAT"] = "wav"
def main():