import re import string def generateReplayGainString(trackGain): return "{0:.2f} dB".format((float(trackGain) + 18.4) * -1) def getBitrateInt(txt): txt = str(txt).lower() if txt in ['flac', 'lossless', '9']: return 9 elif txt in ['mp3', '320', '3']: return 3 elif txt in ['128', '1']: return 1 elif txt in ['360', '360_hq', '15']: return 15 elif txt in ['360_mq', '14']: return 14 elif txt in ['360_lq', '13']: return 13 else: return None def changeCase(str, type): if type == "lower": return str.lower() elif type == "upper": return str.upper() elif type == "start": return string.capwords(str) elif type == "sentence": return str.capitalize() else: return str def removeFeatures(title): clean = title if "(feat." in clean.lower(): pos = clean.lower().find("(feat.") tempTrack = clean[:pos] if ")" in clean: tempTrack += clean[clean.find(")", pos + 1) + 1:] clean = tempTrack.strip() return clean def andCommaConcat(lst): tot = len(lst) result = "" for i, art in enumerate(lst): result += art if tot != i + 1: if tot - 1 == i + 1: result += " & " else: result += ", " return result def getIDFromLink(link, type): if '?' in link: link = link[:link.find('?')] if link.endswith("/"): link = link[:-1] if link.startswith("http") and 'open.spotify.com/' in link: if type == "spotifyplaylist": return link[link.find("/playlist/") + 10:] if type == "spotifytrack": return link[link.find("/track/") + 7:] if type == "spotifyalbum": return link[link.find("/album/") + 7:] elif link.startswith("spotify:"): if type == "spotifyplaylist": return link[link.find("playlist:") + 9:] if type == "spotifytrack": return link[link.find("track:") + 6:] if type == "spotifyalbum": return link[link.find("album:") + 6:] elif type == "artisttop": return re.search(r"\/artist\/(\d+)\/top_track", link)[1] elif type == "artistdiscography": return re.search(r"\/artist\/(\d+)\/discography", link)[1] else: return link[link.rfind("/") + 1:] def getTypeFromLink(link): type = '' if 'spotify' in link: type = 'spotify' if 'playlist' in link: type += 'playlist' elif 'track' in link: type += 'track' elif 'album' in link: type += 'album' elif 'deezer' in link: if '/track' in link: type = 'track' elif '/playlist' in link: type = 'playlist' elif '/album' in link: type = 'album' elif re.search("\/artist\/(\d+)\/top_track", link): type = 'artisttop' elif re.search("\/artist\/(\d+)\/discography", link): type = 'artistdiscography' elif '/artist' in link: type = 'artist' return type def uniqueArray(arr): for iPrinc, namePrinc in enumerate(arr): for iRest, nRest in enumerate(arr): if iPrinc!=iRest and namePrinc.lower() in nRest.lower(): del arr[iRest] return arr