def get_vocal(mix_file, bg_file, lyric_file, out_file="out.wav"):
print("\ndeal with ---", bg_file, "---")
# 讀audio檔
mix, sr = librosa.load(mix_file, sr=None)
bg, sr = librosa.load(bg_file, sr=None)
# 取得前奏的時間
l = lyric_parser.Lyric(lyric_file)
time = l.get_time_before_vocal()
# 單位變換，從ms換成sample
time = ms2sample(time - 1000, sr)
time = time // 2
# 前處理
mix = mute_start(mix)
bg = mute_start(bg)
# 計算位移，音量
shift, vol = compute_t_v(mix, bg, time)
# 前處理
mix, bg = npp.pad_the_same(mix, bg)
mix2 = npp.right_shift(mix, shift)
# 訊號相減
result = mix2 * vol - bg
# try
# result = mix2*1.5151645 - bg
# 輸出
librosa.output.write_wav(out_file, result, sr)
librosa.output.cl
def get_vocal_mp3_void(mix_file, bg_file, lyric_file, out_file="out.mp3"):
# card_sample = ['clear', 'shift', 'volume_ratio',
# 'song_sr', 'bg_sr', 'result_sr',
# 'song_duration', 'bg_duration', 'result_duration', 'result_norm_ave',
# 'error_text']
card = [' ', ' ', ' ',
' ', ' ', ' ',
' ', ' ', ' ', ' ', ' ']
# 讀audio檔
bg, sr_bg = librosa.load(bg_file, sr=None)
mix, sr_mix = librosa.load(mix_file, sr=None)
# TODO: plt
# import matplotlib.pyplot as plt
# mix1, sr_mix = librosa.load(mix_file, sr=None, duration=10)
# bg1, sr_bg = librosa.load(bg_file, sr=None, duration=10)
# import librosa.display as d
# import matplotlib.pyplot as plt
# plt.figure()
# d.waveplot(mix1, sr=sr_mix)
# plt.title('mix')
# plt.figure()
# d.waveplot(bg1, sr=sr_bg)
# plt.title('bg')
# ###
song_duration = len(mix)
bg_duration = len(bg)
if sr_mix != sr_bg:
card = [0, ' ', ' ',
sr_mix, sr_bg, ' ',
song_duration, bg_duration, ' ', ' ', 'sr not match']
# 輸出result with norm_value_fileName，回傳card
fn = os.path.basename(out_file)
fn = "sr_NOTmatch_" + fn
dn = os.path.dirname(out_file)
dn = os.path.join(dn, "error")
fn = os.path.join(dn, fn)
# Write null file with only filename
result = np.array([])
np.savetxt(fn, result)
return card
elif sr_mix == 44100:
# shift = -1728
shift = 0
elif sr_mix == 22050:
# shift = -1728 / 2
shift = 0
else:
card = [0, ' ', ' ',
sr_mix, sr_bg, ' ',
song_duration, bg_duration, ' ', ' ', 'get_vocal error of if(sr_mix != sr_bg) function']
# 輸出result with norm_value_fileName，回傳card
fn = os.path.basename(out_file)
fn = "sr_NOTmatch_" + fn
dn = os.path.dirname(out_file)
dn = os.path.join(dn, "error")
fn = os.path.join(dn, fn)
# Write null file with only filename
result = np.array([])
np.savetxt(fn, result)
# Write null file with only filename
result = np.array([])
np.savetxt(fn, result)
return card
# sr is coming~
sr = sr_mix
sr_result = sr
# get intro time
l = lyric_parser.Lyric(lyric_file)
time = l.get_time_before_vocal()
if time < 1000:
card[0] = 0
card[10] = 'lyric might have wrong start (intro time < 1000)'
# 輸出result with norm_value_fileName，回傳card
fn = os.path.basename(out_file)
fn = "lyric_wrongSTART_" + fn
dn = os.path.dirname(out_file)
dn = os.path.join(dn, "error")
fn = os.path.join(dn, fn)
# Write null file with only filename
result = np.array([])
np.savetxt(fn, result)
return card
# change ms 2 sample rate
time = ms2sample(time - 1000, sr)
time = time // 2
# mute start to avoid blast sound
mix = mute_start(mix)
bg = mute_start(bg)
# calculate volume rate
vol, bg_norm = compute_vol_ratio_bgNorm(mix, bg, time)
# some preprocessing and shift
mix, bg = npp.pad_the_same(mix, bg)
mix_shift = npp.right_shift(mix, shift)
# devocal
result = mix_shift[:song_duration] * vol - bg[:song_duration]
# avoid lst blast （1728+1000）
result = result[:len(result) - 3000]
result_duration = len(result)
# clear or not
result_norm = linalg.norm(result[:time], ord=1)
result_norm_ave = result_norm / time
result_norm_ave = int(result_norm_ave * 10000000)
if result_norm < bg_norm:
fn = os.path.basename(out_file)
fn = str(int(result_norm_ave)) + "_" + fn
dn = os.path.dirname(out_file)
dn = os.path.join(dn, "perfect")
fn = os.path.join(dn, fn)
# TODO : ouput result mp3
write_arr_mp3(fn, result, sr)
print(
" result_norm < bg_norm * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ! ! ! de success ! ! !",
end=" ")
print(datetime.datetime.now())
card = [1, shift, vol,
sr_mix, sr_bg, sr_result,
song_duration, bg_duration, result_duration, result_norm_ave, ' ']
return card
else:
# 輸出result with norm_value_fileName，回傳card
fn = os.path.basename(out_file)
fn = "norm_" + str(int(result_norm_ave)) + "_" + fn
dn = os.path.dirname(out_file)
dn = os.path.join(dn, "error")
fn = os.path.join(dn, fn)
# Write null file with only filename
result = np.array([])
np.savetxt(fn, result)
# write_arr_mp3(fn, result, sr)
print(" " + fn)
print(
" result_norm > bg_norm - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - ! ! ! de fail ! ! !",
end=" ")
print(datetime.datetime.now())
card = [0, shift, vol,
sr_mix, sr_bg, sr_result,
song_duration, bg_duration, result_duration, result_norm_ave, 'de fail']
return card