def worker(_):
sig = np.random.standard_normal(44100 * 8)
effect = AudioEffectsChain()
effect = effect.pitch(np.random.uniform(-300, 300))
effect = effect.tempo(np.random.uniform(0.8, 1.2))
effect = effect.reverb(np.random.uniform(0, 100))
return effect(sig)
def __call__(self, input):
effect = AudioEffectsChain()
if np.random.uniform() > 0.5:
effect = effect.pitch(np.random.uniform(-300, 300))
if np.random.uniform() > 0.5:
effect = effect.tempo(np.random.uniform(0.8, 1.2))
if np.random.uniform() > 0.5:
effect = effect.reverb(np.random.uniform(0, 100))
# if np.random.uniform() > 0.5:
# effect = effect.overdrive(np.random.uniform(0, 10))
# if np.random.uniform() > 0.5:
# effect = effect.limiter(np.random.uniform(-10, 10))
# if np.random.uniform() > 0.5:
# effect = effect.lowshelf()
# if np.random.uniform() > 0.5:
# effect = effect.highshelf()
return effect(input)
def __call__(self, wav_file):
if not Path(wav_file).exists():
print(wav_file)
raise IOError
sr, wav = scipy.io.wavfile.read(wav_file)
if wav.ndim > 1 and wav.shape[1] > 1:
logger.error("wav file has two or more channels")
sys.exit(1)
if type(wav[0]) is np.int32:
wav = wav.astype('float32', copy=False) / 2147483648.0
elif type(wav[0]) is np.int16:
wav = wav.astype('float32', copy=False) / 32768.0
elif type(wav[0]) is np.uint8:
wav = wav.astype('float32', copy=False) / 256.0 - 128.0
fx = AudioEffectsChain()
if self.resample:
if self.sample_rate > sr:
ratio = int(self.sample_rate / sr)
fx.upsample(ratio)
elif self.sample_rate < sr:
ratio = int(sr / self.sample_rate)
fx.custom(f"downsample {ratio}")
if self.tempo:
tempo_change = np.random.uniform(*self.tempo_range)
fx.tempo(tempo_change, opt_flag="s")
if self.pitch:
pitch_change = np.random.uniform(*self.pitch_range)
fx.pitch(pitch_change)
# dithering
fx.custom(f"dither -s")
wav = fx(wav, sample_in=sr, sample_out=self.sample_rate)
#wav = wav / max(abs(wav))
# normalize audio power
gain = 0.1
wav_energy = np.sqrt(np.sum(np.power(wav, 2)) / wav.size)
wav = gain * wav / wav_energy
# sample-domain padding
if self.padding:
wav = np.pad(wav, self.num_padding, mode='constant')
# sample-domain offset
if self.offset:
offset = np.random.randint(*self.offset_range)
wav = np.roll(wav, offset, axis=0)
if self.noise:
snr = 10.0**(np.random.uniform(*self.noise_range) / 10.0)
noise = np.random.normal(0, 1, wav.shape)
noise_energy = np.sqrt(np.sum(np.power(noise, 2)) / noise.size)
wav = wav + snr * gain * noise / noise_energy
#filename = wav_file.replace(".wav", "_augmented.wav")
#scipy.io.wavfile.write(filename, self.sample_rate, wav)
return torch.FloatTensor(wav)
def main():
# Parsing for command line arguments
parser = argparse.ArgumentParser(
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
description=
("Creates a vaporwave (slowed, with reverb) remix of a given MP3 file, with"
" multiple audio effects available, and the option of playing over a looped"
" GIF as a video."),
)
parser.add_argument(
"-o",
"--output",
dest="output_name",
help=
("Name of output file(s), instead of audio file name with the addition of"
" '_vaporised'."),
type=str,
)
required_arguments = parser.add_argument_group("required arguments")
required_arguments.add_argument(
"-a",
"--audio",
dest="audio_input",
help="Input audio file to vaporise (.mp3)",
type=str,
required=True,
)
audio_arguments = parser.add_argument_group(
"audio arguments",
"these arguments control audio effects that will be applied by default",
)
audio_arguments.add_argument(
"-s",
"--speed",
dest="speed_ratio",
help="Ratio of new playback speed to old speed.",
type=float,
default=0.75,
)
audio_arguments.add_argument(
"-p",
"--pitch",
dest="pitch_shift",
help="Pitch shift (100ths of a semitone).",
type=float,
default=-75,
)
audio_arguments.add_argument(
"-l",
"--lowpass",
dest="lowpass_cutoff",
help="Cutoff for lowpass filter (Hz).",
type=int,
default=3500,
)
audio_arguments_optional = parser.add_argument_group(
"extra audio arguments",
"these arguments control extra, optional audio effects")
audio_arguments_optional.add_argument(
"-b",
"--bass",
dest="bass_boost",
help="Add a bass boost effect (e.g. --bass 3).",
type=int,
default=None,
)
audio_arguments_optional.add_argument(
"-ga",
"--gain",
dest="gain_db",
help="Applies gain (dB).",
type=int,
default=None,
)
audio_arguments_optional.add_argument(
"-op",
"--oops",
dest="oops",
help=
("Applies Out Of Phase Stereo effect. This is sometimes known as the"
" ‘karaoke’ effect as it often has the effect of removing most or all of"
" the vocals from a recording."),
action="store_true",
)
audio_arguments_optional.add_argument(
"-ph",
"--phaser",
dest="phaser",
help="Enable phaser effect.",
action="store_true",
)
audio_arguments_optional.add_argument(
"-tr",
"--tremolo",
dest="tremolo",
help="Enable tremolo effect.",
action="store_true",
)
audio_arguments_optional.add_argument(
"-co",
"--compand",
dest="compand",
help="Enable compand, which compresses the dynamic range of the audio.",
action="store_true",
)
video_arguments = parser.add_argument_group(
"video arguments",
"optional arguments, result in an MP4 video output in addition to the MP3"
" audio",
)
video_arguments.add_argument(
"-g",
"--gif",
dest="gif_file",
help=
("Input GIF file to loop. Without a GIF, only an MP3 is created. With a GIF,"
" an MP4 video is also created."),
type=str,
)
video_arguments.add_argument(
"-sb",
"--sobel",
dest="sobel_filter",
help="Applies a Sobel filter to video output.",
action="store_true",
)
args = parser.parse_args()
# Setting name of output file
if args.output_name is None:
# If no output name is given, add "_vaporised" to input audio file name
audio_input_string = re.sub(".mp3", "", str(args.audio_input))
audio_output = audio_input_string + "_vaporised.mp3"
video_output = audio_input_string + "_vaporised.mp4"
else:
# Otherwise, use the output file name given via the command line
output_string = re.sub(".mp3", "", str(args.output_name))
output_string = re.sub(".mp4", "", str(output_string))
audio_output = output_string + ".mp3"
video_output = output_string + ".mp4"
if args.audio_input == args.output_name:
print("ERROR: Input and output name are identical")
sys.exit()
# Creating an audio effects chain, beginning with...
if args.bass_boost:
# ...bass boost effect
bass_boost = f'{"bass "}{args.bass_boost}'
fx = AudioEffectsChain().custom(bass_boost)
fx = fx.pitch(args.pitch_shift)
else:
# ...pitch shift
fx = AudioEffectsChain().pitch(args.pitch_shift)
# Adding OOPS to audio effects chain
if args.oops:
fx = fx.custom("oops")
# Adding tremolo effect to the audio effects chain
if args.tremolo:
fx = fx.tremolo(freq=500, depth=50)
# Adding phaser to the audio effects chain
if args.phaser:
# fx.phaser(gain_in, gain_out, delay, decay, speed)
fx = fx.phaser(0.9, 0.8, 2, 0.2, 0.5)
# Adding gain to the audio effects chain
if args.gain_db is not None:
fx = fx.gain(db=args.gain_db)
# Adding compand to the audio effects chain
if args.compand:
fx = fx.compand()
# Adding reverb, lowpass filter, speed alteration to audio effects chain
fx = fx.speed(args.speed_ratio).lowpass(args.lowpass_cutoff).reverb()
# Applying audio effects
fx(args.audio_input, audio_output)
def apply_sobel(image):
# returns image with Sobel filter applied
return sobel(image.astype(float))
# Create video if a GIF file is provided
if args.gif_file is None:
# If no GIF is provided, exit here
print("Script finished at",
datetime.datetime.now().strftime("%H:%M:%S"))
print("Vaporised MP3 file (audio):", audio_output)
sys.exit()
else:
# If a GIF is provided, loop it for the length of the vaporised audio file
mp3_movedit = movedit.AudioFileClip(audio_output)
gif_movedit = movedit.VideoFileClip(args.gif_file)
number_of_loops = float(mp3_movedit.duration / gif_movedit.duration)
gif_looped = gif_movedit.loop(number_of_loops)
# Applies Sobel filter to looped GIF, if --sobel is used
if args.sobel_filter:
gif_looped = gif_looped.fl_image(apply_sobel)
gif_looped_with_audio = gif_looped.set_audio(mp3_movedit)
gif_looped_with_audio.write_videofile(video_output)
print("Script finished at",
datetime.datetime.now().strftime("%H:%M:%S"))
print("Vaporised MP3 file (audio):", audio_output)
print("Vaporised MP4 file (video):", video_output)