def __wavenet_encode(self, audio):
# Load the model weights.
checkpoint_path = '../wavenet-ckpt/model.ckpt-200000'
# Load and downsample the audio.
# neural_sample_rate = 16000
# audio = utils.load_audio(self._audio_dir + file_path,
# sample_length=400000,
# sr=neural_sample_rate)
# Pass the audio through the first half of the autoencoder,
# to get a list of latent variables that describe the sound.
# Note that it would be quicker to pass a batch of audio
# to fastgen.
audio = np.squeeze(audio)
# print(audio.shape)
# print(len(audio))
if(len(audio.shape) > 1):
encoding = fastgen.encode(audio, checkpoint_path, audio.shape[1])
else:
encoding = fastgen.encode(audio, checkpoint_path, len(audio))
# print("Pre: " + str(encoding.shape))
encoding = self.__std_dev_mean_noise(encoding)
# print("Post: " + str(encoding.shape))
# Reshape to a single sound.
return encoding
def wavenet_encode(self, file_path, **kwargs):
if os.path.exists('../../Pretrained_models/wavenet-ckpt/'):
# Load the model weights.
checkpoint_path = '../../Pretrained_models/wavenet-ckpt/model.ckpt-200000'
else:
raise Exception(
'you should download pretrained model to pretrained_models folder make prediction, the link is: http://download.magenta.tensorflow.org/models/nsynth/wavenet-ckpt.tar'
)
# Load and downsample the audio.
neural_sample_rate = 16000
audio = utils.load_audio(file_path,
sample_length=400000,
sr=neural_sample_rate)
# Pass the audio through the first half of the autoencoder,
# to get a list of latent variables that describe the sound.
# Note that it would be quicker to pass a batch of audio
# to fastgen.
encoding = fastgen.encode(audio, checkpoint_path, len(audio))
# Reshape to a single sound.
return encoding.reshape((-1, 16))
def encode(wav_filenames: List[str],
checkpoint: str = "checkpoints/wavenet-ckpt/model.ckpt-200000",
sample_length: int = 16000,
sample_rate: int = 16000) -> List[np.ndarray]:
"""
Encodes the list of filename to encodings by loading the wav files,
encoding them using fastgen, and returning the result.
:param wav_filenames: the list of filenames to encode, they need to be
present in the "sound" folder
:param checkpoint: the checkpoint folder
:param sample_length: the sample length, can be calculated by multiplying
the desired number of seconds by 16000
:param sample_rate: the sample rate, should be 16000
"""
if not wav_filenames:
return []
# Loads the audio for each filenames
audios = []
for wav_filename in wav_filenames:
audio = utils.load_audio(os.path.join("sounds", wav_filename),
sample_length=sample_length,
sr=sample_rate)
audios.append(audio)
# Encodes the audio for each new wav
audios = np.array(audios)
encodings = fastgen.encode(audios, checkpoint, sample_length)
return encodings
def load_encoding(fname,
sample_lenght=None,
sr=16000,
ckpt='model.ckpt-200000'):
audio = utils.load_audio(fname, sample_length=sample_lenght, sr=sr)
encoding = fastgen.encode(audio, ckpt, sample_lenght)
return audio, encoding
def wavenet_encode(file_path):
neural_sample_rate = 16000
audio = utils.load_audio(file_path,
sample_length=400000,
sr=neural_sample_rate)
encoding = fastgen.encode(audio, '../wavenet-ckpt/model.ckpt-200000',
len(audio))
return encoding.reshape((-1, 16))
def main(unused_argv=None):
os.environ["CUDA_VISIBLE_DEVICES"] = str(FLAGS.gpu_number)
source_path = utils.shell_path(FLAGS.source_path)
checkpoint_path = utils.shell_path(FLAGS.checkpoint_path)
save_path = utils.shell_path(FLAGS.save_path)
if not save_path:
raise RuntimeError("Must specify a save_path.")
tf.logging.set_verbosity(FLAGS.log)
# Generate from wav files
if tf.gfile.IsDirectory(source_path):
files = tf.gfile.ListDirectory(source_path)
exts = [os.path.splitext(f)[1] for f in files]
if ".wav" in exts:
postfix = ".wav"
elif ".npy" in exts:
postfix = ".npy"
else:
raise RuntimeError("Folder must contain .wav or .npy files.")
postfix = ".npy" if FLAGS.npy_only else postfix
files = sorted([
os.path.join(source_path, fname) for fname in files
if fname.lower().endswith(postfix)
])
elif source_path.lower().endswith((".wav", ".npy")):
files = [source_path]
else:
files = []
# Now synthesize from files one batch at a time
batch_size = FLAGS.batch_size
sample_length = FLAGS.sample_length
n = len(files)
for start in range(0, n, batch_size):
end = start + batch_size
batch_files = files[start:end]
save_names = [
os.path.join(
save_path,
"gen_" + os.path.splitext(os.path.basename(f))[0] + ".wav")
for f in batch_files
]
print('loading batch..')
batch_data = fastgen.load_batch(batch_files,
sample_length=sample_length)
# Encode waveforms
encodings = batch_data if postfix == ".npy" else fastgen.encode(
batch_data, checkpoint_path, sample_length=sample_length)
if FLAGS.gpu_number != 0:
with tf.device("/device:GPU:%d" % FLAGS.gpu_number):
fastgen.synthesize(encodings,
save_names,
checkpoint_path=checkpoint_path)
else:
fastgen.synthesize(encodings,
save_names,
checkpoint_path=checkpoint_path)
def load_encoding(fname, sample_length=None, sr=16000, ckpt='model.ckpt-200000'):
'''sound loading'''
audio = utils.load_audio(fname, sample_length=sample_length, sr=sr)
sample_length = audio.shape[0]
print('{} samples, {} seconds'.format(sample_length, sample_length / float(sr)))
'''encoding'''
encoding = fastgen.encode(audio, ckpt, sample_length)
print("(batch_size, time_steps, dimensions) :",encoding.shape)
np.save(fname[fname.rfind('/') + 1:] + '.npy', encoding)
return audio, encoding
def nsynth_encode(wav_path):
# Checkpoint path
checkpoint_path = './wavenet-ckpt/model.ckpt-200000'
# Load Audio
sr = 16000
audio, _ = lr.load(wav_path, sr=sr)
# Encode
encoding = fastgen.encode(audio, checkpoint_path, audio.shape[0])
return encoding.reshape((-1, 16))
def wavenet_encode(audio):
# Load the model weights.
checkpoint_path = './wavenet-ckpt/model.ckpt-200000'
# Pass the audio through the first half of the autoencoder,
# to get a list of latent variables that describe the sound.
# Note that it would be quicker to pass a batch of audio
# to fastgen.
encoding = fastgen.encode(audio, checkpoint_path, len(audio))
# Reshape to a single sound.
return encoding.reshape([-1, 16])
def main(unused_argv=None):
os.environ["CUDA_VISIBLE_DEVICES"] = str(FLAGS.gpu_number)
source_path = utils.shell_path(FLAGS.source_path)
checkpoint_path = utils.shell_path(FLAGS.checkpoint_path)
save_path = utils.shell_path(FLAGS.save_path)
if not save_path:
raise RuntimeError("Must specify a save_path.")
tf.logging.set_verbosity(FLAGS.log)
# Generate from wav files
if tf.gfile.IsDirectory(source_path):
files = tf.gfile.ListDirectory(source_path)
exts = [os.path.splitext(f)[1] for f in files]
if ".wav" in exts:
postfix = ".wav"
elif ".npy" in exts:
postfix = ".npy"
else:
raise RuntimeError("Folder must contain .wav or .npy files.")
postfix = ".npy" if FLAGS.npy_only else postfix
files = sorted([
os.path.join(source_path, fname)
for fname in files
if fname.lower().endswith(postfix)
])
elif source_path.lower().endswith((".wav", ".npy")):
files = [source_path]
else:
files = []
# Now synthesize from files one batch at a time
batch_size = FLAGS.batch_size
sample_length = FLAGS.sample_length
n = len(files)
for start in range(0, n, batch_size):
end = start + batch_size
batch_files = files[start:end]
save_names = [
os.path.join(save_path,
"gen_" + os.path.splitext(os.path.basename(f))[0] + ".wav")
for f in batch_files
]
batch_data = fastgen.load_batch(batch_files, sample_length=sample_length)
# Encode waveforms
encodings = batch_data if postfix == ".npy" else fastgen.encode(
batch_data, checkpoint_path, sample_length=sample_length)
if FLAGS.gpu_number != 0:
with tf.device("/device:GPU:%d" % FLAGS.gpu_number):
fastgen.synthesize(
encodings, save_names, checkpoint_path=checkpoint_path)
else:
fastgen.synthesize(encodings, save_names, checkpoint_path=checkpoint_path)
def load_encoding(_file,
sample_length=None,
sample_rate=16000,
ckpt='model.ckpt-200000'):
'''
Resamples signal to <sample_rate> and truncates it to <sample_length> elements
Then encodes it through the model <ckpt>
Returns a tuple (signal, encoded_signal)
'''
audio = utils.load_audio(_file,
sample_length=sample_length,
sr=sample_rate)
encoding = fastgen.encode(audio, ckpt, sample_length)
return audio, encoding
def encode(path, filename):
print('encoding..')
sr = 16000
audio = utils.load_audio(path, sample_length=40000, sr=sr)
sample_length = audio.shape[0]
print('{} samples, {} seconds'.format(sample_length,
sample_length / float(sr)))
model_path = '/home/paperspace/data/wavenet-ckpt/model.ckpt-200000'
encoding = fastgen.encode(audio, model_path, sample_length)
print(encoding.shape)
print('finished encoding..')
# np.save(fname + '.npy', encoding)
decode(encoding, path, filename, sample_length, model_path)
def encode(paths: List[str],
sample_length: int = 16000,
sample_rate: int = 16000,
checkpoint: str = "checkpoints/wavenet-ckpt/model.ckpt-200000") \
-> np.ndarray:
audios = []
for path in paths:
audio = utils.load_audio(path,
sample_length=sample_length,
sr=sample_rate)
audios.append(audio)
audios = np.array(audios)
encodings = fastgen.encode(audios, checkpoint, sample_length)
return encodings
def load_encoding(fname,
sample_length=None,
sr=16000,
ckpt='model.ckpt-200000'):
'''sound loading'''
audio = utils.load_audio(fname, sample_length=sample_length, sr=sr)
sample_length = audio.shape[0]
print('{} samples, {} seconds'.format(sample_length,
sample_length / float(sr)))
'''encoding'''
encoding = fastgen.encode(audio, ckpt, sample_length)
print("(batch_size, time_steps, dimensions) :", encoding.shape)
np.save(fname[fname.rfind('/') + 1:] + '.npy', encoding)
return audio, encoding
def wavenet_encode(file_path):
# Load the model weights.
checkpoint_path = './wavenet-ckpt/model.ckpt-200000'
# Load and downsample the audio.
neural_sample_rate = 16000
audio = utils.load_audio(file_path,
sample_length=400000,
sr=neural_sample_rate)
# Pass the audio through the first half of the autoencoder,
# to get a list of latent variables that describe the sound.
# Note that it would be quicker to pass a batch of audio
# to fastgen.
encoding = fastgen.encode(audio, checkpoint_path, len(audio))
# Reshape to a single sound.
return encoding.reshape((-1, 16))
def encode():
# from https://www.freesound.org/people/MustardPlug/sounds/395058/
# fname = '395058__mustardplug__breakbeat-hiphop-a4-4bar-96bpm.wav'
fname = './wav/mehldau-1.wav'
sr = 44100
audio = utils.load_audio(fname, sample_length=44100, sr=sr)
sample_length = audio.shape[0]
print('{} samples, {} seconds'.format(sample_length,
sample_length / float(sr)))
encoding = fastgen.encode(audio, './wavenet-ckpt/model.ckpt-200000',
sample_length)
print(encoding.shape)
np.save(fname + '.npy', encoding)
fig, axs = plt.subplots(2, 1, figsize=(10, 5))
axs[0].plot(audio)
axs[0].set_title('Audio Signal')
axs[1].plot(encoding[0])
axs[1].set_title('NSynth Encoding')
return encoding
def Plot_SingleFile(file_name, sampleRate):
audio = utils.load_audio(
file_name,
sample_length=70000) #sample_length for how long will it be.
sample_length = audio.shape[0]
print('{} samples, {} seconds'.format(sample_length,
sample_length / float(sampleRate)))
#Encoding for new sound part.
encoding = fastgen.encode(audio, 'model.ckpt-200000', sample_length)
print(encoding.shape)
np.save(file_name + '.npy', encoding)
fig, axs = plt.subplots(2, 1, figsize=(10, 5))
axs[0].plot(audio)
axs[0].set_title('Audio Signal')
axs[1].plot(encoding[0])
axs[1].set_title('NSynth Encoding')
#synthesis
fastgen.synthesize(encoding,
save_paths=['gen_' + file_name],
samples_per_save=sample_length)
def main(unused_argv=None):
tf.logging.set_verbosity(FLAGS.log)
if FLAGS.checkpoint_path:
checkpoint_path = utils.shell_path(FLAGS.checkpoint_path)
else:
expdir = utils.shell_path(FLAGS.expdir)
tf.logging.info("Will load latest checkpoint from %s.", expdir)
while not tf.gfile.Exists(expdir):
tf.logging.fatal("\tExperiment save dir '%s' does not exist!", expdir)
sys.exit(1)
try:
checkpoint_path = tf.train.latest_checkpoint(expdir)
except tf.errors.NotFoundError:
tf.logging.fatal("There was a problem determining the latest checkpoint.")
sys.exit(1)
if not tf.train.checkpoint_exists(checkpoint_path):
tf.logging.fatal("Invalid checkpoint path: %s", checkpoint_path)
sys.exit(1)
tf.logging.info("Will restore from checkpoint: %s", checkpoint_path)
source_path = utils.shell_path(FLAGS.source_path)
tf.logging.info("Will load Wavs from %s." % source_path)
save_path = utils.shell_path(FLAGS.save_path)
tf.logging.info("Will save embeddings to %s." % save_path)
if not tf.gfile.Exists(save_path):
tf.logging.info("Creating save directory...")
tf.gfile.MakeDirs(save_path)
sample_length = FLAGS.sample_length
batch_size = FLAGS.batch_size
def is_wav(f):
return f.lower().endswith(".wav")
wavfiles = sorted([
os.path.join(source_path, fname)
for fname in tf.gfile.ListDirectory(source_path) if is_wav(fname)
])
for start_file in xrange(0, len(wavfiles), batch_size):
batch_number = (start_file / batch_size) + 1
tf.logging.info("On file number %s (batch %d).", start_file, batch_number)
end_file = start_file + batch_size
wavefiles_batch = wavfiles[start_file:end_file]
# Ensure that files has batch_size elements.
batch_filler = batch_size - len(wavefiles_batch)
wavefiles_batch.extend(batch_filler * [wavefiles_batch[-1]])
wav_data = np.array(
[utils.load_audio(f, sample_length) for f in wavefiles_batch])
try:
tf.reset_default_graph()
# Load up the model for encoding and find the encoding
encoding = encode(wav_data, checkpoint_path, sample_length=sample_length)
if encoding.ndim == 2:
encoding = np.expand_dims(encoding, 0)
tf.logging.info("Encoding:")
tf.logging.info(encoding.shape)
tf.logging.info("Sample length: %d" % sample_length)
for num, (wavfile, enc) in enumerate(zip(wavefiles_batch, encoding)):
filename = "%s_embeddings.npy" % wavfile.split("/")[-1].strip(".wav")
with tf.gfile.Open(os.path.join(save_path, filename), "w") as f:
np.save(f, enc)
if num + batch_filler + 1 == batch_size:
break
except Exception as e:
tf.logging.info("Unexpected error happened: %s.", e)
raise
def main(unused_argv=None):
os.environ["CUDA_VISIBLE_DEVICES"] = str(FLAGS.gpu_number)
source_path = utils.shell_path(FLAGS.source_path)
checkpoint_path = utils.shell_path(FLAGS.checkpoint_path)
save_path = utils.shell_path(FLAGS.save_path)
if not save_path:
raise ValueError("Must specify a save_path.")
tf.logging.set_verbosity(FLAGS.log)
# Use directory of files
if tf.gfile.IsDirectory(source_path):
files = tf.gfile.ListDirectory(source_path)
file_extensions = [os.path.splitext(f)[1] for f in files]
if ".wav" in file_extensions:
file_extension = ".wav"
elif ".npy" in file_extensions:
file_extension = ".npy"
else:
raise RuntimeError("Folder must contain .wav or .npy files.")
file_extension = ".npy" if FLAGS.npy_only else file_extension
files = sorted([
os.path.join(source_path, fname)
for fname in files
if fname.lower().endswith(file_extension)
])
# Use a single file
elif source_path.lower().endswith((".wav", ".npy")):
file_extension = os.path.splitext(source_path.lower())[1]
files = [source_path]
else:
raise ValueError(
"source_path {} must be a folder or file.".format(source_path))
# Now synthesize from files one batch at a time
batch_size = FLAGS.batch_size
sample_length = FLAGS.sample_length
n = len(files)
for start in range(0, n, batch_size):
end = start + batch_size
batch_files = files[start:end]
save_names = [
os.path.join(save_path,
"gen_" + os.path.splitext(os.path.basename(f))[0] + ".wav")
for f in batch_files
]
# Encode waveforms
if file_extension == ".wav":
batch_data = fastgen.load_batch_audio(
batch_files, sample_length=sample_length)
encodings = fastgen.encode(
batch_data, checkpoint_path, sample_length=sample_length)
# Or load encodings
else:
encodings = fastgen.load_batch_encodings(
batch_files, sample_length=sample_length)
# Synthesize multi-gpu
if FLAGS.gpu_number != 0:
with tf.device("/device:GPU:%d" % FLAGS.gpu_number):
fastgen.synthesize(
encodings, save_names, checkpoint_path=checkpoint_path)
# Single gpu
else:
fastgen.synthesize(
encodings, save_names, checkpoint_path=checkpoint_path)
import os
import numpy as np
from magenta.models.nsynth import utils
from magenta.models.nsynth.wavenet import fastgen
filename = '/data/input/battle1.wav'
sr = 44100
audio = utils.load_audio(filename, sample_length=(sr * 4), sr=sr)
sample_length = audio.shape[0]
print('{} samples, {} seconds'.format(sample_length,
sample_length / float(sr)))
encoding = fastgen.encode(audio, '/data/model/wavenet-ckpt/model.ckpt-200000',
sample_length)
print(encoding.shape)
np.save(filename.replace('.wav', '') + '_encoded.npy', encoding)
fastgen.synthesize(
encoding,
save_paths=['/data/output/test.wav'],
samples_per_save=sample_length,
checkpoint_path="/data/model/wavenet-ckpt/model.ckpt-200000")
def load_encoding(fname, sample_length=None, sr=16000, ckpt=MODEL):
audio = utils.load_audio(fname, sample_length=sample_length, sr=sr)
encoding = fastgen.encode(audio, ckpt, sample_length)
return audio, encoding
from magenta.models.nsynth import utils
from magenta.models.nsynth.wavenet import fastgen
from IPython.display import Audio
#%matplotlib inline
#%config InlineBackend.figure_format = 'jpg'
fname = "03 Plimsoll Punks.wav"
ckpt = "model.ckpt-200000"
sr = 16000
audio = utils.load_audio(fname, sample_length=16000, sr=sr)
sample_length = audio.shape[0]
print ("{} samples , {} seconds".format(sample_length, sample_length/float(sr)))
encoding = fastgen.encode(audio, ckpt, sample_length)
print(encoding.shape)
np.save(fname.split(".")[0] + ".npy", encoding)
fig, axs = plt.subplots(2, 1, figsize=(10, 5))
axs[0].plot(audio);
axs[0].set_title("Audio Signal")
axs[1].plot(encoding[0]);
axs[1].set_title("NSynth Encoding")
# Verify fast to generate encoding
fastgen.synthesize(encoding, save_paths=["gen_" + fname], samples_per_save=sample_length)
sr = 16000
def unused():
# from https://www.freesound.org/people/MustardPlug/sounds/395058/
fname = '395058__mustardplug__breakbeat-hiphop-a4-4bar-96bpm.wav'
sr = 16000
audio = utils.load_audio(fname, sample_length=40000, sr=sr)
sample_length = audio.shape[0]
print('{} samples, {} seconds'.format(sample_length, sample_length / float(sr)))
encoding = fastgen.encode(audio, 'model.ckpt-200000', sample_length)
print(encoding.shape)
np.save(fname + '.npy', encoding)
fig, axs = plt.subplots(2, 1, figsize=(10, 5))
axs[0].plot(audio);
axs[0].set_title('Audio Signal')
axs[1].plot(encoding[0]);
axs[1].set_title('NSynth Encoding')
fastgen.synthesize(encoding, save_paths=['gen_' + fname], samples_per_save=sample_length)
sr = 16000
synthesis = utils.load_audio('gen_' + fname, sample_length=sample_length, sr=sr)
def load_encoding(fname, sample_length=None, sr=16000, ckpt='model.ckpt-200000'):
audio = utils.load_audio(fname, sample_length=sample_length, sr=sr)
encoding = fastgen.encode(audio, ckpt, sample_length)
return audio, encoding
# from https://www.freesound.org/people/maurolupo/sounds/213259/
fname = '213259__maurolupo__girl-sings-laa.wav'
sample_length = 32000
audio, encoding = load_encoding(fname, sample_length)
fastgen.synthesize(
encoding,
save_paths=['gen_' + fname],
samples_per_save=sample_length)
synthesis = utils.load_audio('gen_' + fname,
sample_length=sample_length,
sr=sr)
# use image interpolation to stretch the encoding: (pip install scikit-image)
from skimage.transform import resize
def timestretch(encodings, factor):
min_encoding, max_encoding = encoding.min(), encoding.max()
encodings_norm = (encodings - min_encoding) / (max_encoding - min_encoding)
timestretches = []
for encoding_i in encodings_norm:
stretched = resize(encoding_i, (int(encoding_i.shape[0] * factor), encoding_i.shape[1]), mode='reflect')
stretched = (stretched * (max_encoding - min_encoding)) + min_encoding
timestretches.append(stretched)
return np.array(timestretches)
# from https://www.freesound.org/people/MustardPlug/sounds/395058/
fname = '395058__mustardplug__breakbeat-hiphop-a4-4bar-96bpm.wav'
sample_length = 40000
audio, encoding = load_encoding(fname, sample_length)
audio = utils.load_audio('gen_slower_' + fname, sample_length=None, sr=sr)
Audio(audio, rate=sr)
encoding_slower = timestretch(encoding, 1.5)
encoding_faster = timestretch(encoding, 0.5)
fig, axs = plt.subplots(3, 1, figsize=(10, 7), sharex=True, sharey=True)
axs[0].plot(encoding[0]);
axs[0].set_title('Encoding (Normal Speed)')
axs[1].plot(encoding_faster[0]);
axs[1].set_title('Encoding (Faster))')
axs[2].plot(encoding_slower[0]);
axs[2].set_title('Encoding (Slower)')
fastgen.synthesize(encoding_faster, save_paths=['gen_faster_' + fname])
fastgen.synthesize(encoding_slower, save_paths=['gen_slower_' + fname])
sample_length = 80000
# from https://www.freesound.org/people/MustardPlug/sounds/395058/
aud1, enc1 = load_encoding('395058__mustardplug__breakbeat-hiphop-a4-4bar-96bpm.wav', sample_length)
# from https://www.freesound.org/people/xserra/sounds/176098/
aud2, enc2 = load_encoding('176098__xserra__cello-cant-dels-ocells.wav', sample_length)
enc_mix = (enc1 + enc2) / 2.0
fig, axs = plt.subplots(3, 1, figsize=(10, 7))
axs[0].plot(enc1[0]);
axs[0].set_title('Encoding 1')
axs[1].plot(enc2[0]);
axs[1].set_title('Encoding 2')
axs[2].plot(enc_mix[0]);
axs[2].set_title('Average')
fastgen.synthesize(enc_mix, save_paths='mix.wav')
def fade(encoding, mode='in'):
length = encoding.shape[1]
fadein = (0.5 * (1.0 - np.cos(3.1415 * np.arange(length) /
float(length)))).reshape(1, -1, 1)
if mode == 'in':
return fadein * encoding
else:
return (1.0 - fadein) * encoding
fig, axs = plt.subplots(3, 1, figsize=(10, 7))
axs[0].plot(enc1[0]);
axs[0].set_title('Original Encoding')
axs[1].plot(fade(enc1, 'in')[0]);
axs[1].set_title('Fade In')
axs[2].plot(fade(enc1, 'out')[0]);
axs[2].set_title('Fade Out')
def crossfade(encoding1, encoding2):
return fade(encoding1, 'out') + fade(encoding2, 'in')
fig, axs = plt.subplots(3, 1, figsize=(10, 7))
axs[0].plot(enc1[0]);
axs[0].set_title('Encoding 1')
axs[1].plot(enc2[0]);
axs[1].set_title('Encoding 2')
axs[2].plot(crossfade(enc1, enc2)[0]);
axs[2].set_title('Crossfade')
fastgen.synthesize(crossfade(enc1, enc2), save_paths=['crossfade.wav'])
def load_encoding(fname, sample_length=None, sr=16000, ckpt='../wavenet-ckpt/model.ckpt-200000'):
audio = utils.load_audio(fname, sample_length=sample_length, sr=sr)
encoding = fastgen.encode(audio, ckpt, sample_length)
return audio, encoding
axs[1].plot(SecondSong_begin)
axs[1].set_title('Second Song')
st.pyplot()
# Save original snippets
output_dir = '/home/ubuntu/DeepBass/src/notebooks/'
output_name1 = 'originalend_' + FirstSong_fname + '.wav'
Save(output_dir, output_name1, FirstSong_end, sr)
output_name2 = 'originalbegin_' + SecondSong_fname + '.wav'
Save(output_dir, output_name2, SecondSong_begin, sr)
model_dir = '/home/ubuntu/DeepBass/src/notebooks/wavenet-ckpt/model.ckpt-200000'
# Create encodings
start = time.time()
enc1 = fastgen.encode(FirstSong_end, model_dir, sample_length)
enc2 = fastgen.encode(SecondSong_begin, model_dir, sample_length)
end = time.time()
st.write('Encoding took ' + str((end - start)) + ' seconds')
# Create cross fading in the latent space
fade_type = 'LinearFade'
xfade_encoding = crossfade(enc1, enc2, fade_type)
fig, axs = plt.subplots(3, 1, figsize=(10, 7))
axs[0].plot(enc1[0])
axs[0].set_title('Encoding 1')
axs[1].plot(enc2[0])
axs[1].set_title('Encoding 2')
axs[2].plot(xfade_encoding[0])
axs[2].set_title('Crossfade')
padding = SAMPLE_LENGTH - len(audio_list[i])
audio_list[i] = np.pad(audio_list[i], (0, padding), 'constant')
audio_list = np.array(audio_list)
except Exception as e:
print('Upload Cancelled')
print(e)
"""# Encode
Next we need to encode the audio. This should be relatively fast on a GPU, we will also create interpolations (the midpoints between each encoding) from which to re-synthesize audio.
"""
#@title Generate Encodings
audio = np.array(audio_list)
z = fastgen.encode(audio, ckpt_path, SAMPLE_LENGTH)
print('Encoded %d files' % z.shape[0])
# Start with reconstructions
z_list = [z_ for z_ in z]
name_list = ['recon_' + name_ for name_ in names]
# Add all the mean interpolations
n = len(names)
for i in range(n - 1):
for j in range(i + 1, n):
new_z = (z[i] + z[j]) / 2.0
new_name = 'interp_' + names[i] + '_X_'+ names[j]
z_list.append(new_z)
name_list.append(new_name)
def wavenet_encode(wave):
model_path = './wavenet-ckpt/wavenet-ckpt/model.ckpt-200000' #模型位置
# audio = np.load(file_path)
encoding = fastgen.encode(wave, model_path, len(wave))
print(encoding.reshape((-1, 16)).shape)
return encoding.reshape((-1, 16))
# Take the last four seconds
t_len = 1
silence_len = 7
x1 = x1[:silence_len * sr]
x1 = x1[-sr * t_len:]
sample_length = x1.shape[0]
output_dir = '/home/ubuntu/DeepBass/src/notebooks/'
output_name = 'original_' + filenames[0] + '.wav'
Save(output_dir, output_name, x1, sr)
model_dir = '/home/ubuntu/DeepBass/src/notebooks/wavenet-ckpt/model.ckpt-200000'
# Create encoding
start = time.time()
encoding = fastgen.encode(x1, model_dir, sample_length)
end = time.time()
st.write('Encoding took ' + str((end - start)) + ' seconds')
st.write('Encoding shape ' + str(encoding.shape))
# Save encoding
np.save(filenames[0] + '.npy', encoding)
# Plot PCM and encoding
fig, axs = plt.subplots(2, 1, figsize=(10, 5))
axs[0].plot(x1)
axs[0].set_title('Audio Signal')
axs[1].plot(encoding[0])
axs[1].set_title('NSynth Encoding')
st.pyplot()
def main(unused_argv=None):
os.environ["CUDA_VISIBLE_DEVICES"] = str(FLAGS.gpu_number)
source_path = utils.shell_path(FLAGS.source_path)
checkpoint_path = utils.shell_path(FLAGS.checkpoint_path)
save_path = utils.shell_path(FLAGS.save_path)
if not save_path:
raise ValueError("Must specify a save_path.")
tf.logging.set_verbosity(FLAGS.log)
# Use directory of files
if tf.gfile.IsDirectory(source_path):
files = tf.gfile.ListDirectory(source_path)
file_extensions = [os.path.splitext(f)[1] for f in files]
if ".wav" in file_extensions:
file_extension = ".wav"
elif ".npy" in file_extensions:
file_extension = ".npy"
else:
raise RuntimeError("Folder must contain .wav or .npy files.")
file_extension = ".npy" if FLAGS.npy_only else file_extension
files = sorted([
os.path.join(source_path, fname) for fname in files
if fname.lower().endswith(file_extension)
])
# Use a single file
elif source_path.lower().endswith((".wav", ".npy")):
file_extension = os.path.splitext(source_path.lower())[1]
files = [source_path]
else:
raise ValueError(
"source_path {} must be a folder or file.".format(source_path))
# Now synthesize from files one batch at a time
batch_size = FLAGS.batch_size
sample_length = FLAGS.sample_length
n = len(files)
for start in range(0, n, batch_size):
end = start + batch_size
batch_files = files[start:end]
save_names = [
os.path.join(
save_path,
"gen_" + os.path.splitext(os.path.basename(f))[0] + ".wav")
for f in batch_files
]
# Encode waveforms
if file_extension == ".wav":
batch_data = fastgen.load_batch_audio(batch_files,
sample_length=sample_length)
encodings = fastgen.encode(batch_data,
checkpoint_path,
sample_length=sample_length)
# Or load encodings
else:
encodings = fastgen.load_batch_encodings(
batch_files, sample_length=sample_length)
# Synthesize multi-gpu
if FLAGS.gpu_number != 0:
with tf.device("/device:GPU:%d" % FLAGS.gpu_number):
fastgen.synthesize(encodings,
save_names,
checkpoint_path=checkpoint_path)
# Single gpu
else:
fastgen.synthesize(encodings,
save_names,
checkpoint_path=checkpoint_path)
import os
import numpy as np
from magenta.models.nsynth import utils
from magenta.models.nsynth.wavenet import fastgen
fname = 'aggression.wav'
sr = 16000
audio = utils.load_audio(fname, sample_length=40000, sr=sr)
sample_length = audio.shape[0]
print('{} samples, {} seconds'.format(sample_length,
sample_length / float(sr)))
encoding = fastgen.encode(audio, os.path.abspath('model.ckpt-200000'),
sample_length)
print(encoding.shape)
np.save(fname + '.npy', encoding)
fastgen.synthesize(encoding,
save_paths=['gen_' + fname],
samples_per_save=sample_length)
sr = 16000
synthesis = utils.load_audio('gen_' + fname,
sample_length=sample_length,
sr=sr)
print('Magenta Test')
def encode(self):
self.encoding = fastgen.encode(self.audio, self.ckpt, self.sample_length)
def load_encoding(fname):
audio = utils.load_audio(fname, sample_length=sl, sr=sr)
print 'Encoding.. ', fname
encoding = fastgen.encode(audio, ckpt, sl)
print 'Encoded successfully'
return audio, encoding
