def ppr_to_audio(ppr,
save_dir,
sfpath=soundfont(),
tempo=120,
save_npy=False,
save_midi=True,
convert_mp3=True):
song_name = ppr.name
wave_file_path = os.path.join(save_dir, f"{song_name}.wav")
pm = ppr.to_pretty_midi(constant_tempo=tempo)
audio = pm_to_wave(pm, wave_file_path, sfpath)
print("wave file length:", len(audio))
print("wave file saved to", wave_file_path)
if save_npy:
npy_path = os.path.join(save_dir, f'{song_name}.npy')
np.save(npy_path, ppr)
print(f"{song_name}.npy saved!")
if save_midi:
midi_path = os.path.join(save_dir, f'{song_name}.midi')
ppr.write(midi_path)
print(f"{song_name}.midi file saved!")
if convert_mp3:
sound = AS.from_wav(wave_file_path)
mp3_file_path = f"{wave_file_path[:-4]}.mp3"
sound.export(mp3_file_path, format="mp3")
os.remove(wave_file_path)
print("The wave file is replaced to", mp3_file_path, '\n')
else:
return Audio(wave_file_path)
return Audio(mp3_file_path)
def get_sample(search_dir, fmt='mp3'):
sound_paths = glob.glob(os.path.join(search_dir, f"*.{fmt}"))
sound_paths.sort()
if len(sound_paths) > 1:
print(f"{len(sound_paths)} sounds found in {search_dir}")
for i, path in enumerate(sound_paths):
print(f"{i}: {path.split('/')[-1]}")
idx = input("input the number of sound(empty for the last):") or -1
sound_path = sound_paths[int(idx)]
elif sound_paths:
sound_path = sound_paths[0]
else:
print(f"no sound file found in {search_dir}")
return (None, None)
print(f"sound is loaded from {sound_path}")
song_name = sound_path.split('/')[-1].split('.')[0]
midi_path = os.path.join(search_dir, f'{song_name}.midi')
if os.path.exists(midi_path):
ppr = Multitrack(midi_path)
print(f"midi is loaded from {midi_path}")
return (Audio(sound_path), ppr)
else:
print(f"failed to load midi from {midi_path}")
return (Audio(sound_path), None)
def demo(self, track, artist=None, album=None):
producer_proba, top_songs, producer = self.query(track,
artist,
album,
use_spotify=True)
if type(top_songs) == type(None):
print('Song mp3 not available. :-( ')
return producer_proba, top_songs, producer
# get top song info
query_preview_url = self.query_preview_url
top_track = top_songs['track'].iloc[0]
top_artist = top_songs['artist'].iloc[0]
top_song_url = self.collection.find_one({'track':
top_track})['preview_url']
# process audio:
y_query, sr_query = load_mp3_from_url(query_preview_url)
y_top, sr_top = load_mp3_from_url(top_song_url)
if type(producer) == type(None):
print("True Producer: Not Found")
else:
print("True Producer: {}".format(producer))
print()
print("Producer Probabilities:")
print(producer_proba)
print()
print("Query Song: {} by {}".format(track, artist))
IPython.display.display(Audio(data=y_query, rate=sr_query))
print()
print("Most Similar Song: {} by {}".format(top_track, top_artist))
IPython.display.display(Audio(data=y_top, rate=sr_top))
return producer_proba, top_songs, producer
def show(self, ax=None, figsize=(5, 1), player=True, title=None, **kwargs):
if ax is None:
_, ax = plt.subplots(figsize=figsize)
if title:
ax.set_title("Class: " + str(title) + " \nfilename: " +
str(self.fn))
timesteps = np.arange(self.original_signal.shape[1]) / self.sample_rate
ax.plot(timesteps, self.original_signal[0])
if self.original_signal.size(0) > 1: # Check if mono or stereo
ax.plot(timesteps, self.original_signal[1])
ax.set_xlabel('Original Signal Time (s)')
plt.show()
timesteps = np.arange(
self.processed_signal.shape[1]) / self.sample_rate
_, ax = plt.subplots(figsize=figsize)
if title:
ax.set_title("Class: " + str(title) + " \nfilename: " +
str(self.fn))
ax.plot(timesteps, self.processed_signal[0])
if self.processed_signal.size(0) > 1: # Check if mono or stereo
ax.plot(timesteps, self.processed_signal[1])
ax.set_xlabel('Processed Signal Time (s)')
plt.show()
if player:
# unable to display an IPython 'Audio' player in plt axes
display("Original signal")
display(Audio(self.original_signal, rate=self.sample_rate))
display("Processed signal")
display(Audio(self.processed_signal, rate=self.sample_rate))
def audio_to_html(data_or_str, embed=False, max_audio_length=20):
html = ''
if not is_audio(data_or_str):
raise ValueError(f'Unknown audio format: {data_or_str}')
if embed:
if is_audio_path(data_or_str):
audio_data, sample_rate = load_audio(data_or_str,
return_sample_rate=True)
else:
assert is_audio_array(data_or_str)
audio_data = data_or_str
sample_rate = 16000 # assume sampling rate of 16kHz
if max_audio_length is not None and \
max_audio_length * sample_rate <= audio_data.shape[0]:
html = f'(original length was ' \
f'{int(audio_data.shape[0]/sample_rate)}s)'
audio_data = audio_data[:max_audio_length * sample_rate]
html += Audio(audio_data, rate=sample_rate)._repr_html_()
else:
assert is_audio_path(data_or_str)
if is_nist_sphere_file(data_or_str):
raise ValueError(
f'Audio at {data_or_str} is in nist/sphere format, '
'which the ipython Audio applet cannot play. '
'Use embed_audio=True instead.')
path, length = cache_audio_local(data_or_str, max_audio_length)
if path is None:
html = Templates.warning.format(
content=f'Audio too long to display ({length} seconds)')
else:
html = Audio(filename=str(path))._repr_html_()
return html
def play_chords(chords, tempo=160, amplitude=0.1, sample_rate=44100, filepath=None):
samples = amplitude * chords_to_samples(chords, tempo, sample_rate)
if filepath:
from scipy.io import wavfile
samples = (2**15 * samples).astype(np.int16)
wavfile.write(filepath, sample_rate, samples)
return display(Audio(filepath))
else:
return display(Audio(samples, rate=sample_rate))
def sound( x, rate=8000, label=''):
from IPython.display import display, Audio, HTML
if label is '':
display( Audio( x, rate=rate))
else:
display( HTML(
'<style> table, th, td {border: 0px; }</style> <table><tr><td>' + label +
'</td><td>' + Audio( x, rate=rate)._repr_html_()[3:] + '</td></tr></table>'
))
def play_audio(waveform, sample_rate):
waveform = waveform.numpy()
num_channels, num_frames = waveform.shape
if num_channels == 1:
display(Audio(waveform[0], rate=sample_rate))
elif num_channels == 2:
display(Audio((waveform[0], waveform[1]), rate=sample_rate))
else:
raise ValueError("Waveform with more than 2 channels are not supported.")
def play(filename,
player_id,
autoplay=True,
normalize=True,
start=None,
stop=None):
if start is not None and stop is not None:
raw_wav = raw_wav_segment(filename, start, stop)
update_display(Audio(data=raw_wav,
autoplay=autoplay,
normalize=normalize),
display_id=player_id)
else:
update_display(Audio(filename, autoplay=autoplay, normalize=normalize),
display_id=player_id)
def extractPeriod(data, rate, t_start, t_end):
t = np.arange(0, len(data)) / rate
plt.plot(t, data)
duration = t_end - t_start
plt.xlabel('$t$')
plt.ylabel('$x(t)$')
sample_start = int(t_start * rate)
sample_end = int(t_end * rate)
print(rate)
period = data[sample_start:sample_end]
audioSideBySide("Original", Audio(data=data, rate=rate),
"Extracted period",
Audio(np.tile(period, int(1 / duration)), rate=rate))
return period, rate
def play_drum_matrix(mat, tempo=120.0, threshold=0.0):
# generate audio
audio_data, mididata = get_audio_from_drum_matrix(mat,
tempo=tempo,
threshold=threshold)
display(Audio(audio_data, rate=44100))
return audio_data, mididata
def play(self):
left_channel = self.data[:, 0]
right_channel = self.data[:, 1]
display(
Audio([left_channel, right_channel], rate=self.rate,
autoplay=True))
def play(abc,
title='',
lyrics='',
tempo=TEMPO,
key=KEY,
soundfont_path='./GeneralUser GS v1.471.sf2'):
if not os.path.exists(soundfont_path):
raise FileNotFoundError(f'Could not find {soundfont_path}')
print(abc)
header = HEADER.format(title=title, tempo=str(tempo), key=key)
random_string = str(uuid4())
abc_filename = random_string + '.abc'
with open(abc_filename, 'w') as f:
f.write('\n'.join([header, abc, lyrics]))
svg_filename = 'Out001.svg'
subprocess.run(['abcm2ps', '-g', abc_filename])
display(SVG(svg_filename))
midi_filename = random_string + '.mid'
subprocess.run(['abc2midi', abc_filename, '-o', midi_filename])
wav_filename = random_string + '.wav'
subprocess.run([
'fluidsynth', '-i', '-F', wav_filename, soundfont_path, midi_filename
])
display(Audio(wav_filename))
os.remove(abc_filename)
os.remove(svg_filename)
os.remove(midi_filename)
os.remove(wav_filename)
def melspec_to_audio(self,
mel_spectrogram,
log=True,
phase=None,
transpose=True,
audio_out=True):
if transpose:
mel_spectrogram = mel_spectrogram.T
if log:
mel_spectrogram = librosa.db_to_power(mel_spectrogram)
mel_spectrogram = mel_spectrogram**0.5
magnitude = np.dot(np.linalg.pinv(self._MEL_FILTER), mel_spectrogram)
if phase is not None:
inverted_signal = librosa.istft(magnitude * phase,
hop_length=self._HOP_LENGTH)
else:
inverted_signal = griffin_lim(magnitude,
self._N_FFT,
self._HOP_LENGTH,
n_iterations=10)
if audio_out:
return Audio(inverted_signal, rate=self._SAMPLE_RATE)
else:
return inverted_signal
def beep(inp=1, duration=.1, n=1):
rate = 10000
mult = 1.6 * inp if inp else .08
wave = np.sin(mult*np.arange(rate*duration))
for i in range(n):
display(Audio(wave, rate=10000, autoplay=True))
time.sleep(duration / .1)
def generate_samples_for_spkr_list(
spkr_id_list,
npz='',
text='',
checkpoint='checkpoints/vctk-16khz-cmu-no-boundaries-all-noise-2/bestmodel.pth',
output_dir='./',
npz_path='/home/ubuntu/loop/data/vctk-16khz-cmu-no-boundaries-all/numpy_features'
):
out = []
for spkr in spkr_id_list:
output_file_override = 'gen_test_' + str(spkr)
loop_dict = generate_sample_with_loop(
spkr_id=spkr,
npz=npz,
text=text,
checkpoint=checkpoint,
output_dir='./',
npz_path=
'/home/ubuntu/loop/data/vctk-16khz-cmu-no-boundaries-all/numpy_features',
output_file_override=output_file_override)
out.append(loop_dict)
IPython.display.display(
Audio(loop_dict['output_file'] + '.wav', autoplay=True))
return out
def play_sound(self):
if self.fs >= 50000:
raise ValueError('frequency to high')
if self.domain == 'frequency':
raise DomainError('cannot play frequency domain signals')
#sd.play(self.__y_val, self.fs)
Audio(data=self.__y_val, rate=self.fs, autoplay=True)
def play(self, seq, show_player=True, instrument=None):
"""
Play sequence of chords
:param seq: a string containing a space-separated sequence of chords
:param show_player: whether to show the player or not
:param instrument: user-supplied sf2 sound font (do not set it to use default)
:return: None
"""
if instrument is None:
# Use default sound font
instrument = get_path_of_data_file("piano_soundfont.sf2")
seq = str(seq)
audio = []
for chord in seq.split(" "):
audio = np.append(audio, self._make_chord(chord, instrument))
if show_player:
display(Audio(audio, rate=self._rate, autoplay=False))
return audio
def shift_silent_right(sample_rate, samples):
samples[(samples > 200) | (samples < -200)]
sampling = samples[(samples > 200) | (samples < -200)]
shifted_silent = sampling.tolist() + np.zeros(
(samples.shape[0] - sampling.shape[0])).tolist()
Audio(shifted_silent, rate=sample_rate)
return sample_rate, shifted_silent
def show_sample(melsg,
file_id=None,
label="",
offset=0,
data_dir='data',
load_clip=False):
fig = plt.figure(figsize=(7, 5))
if file_id or label != "":
fig.suptitle(' '.join([("XC%s" % file_id) if file_id else "", label]))
gs = GridSpec(4, 1, fig, hspace=.1, wspace=0, top=.93)
melsg_ax = fig.add_subplot(gs[0:3])
specshow(melsg.squeeze(), y_axis='mel', x_axis='s', ax=melsg_ax)
plt.colorbar(melsg_ax.collections[0], ax=melsg_ax, pad=.01)
#mfcc_ax = fig.add_subplot(gs[3])
#specshow(mfcc.squeeze(), ax=mfcc_ax, x_axis='s')
#mfcc_ax.set_ylabel("MFCC")
#mfcc_ax.set_yticks([0,5,10,15])
# TODO: Ensure 22050 is correct frame rate
#mfcc_ax.set_xticklabels(["%0.1f"%(t+offset/(22050/512))
# for t in mfcc_ax.get_xticks()])
#plt.colorbar(mfcc_ax.collections[0], ax=mfcc_ax, aspect=7, pad=.01)
plt.show()
if file_id and load_clip:
file_path = os.path.join(data_dir, 'audio', "XC%s.mp3" % file_id)
print(file_path)
import warnings
warnings.simplefilter('ignore')
data, samplerate = librosa.load(file_path)
display(Audio(data, rate=samplerate))
def mystery_audio_demo2_play(signal_name='mystery1',
centre=200, width=20):
fmin = centre - 0.5 * width
fmax = centre + 0.5 * width
if fmin < 0:
fmin = 0
fs, x = scipy.io.wavfile.read('data/%s.wav' % signal_name)
try:
x = x[:, 0]
except:
pass
N = len(x)
# Round to power of 2 for FFT efficiency
Nz = 2 << (N - 1).bit_length()
f = np.arange(Nz // 2 + 1) / Nz * fs / 2
H = (f > fmin) & (f < fmax)
X = np.fft.rfft(x, Nz)
Y = X * H
y = np.fft.irfft(Y)[0:N]
y = y / y.max()
t = np.arange(N) / fs
signal_plot_with_dft(t, x, f, X, lollipop=False, mode='magnitude')
display(Audio(y, rate=fs))
def load_audio_frame_raw(filename, frame, context):
audio, sr = librosa.load(filename,
sr=None,
mono=False,
offset=frame - context,
duration=1 + context)
return Audio(audio, rate=sr, normalize=False)
def notebook_audio(audio):
"""Display IPython Audio object in the notebook
:param audio: path to the audio file
"""
if audio:
display(Audio(filename=audio))
def show_audio(a: Tuple[int, np.ndarray])->None: # a: (sample_rate, audio_array)
fig, ax = plt.subplots()
time_axis = np.linspace(start=0, stop=(len(a[1])/a[0]),num=np.round(len(a[1])))
ax.plot(time_axis, a[1])
ax.set_xlabel('Time (seconds)')
ax.set_ylabel('Amplitude')
display(Audio(a[1], rate=a[0]))
def automate_trade(self):
if (self.automate == True):
for indx in self.proposals.index :
if(indx not in self.transactions.index):
if(self.alert == True): display(Audio(numpy.sin(numpy.linspace(0, 9000, 10000)), rate=200000, autoplay=True));
self.ws.send(json.dumps({"buy": indx, "price": self.proposals.loc[indx]['amount']}))
return
def show_audio(wav,
sample_rate=DEFAULT_SAMPLE_RATE,
focus_points=[0.3],
focus_windows=[1000]):
"""Show all kind of things about this wav
"""
try:
wav = wav.numpy()
except:
assert isinstance(wav, np.ndarray)
wav = wav.squeeze()
display(Audio(data=wav, rate=sample_rate))
specplot(wav)
_, axes = plt.subplots(nrows=1 + len(focus_points),
ncols=1,
figsize=(15, 2 + 2 * len(focus_points)),
sharey=True)
t = np.linspace(0, len(wav) // sample_rate, len(wav))
axes[0].plot(t, wav)
for i, (focus_point,
focus_window) in enumerate(zip(focus_points, focus_windows)):
c = int(len(wav) * focus_point)
idx = list(range(c - focus_window // 2, c + focus_window // 2))
t_focused = t[idx]
wav_focused = wav[idx]
axes[1 + i].plot(t_focused, wav_focused)
axes[1 +
i].set_title(f"Focused on t={focus_point*len(wav)/sample_rate}")
axes[-1].set_xlabel("Time/s")
plt.tight_layout()
plt.show()
def plot_td_and_fd(t, hprime, f, htildeprime, h=None, htilde=None):
from IPython.display import display, clear_output, Audio
print('Contrast: {0:.4f}'.format(np.max(np.abs(hprime)) / np.sqrt(np.mean(np.abs(hprime)**2))))
sampling_rate = 1.0/(t[1]-t[0])
plt.close('all')
fig, (ax1, ax2) = plt.subplots(1, 2)
if h is not None:
ax1.plot(t, h, label='Raw data')
ax1.plot(t, hprime, label='Filtered data')
ax1.legend(loc='lower left');
else:
ax1.plot(t, hprime)
ax1.set_xlabel('Time (seconds)')
ax1.set_ylabel('Detector strain $h$ (dimensionless)')
ax1.set_xlim(xmax=t[-1])
ax1.set_ylim(1.1*np.min(hprime), 1.1*np.max(hprime))
ax1.set_title('Time domain')
ax1.grid()
if htilde is not None:
ax2.loglog(f, abs(htilde), label='Raw data')
ax2.loglog(f, abs(htildeprime), label='Filtered data')
ax2.legend(loc='lower left');
else:
ax2.loglog(f, abs(htildeprime))
ax2.set_xlabel('Frequency (Hz)')
ax2.set_ylabel(r'Detector strain Fourier transform $\tilde{h}$ (seconds)')
ax2.set_xlim(1, sampling_rate/2)
ax2.set_title('Frequency domain')
ax2.grid()
fig.tight_layout()
display(Audio(data=hprime, rate=int(sampling_rate), autoplay=False))
return fig, (ax1, ax2)
def iir_notch_play(fnotch=50, alpha=0.5, bode=True):
N = 40000
fs = 20e3
t = np.arange(N) / fs
hum = 0.5 * np.cos(2 * np.pi * 50 * t)
s = 0.2 * np.cos(2 * np.pi * 220 * t)
x = hum + s
omega0 = 2 * np.pi * fnotch / fs
beta = np.cos(omega0)
K = 0.5 * (1 + alpha)
b = (K, -2 * K * beta, K)
a = (1, -2 * K * beta, alpha)
f = np.logspace(0, 4, 200)
filter_plot(b, a, fs, f=f, bode=bode)
y = signal.lfilter(b=b, a=a, x=x)
signal_plot(t[0:2000:10], y[0:2000:10])
show()
display(Audio(y, rate=fs))
def process_account(self, message):
try:
sl = message['authorize']
if ('account_list' in sl): sl.pop('account_list')
if ('scopes' in sl): sl.pop('scopes')
if ('upgradeable_landing_companies' in sl):
sl.pop('upgradeable_landing_companies')
self.account = pandas.DataFrame(columns=[0])
self.transaction_stream()
for key in sl:
self.account.loc[key] = {0: str(sl[key])}
#for sym in self.forex_major:
#self.subscribe(sym)
for sym in self.volatility_indices:
self.subscribe(sym)
#tr = threading.Timer(1, self.ping).start()
threading.Thread(target=self.analysis).start()
threading.Thread(target=self.ping).start()
display(
Audio(numpy.sin(numpy.linspace(0, 4 * 2 * numpy.pi, 25)),
rate=20000,
autoplay=True))
except:
pass
return
def generate(self, parm_var, do_postfilter=True):
config = self.analysis_config
for path in self.paths:
file_id = splitext(basename(path))[0]
print('Synthesizing %s ... ' % (file_id), end='')
mgc, lf0, vuv, bap = self._generate_parameters(path, parm_var)
if do_postfilter:
mgc = merlin_post_filter(mgc, config.alpha)
sp = pysptk.mc2sp(mgc,
fftlen=config.fft_length,
alpha=config.alpha)
ap = pyworld.decode_aperiodicity(bap.astype(np.float64),
config.sampling_rate,
config.fft_length)
f0 = self._lf0_to_f0(lf0, vuv)
generated = pyworld.synthesize(f0.flatten().astype(np.float64),
sp.astype(np.float64),
ap.astype(np.float64),
config.sampling_rate,
config.frame_period)
with open(join(self.out_dir, file_id + '.wav'), 'wb') as f:
f.write(Audio(generated, rate=config.sampling_rate).data)
print('done!')
