def draw_waveform(signals, sr=16000, titles=None):
"""draw waveform of signal"""
if not isinstance(signals, list):
signals = [signals]
if titles is not None:
if not isinstance(titles, list):
titles = [titles]
if len(signals) != len(titles):
titles = None
fig = plt.figure(figsize=(18, 9))
fig.subplots_adjust(left=0.2, right=0.95, hspace=0.3)
n = len(signals)
for i, sig in enumerate(signals):
ax = fig.add_subplot(n, 1, i + 1)
display.waveplot(sig, sr=sr, ax=ax)
if titles is not None:
ax.set_title(titles[i], x=-0.07, y=0.4, va='center', ha='right')
buf = io.BytesIO()
plt.savefig(buf, format='png')
buf.seek(0)
plt.close(fig)
arr = buf2ndarray(buf)
return arr
def draw_waveform_animated_faster(animation_path, signals, sr=16000, titles=None, fps=30):
"""draw waveform of signal"""
if not isinstance(signals, list):
signals = [signals]
if titles is not None:
if not isinstance(titles, list):
titles = [titles]
if len(signals) != len(titles):
titles = None
fig = plt.figure(figsize=(18, 9))
canvas_width, canvas_height = fig.canvas.get_width_height()
fig.subplots_adjust(left=0.2, right=0.95, hspace=0.3)
n = len(signals)
lines = []
for i, sig in enumerate(signals):
ax = fig.add_subplot(n, 1, i + 1)
display.waveplot(sig, sr=sr, ax=ax)
if titles is not None:
ax.set_title(titles[i], x=-0.07, y=0.4, va='center', ha='right')
# draw animated line
X_VALS = np.linspace(0, len(sig)/sr, num=int(len(sig)/sr*fps))
padding = 0
min_line = min(sig) - padding
max_line = max(sig) + padding
l, v = plt.plot(X_VALS[0], min_line, X_VALS[-1], max_line, linewidth=2, color='#ff0000')
lines.append(l)
def animate(num):
i = X_VALS[num]
for line in lines:
line.set_data([i, i], [-1, 1])
# Open an ffmpeg process
cmdstring = ('ffmpeg',
'-y', '-r', '%d' % fps, # overwrite, 1fps
'-s', '%dx%d' % (canvas_width, canvas_height), # size of image string
'-pix_fmt', 'argb', # format
'-f', 'rawvideo', '-i', '-', # tell ffmpeg to expect raw video from the pipe
# '-q:v', '5', # video quality when -vcodec=mpeg4
'-vcodec', 'h264', animation_path) # output encoding
p = subprocess.Popen(cmdstring, stdin=subprocess.PIPE)
# Draw frames and write to the pipe
for frame in range(len(X_VALS)):
# draw the frame
animate(frame)
fig.canvas.draw()
# extract the image as an ARGB string
string = fig.canvas.tostring_argb()
# write to pipe
p.stdin.write(string)
# Finish up
p.communicate()
def isolate_events(loadpath, min_silence_len=500, silence_thresh=-35):
'''
Function to isolate singular audio events from an input file.
args:
loadpath : path to audio file
min_silence_len : minimum time in ms to consider as silence
silence_thresh : minimum level in dB for detecting a sound
returns:
events : list of arrays representing each audio event
'''
# RAW AUDIO SHOULD BE MONO WAV AT 22050 Hz !!!!!!!!!!!!!!!!
raw_audio = AudioSegment.from_wav(loadpath)
# Present raw audio:
audio_array, sample_rate = lb.load(loadpath, sr=None)
plt.figure(figsize=(12, 4))
lbd.waveplot(audio_array, sr=sample_rate)
plt.title('Recorded Audio')
plt.show()
# Isolate events:
events = split_on_silence(
raw_audio,
# must be silent for at least half a second
min_silence_len=min_silence_len,
# consider it silent if quieter than
silence_thresh=silence_thresh)
return events
def plot_wave(filename, sr=16000, axis=False):
data, sr = librosa.load(filename, sr)
plt.figure()
axis = "on" if axis else "off"
plt.axis(axis)
display.waveplot(data, sr)
plt.show()
def plot_fig(y,save='',x_axis='time', max_points=50000.0, offset=0.0, color='#333333', alpha=1.0, show_filename=True, plot=True):
plt.figure()
waveplot(y=y,sr=44100,x_axis=x_axis,max_points=max_points,
offset=offset,color=color,alpha=alpha)
plt.show()
if save != '':
plt.savefig(save)
def plot_band_pass_filter(data_path, bpf_kernel_size=33):
plt.rcParams["font.size"] = axis_font_size
y, sr = librosa.load(data_path, sr=16000)
plt.figure(figsize=(24, 6))
plt.subplot(2, 5, 1)
display.waveplot(y, sr=sr, x_axis='none')
plt.title("raw waveform")
# raw spectrogram
plt.subplot(2, 5, 6)
display.specshow(librosa.amplitude_to_db(np.abs(librosa.stft(y)),
ref=np.max),
sr=sr,
y_axis='linear')
plt.colorbar(format="%+2.0f dB")
plt.title("Linear power spectrogram")
for low_hz, plot_idx in zip([0, 1000, 2000, 3000, 4000, 5000, 6000, 7000],
[2, 3, 4, 5, 7, 8, 9, 10]):
high_hz = low_hz + 1000
y_input = torch.tensor(y).unsqueeze(0).unsqueeze(0)
bpf = BandPassFilter(kernel_size=bpf_kernel_size,
stride=1,
padding=bpf_kernel_size // 2,
low_hz=low_hz,
high_hz=high_hz)
y_pass = bpf.forward(y_input).squeeze().numpy()
plt.subplot(2, 5, plot_idx)
display.specshow(librosa.amplitude_to_db(np.abs(librosa.stft(y_pass)),
ref=np.max),
sr=sr)
plt.colorbar(format="%+2.0f dB")
plt.title("frequency band:[{}, {}]".format(low_hz, high_hz))
plt.tight_layout()
plt.show()
def animate(t):
# Access our initialized variables from outside the function
global last_t
global tempgraph
# Round time to nearest 0.5 second. This limits the number of times we
# update the graph. Each graph update costs 10-20 seconds to draw, so by
# limiting this, we can render the graph an order of magnitude faster
temp_t = np.round(t * 2) / 2.0
# Only update the graph if we are at the next 0.5 seconds
if t > 0 and temp_t != last_t:
# Update our timekeeping variable
last_t = temp_t
# Delete the previous graph, otherwise we will keep plotting over
# the same graphs again and again
for coll in (ax.collections):
ax.collections.remove(coll)
# Load only the played portion of the mp3 and plot them
y2, sr2 = librosa.load('extracted_audio.mp3',
mono=False,
duration=t)
waveplot(y2, sr=sr2, color='b', alpha=0.8)
waveplot(y, sr=sr, color='b', alpha=0.25)
# Update the output graph
tempgraph = mplfig_to_npimage(fig)
return tempgraph
def compare_values(attribute, values, file):
"""
attributes (str): the names of the attribute you like to test.
values (list): list of values to compare.
file (str): file name to perform the tests.
"""
wavs = []
for idx, val in enumerate(values):
set_val_cmd = "AP.{}={}".format(attribute, val)
exec(set_val_cmd)
wav = AP.load_wav(file)
spec = AP.spectrogram(wav)
spec_norm = AP._denormalize(spec.T)
plt.subplot(len(values), 2, 2*idx + 1)
plt.imshow(spec_norm.T, aspect="auto", origin="lower")
# plt.colorbar()
plt.tight_layout()
wav_gen = AP.inv_spectrogram(spec)
wavs.append(wav_gen)
plt.subplot(len(values), 2, 2*idx + 2)
display.waveplot(wav, alpha=0.5)
display.waveplot(wav_gen, alpha=0.25)
plt.title("{}={}".format(attribute, val))
plt.tight_layout()
wav = AP.load_wav(file)
print(" > Ground-truth")
IPython.display.display(IPython.display.Audio(wav, rate=AP.sample_rate))
for idx, wav_gen in enumerate(wavs):
val = values[idx]
print(" > {} = {}".format(attribute, val))
IPython.display.display(IPython.display.Audio(wav_gen, rate=AP.sample_rate))
def plot_audio(filepath):
x, fs = load(filepath, sr = None, mono = True)
plt.figure(figsize=(16,4))
waveplot(x,sr=fs)
plt.title("Waveform for {}".format(filepath))
plt.tight_layout()
plt.show()
def plot_waves(sound_names, raw_sounds):
i = 1
for n, f in zip(sound_names, raw_sounds):
plt.subplot(6, 1, i)
libdisplay.waveplot(np.array(f), sr=22050)
plt.title(n.title() + ' - Waveplot')
i += 1
def waveplot(y: np.ndarray, sr: int, processed=None):
plot_wave = st.checkbox("Waveplot")
if plot_wave:
st.sidebar.markdown("#### Waveplot settings")
start_second = st.sidebar.number_input("start second",
min_value=0,
max_value=len(y) // sr,
value=0,
step=1,
key="waveplot_start")
end_second = st.sidebar.number_input("end second",
min_value=0,
max_value=len(y) // sr,
value=len(y) // sr,
step=1,
key="waveplot_end")
start_index = start_second * sr
if end_second == len(y) // sr:
end_index = len(y)
else:
end_index = end_second * sr
fig = plt.figure(figsize=(12, 4))
plt.grid(True)
display.waveplot(y[start_index:end_index], sr=sr, alpha=0.5)
if processed is not None:
display.waveplot(processed[start_index:end_index],
sr=sr,
alpha=0.5,
color="red")
st.pyplot(fig)
def main():
# y_or, sr = librosa.load("khoosoothunhus.wav", duration=10)
y_or, sr = librosa.load("A96.wav", duration=10)
# print(y_or)
fig1 = plt.figure()
fig1.suptitle('mouse hover over figure or axes to trigger events')
ax1 = fig1.add_subplot(411)
ax1.set_title("do thi am")
display.waveplot(y_or, sr=sr)
ax2 = fig1.add_subplot(412)
ax2.set_title('tu tuong quan')
fig2 = plt.figure()
ax3 = fig2.add_subplot(211)
ax3.set_title("bien thien f0 tu_tuong_quan")
ax4= fig1.add_subplot(413)
ax5 =fig2.add_subplot(212)
ax5.set_title("bien thien f0 amdf")
print("sample rate: ", sr)
ax_lpc_fft = fig1.add_subplot(414)
# fig3 = plt.figure()
# ax_lpc_fft = fig3.add_subplot(111)
cursor = SnaptoCursor(ax1, ax2, ax3, ax4, ax5, ax_lpc_fft, y_or, sr)
# plt.connect('motion_notify_event', cursor.mouse_move)
fig1.canvas.mpl_connect('button_press_event', cursor.mouse_click)
# plt.connect('button_press_event', cursor.mouse_click)
plt.show()
def show_audio_signal(ai, ctx, **kwargs):
if(ai.nchannels > 1):
_,axs = plt.subplots(ai.nchannels, 1, figsize=(6,4*ai.nchannels))
for i,channel in enumerate(ai.sig):
waveplot(channel.numpy(), ai.sr, ax=axs[i], **kwargs)
else:
axs = plt.subplots(ai.nchannels, 1)[1] if ctx is None else ctx
waveplot(ai.sig.squeeze(0).numpy(), ai.sr, ax=axs, **kwargs)
def update_waveform(self, audio_data, audio_sample_rate):
self.clear_axes()
if audio_data is not None:
libdisplay.waveplot(audio_data,
audio_sample_rate,
x_axis="time",
ax=self.axes)
self.update_canvas()
def audio_display():
sr = 1000
sound1, sound2, base = load_data(sr)
plt.subplot(2, 1, 1)
display.waveplot(sound1, sr=sr, alpha=0.25)
plt.subplot(2, 1, 2)
display.waveplot(base, sr=sr, alpha=0.25)
plt.tight_layout()
plt.show()
def plot_wav(path, y_hat, y_target, key, global_step, sample_rate):
plt.figure(figsize=(16, 9))
plt.subplot(2, 1, 1)
waveplot(y_target, sr=sample_rate)
plt.subplot(2, 1, 2)
waveplot(y_hat, sr=sample_rate)
plt.tight_layout()
plt.suptitle(f"record: {key}, global step: {global_step}")
plt.savefig(path, format="png")
plt.close()
def waveform(file_path: str) -> None:
"""
Draw audio waveform
Args:
file_path (str): audio file path
"""
data, sampling_rate = librosa.load(file_path)
plt.figure(figsize=(15, 5))
display.waveplot(data, sr=sampling_rate)
plt.show()
def plot_signal(self, title=None, out=None):
import matplotlib.pyplot as plt
if title is None:
title = self.filename
plt.title(title)
waveplot(self.signal, sr=self.samplerate)
if out is None:
plt.show()
else:
plt.savefig(out, dpi=300)
plt.close()
def display_features(audio_filepath):
x, sr, X, Xdb, mfccs = extract_features(audio_filepath)
plt.figure()
lbdisplay.waveplot(x, sr=sr)
plt.show()
plt.figure()
spectrogram = lbdisplay.specshow(Xdb, sr=sr, x_axis="time", y_axis="hz")
plt.colorbar()
plt.figure()
mfcc = lbdisplay.specshow(mfccs, sr=sr, x_axis="time")
plt.show()
def plot_amplitude(original_samples, shifted_samples, sampling_rate):
librosa_display.waveplot(original_samples, sr=sampling_rate, label='Original')
librosa_display.waveplot(shifted_samples, sr=sampling_rate, label='Shifted')
plt.title('Original vs Shifted samples')
plt.legend()
plt.ylabel('Amplitude')
plt.show()
return
def waves(raw_sounds):
i = 1
fig = plt.figure(figsize=(12, 15))
for file, name in zip(raw_sounds, sound_names):
plt.subplot(10, 1, i)
lsd.waveplot(np.array(file), x_axis=None)
plt.title(name.title())
i += 1
plt.suptitle("Waveplot", x=0.52, y=1.02, fontsize=15)
fig.tight_layout()
plt.savefig('waveplot.jpg')
def waveplot(y: np.ndarray,
sr: int,
processed=None,
tp: pd.DataFrame = None,
fp: pd.DataFrame = None):
plot_wave = st.checkbox("Waveplot")
if plot_wave:
st.sidebar.markdown("#### Waveplot settings")
start_second = st.sidebar.number_input("start second",
min_value=0,
max_value=len(y) // sr,
value=0,
step=1,
key="waveplot_start")
end_second = st.sidebar.number_input("end second",
min_value=0,
max_value=len(y) // sr,
value=len(y) // sr,
step=1,
key="waveplot_end")
start_index = start_second * sr
if end_second == len(y) // sr:
end_index = len(y)
else:
end_index = end_second * sr
fig = plt.figure(figsize=(12, 4))
plt.grid(True)
display.waveplot(y[start_index:end_index], sr=sr, alpha=0.5)
if processed is not None:
display.waveplot(processed[start_index:end_index],
sr=sr,
alpha=0.5,
color="red")
if tp is not None and len(tp) > 0:
for _, row in tp.iterrows():
plt.axvspan(row["t_min"],
row["t_max"],
color="g",
alpha=0.5,
label=str(row["species_id"]))
if fp is not None and len(fp) > 0:
for _, row in fp.iterrows():
plt.axvspan(row["t_min"],
row["t_max"],
color="r",
alpha=0.5,
label=str(row["species_id"]))
plt.legend()
st.pyplot(fig)
def show_audio_signal(ai, ctx, ax=None, title="", **kwargs):
ax = ifnone(ax, ctx)
if ax is None:
_, ax = plt.subplots()
ax.axis(False)
for i, channel in enumerate(ai):
# x_start, y_start, x_lenght, y_lenght, all in percent
ia = ax.inset_axes((i / ai.nchannels, 0.2, 1 / ai.nchannels, 0.7))
waveplot(channel.cpu().numpy(), ai.sr, ax=ia, **kwargs)
ia.set_title(f"Channel {i}")
ax.set_title(title)
return ax
def waveplot(sample, samplerate):
""" Create a waveplot figure of the raw sample
"""
plt.figure(figsize=(12, 6))
display.waveplot(y=sample, sr=samplerate)
plt.title('Waveplot of sound sample')
plt.xlabel('Time (sec)')
plt.ylabel('Amplitude')
plt.savefig(os.path.join(paths.path2Output, 'sample_waveplot.png'))
plt.close()
def show_wav(wav,
sr=8000,
figsize=(10, 4),
specgram_lib='librosa',
save_to=None,
n_fft=2048,
hop_length=None):
if hop_length is None:
hop_length = n_fft // 4
if type(wav) == str:
wav, sr = lr.load(wav)
elif type(wav) == torch.Tensor:
wav = wav.detach().numpy()
fig, axes = plt.subplots(nrows=1, ncols=2, figsize=figsize)
plt.sca(axes[0])
lrd.waveplot(wav, sr=sr, ax=axes[0])
plt.ylabel('Amplitude')
tticks, tlabels = plt.xticks()
plt.sca(axes[1])
if specgram_lib == 'librosa':
S_db = lr.amplitude_to_db(
np.abs(lr.stft(wav, n_fft=n_fft, hop_length=hop_length)))
img = lrd.specshow(S_db,
sr=sr,
ax=axes[1],
hop_length=hop_length,
x_axis='s',
y_axis='hz')
fig.colorbar(img, ax=axes[1])
elif specgram_lib == 'matplotlib':
plt.specgram(wav, Fs=sr, mode='magnitude')
plt.xlabel('Time')
plt.ylabel('Frequency')
else:
raise ValueError(
f'Invalid `specgram_lib={spegram_lib}`, should be one of (`librosa`, `matplotlib`)'
)
plt.xticks(tticks)
fig.tight_layout()
if save_to is not None:
plt.savefig(save_to, bbox_inches='tight')
plt.show()
ipd.display(ipd.Audio(wav, rate=sr))
def plot_waves(self, sound_names, raw_sounds):
i = self.i
fig = plt.figure(figsize=self.figsize, dpi=self.dpi)
for n, f in zip(sound_names, raw_sounds):
plt.subplot(10, 1, i)
display.waveplot(np.array(f), sr=22050)
plt.title(n.title())
i += 1
plt.suptitle("Figure 1: Waveplot",
x=self.x,
y=self.y,
fontsize=self.fontsize)
plt.tight_layout()
plt.show()
def plot_sound():
amplitudes, frequencies = librosa.load("microphone-results.wav")
plt.subplot(211)
display.waveplot(y=amplitudes, sr=frequencies)
# calcul de la transformee de Fourier et des frequences
fourier = fft(amplitudes)
# affichage de la transformee de Fourier
plt.subplot(212)
plt.plot(fourier)
plt.grid()
plt.xlabel("frequency domaine")
plt.ylabel("Magnitude")
plt.legend()
plt.show()
def draw_waveform_animated_better_quality(animation_path, signals, sr=16000, titles=None, fps=30):
"""draw waveform of signal"""
if not isinstance(signals, list):
signals = [signals]
if titles is not None:
if not isinstance(titles, list):
titles = [titles]
if len(signals) != len(titles):
titles = None
fig = plt.figure(figsize=(18, 9))
fig.subplots_adjust(left=0.2, right=0.95, hspace=0.3)
n = len(signals)
lines = []
for i, sig in enumerate(signals):
ax = fig.add_subplot(n, 1, i + 1)
display.waveplot(sig, sr=sr, ax=ax)
if titles is not None:
ax.set_title(titles[i], x=-0.07, y=0.4, va='center', ha='right')
# draw animated line
X_VALS = np.linspace(0, len(sig)/sr, num=int(len(sig)/sr*fps))
padding = 0
min_line = min(sig) - padding
max_line = max(sig) + padding
l, v = plt.plot(X_VALS[0], min_line, X_VALS[-1], max_line, linewidth=2, color='#ff0000')
lines.append(l)
def init():
for line in lines:
line.set_data([], [])
return lines
def animate(num):
i = X_VALS[num]
for line in lines:
line.set_data([i, i], [-1, 1])
return lines
line_anim = animation.FuncAnimation(fig, animate, frames=len(X_VALS), init_func=init, blit=True)
# save animated plot
# writer = FFMpegWriter(fps=fps)
writer = animation.writers['ffmpeg'](fps=fps)
print('Saving animated plot to', animation_path)
line_anim.save(animation_path, writer=writer)
print('Done')
def plot_audio(self, outside_series=None, outside_sr=None):
"""
Plot audio
:param outside_series:
:param outside_sr:
:return:
"""
y = self.select_series(outside_series)
sr = self.select_sr(outside_sr)
plt.figure(figsize=(14, 5))
waveplot(y, sr=sr)
plt.show()
def plot_waveform(temp, wav, channels):
y, sr = librosa.load(wav, mono=False)
if channels == 1:
y = y.reshape(-1, len(y))
DPI = 72
plt.figure(1, figsize=(16, 9), dpi=DPI)
plt.subplots_adjust(wspace=0, hspace=0)
for n in range(channels):
plt.subplot(2, 1, n + 1, facecolor='200')
display.waveplot(y[n], sr)
plt.grid(True, color='w')
waveform = str(Path(temp) / 'waveform.png')
plt.savefig(waveform, dpi=DPI)
waveform = str(base64.b64encode(open(waveform, 'rb').read()))[2: -1]
return waveform
def waveplot(path, y_hat, y_target, hparams):
sr = hparams.sample_rate
plt.figure(figsize=(12, 4))
if y_target is not None:
ax = plt.subplot(2, 1, 1)
dsp.waveplot(y_target, sr=sr)
ax.set_title('Target waveform')
ax = plt.subplot(2, 1, 2)
dsp.waveplot(y_hat, sr=sr)
ax.set_title('Prediction waveform')
else:
ax = plt.subplot(1, 1, 1)
dsp.waveplot(y_hat, sr=sr)
ax.set_title('Generated waveform')
plt.tight_layout()
plt.savefig(path, format="png")
plt.close()
