def reduce_noise(click, url, select_data):
if url is not None and select_data is not None:
fs, y = audio_io.read_wave_local(TEMP_STORAGE + url, as_float=True)
onsets = [point['customdata'] for point in select_data['points']]
noises = [
y[int(start_s * fs):int(end_s * fs)] for start_s, end_s in onsets
]
noises = np.concatenate(noises)
y = nr.reduce_noise(audio_clip=y, noise_clip=noises)
audio_io.save_wav(y, fs, path=TEMP_STORAGE + url)
return 1
def full_spectrogram_graph(select_data, url, selection, n_clicks, bandpass,
features, fig):
if url is not None:
if relayout_autosize_triggered():
raise PreventUpdate
temp_path = f'/tmp/{os.path.splitext(url)[0]}'
spectrum_path = temp_path + '_spectrum.npy'
time_path = temp_path + '_time.npy'
if select_data and event_triggered('embedding-graph.selectedData'):
start = fig['data'][0]['x'][0]
end = fig['data'][0]['x'][-1]
onsets = [point['customdata'] for point in select_data['points']]
shapes = visualize.shapes_from_onsets(onsets,
x_min=start,
x_max=end,
color='red')
fig['layout']['shapes'] = shapes
elif selection is not None and 'xaxis.range[0]' in selection and 'xaxis.range[1]' in selection:
start = selection['xaxis.range[0]']
end = selection['xaxis.range[1]']
Sxx = np.load(spectrum_path)
time = np.load(time_path)
fig = visualize.spectrogram_shaded(S=Sxx,
time=time,
fs=SAMPLING_RATE,
start_time=start,
end_time=end)
elif os.path.exists(spectrum_path
) and not event_triggered('apply-button.n_clicks'):
Sxx = np.load(spectrum_path)
time = np.load(time_path)
fig = visualize.spectrogram_shaded(S=Sxx,
time=time,
fs=SAMPLING_RATE)
else:
fs, y = audio_io.read_wave_local(TEMP_STORAGE + url)
lowcut, higcut = bandpass
y = filters.frequency_filter(y,
fs=SAMPLING_RATE,
lowcut=lowcut,
highcut=higcut)
freq, time, Sxx = visualize.calculate_spectrogram(y,
fs,
backend='yaafe')
np.save(spectrum_path, Sxx)
np.save(time_path, time)
fig = visualize.spectrogram_shaded(S=Sxx,
time=time,
fs=SAMPLING_RATE)
return fig
else:
raise PreventUpdate
def display_click_image(click_data, select_data, n_clicks, url, bandpass,
fft_size):
if url:
lowcut, higcut = bandpass
if click_data is not None and event_triggered(
'embedding-graph.clickData'):
start, end = click_data['points'][0]['customdata']
wav = audio_io.read_wav_part_from_local(path=TEMP_STORAGE + url,
start_s=start -
AUDIO_MARGIN,
end_s=end + AUDIO_MARGIN)
wav = filters.frequency_filter(wav,
fs=SAMPLING_RATE,
lowcut=lowcut,
highcut=higcut)
im = visualize.specgram_base64(y=wav,
fs=SAMPLING_RATE,
start=start,
end=end,
margin=AUDIO_MARGIN)
return html.Img(src='data:image/png;base64, ' + im,
style={
'height': '25vh',
'display': 'block',
'margin': 'auto'
})
else:
if select_data is not None:
onsets = [
point['customdata'] for point in select_data['points']
]
if onsets:
wavs = audio_io.read_wav_parts_from_local(
path=TEMP_STORAGE + url, onsets=onsets, as_float=True)
else:
raise PreventUpdate
else:
fs, wavs = audio_io.read_wave_local(TEMP_STORAGE + url,
as_float=True)
wavs = filters.frequency_filter(wavs,
fs=SAMPLING_RATE,
lowcut=lowcut,
highcut=higcut)
fig = visualize.power_spectrum(wavs,
fs=SAMPLING_RATE,
block_size=fft_size,
scaling='spectrum',
cutoff=-90)
return dcc.Graph(id='spectrum', figure=fig)
else:
raise PreventUpdate
def plot_embeddings(filename, n_clicks, embedding_type, fftsize, bandpass,
onset_threshold, sample_len, neighbours,
selected_features):
if filename is not None:
filepath = TEMP_STORAGE + filename
lowpass, highpass = bandpass
min_duration = sample_len - 0.05
fs, X = audio_io.read_wave_local(filepath, as_float=True)
features = get(X,
fs,
n_jobs=1,
selected_features=selected_features,
lowcut=lowpass,
highcut=highpass,
block_size=fftsize,
onset_detector_type='hfc',
onset_silence_threshold=-90,
onset_threshold=onset_threshold,
min_duration_s=min_duration,
sample_len=sample_len)
params = map_parameters(embedding_type, neighbours)
style = {
'display': 'inline-block',
'margin-left': 'auto',
'margin-right': '20px',
'float': 'right'
}
try:
embeddings, algo, msg = get_embeddings(
data=features.drop(columns=['onset', 'offset']),
type=embedding_type,
n_jobs=1,
**params)
# features.insert(0, column='filename', value=filenames[-1])
extra_data = ['onset', 'offset']
if 'freq_mean' in features:
mean_freq = features['freq_mean'].astype(int).astype(
str) + ' Hz<br>'
elif 'pitch_median' in features:
mean_freq = features['pitch_median'].astype(int).astype(
str) + ' Hz<br>'
else:
mean_freq = ''
interval = features['onset'].round(2).astype(
str) + ' - ' + features['offset'].round(2).astype(str) + 's'
text = mean_freq + interval
figure = visualize.scatter_plot(x=embeddings[:, 0],
y=embeddings[:, 1],
customdata=features[extra_data],
text=text)
if msg is None:
msg = f'Found {len(embeddings)} samples'
else:
style['color'] = 'red'
features.insert(0, 'id', features.index)
return figure, features.round(2).to_dict(orient='rows'), msg, style
except Exception as ex:
style['color'] = 'red'
return go.Figure(), None, str(ex), style
else:
raise PreventUpdate