if __name__ == "__main__":
if len(sys.argv) < 5:
print(
"Usage: python3 gen_tone.py <type> <frequency> <volume> <duration> <samp_rate>"
)
print("type:\t\tsine, square, sawtooth;")
print("frequency:\tnumber;")
print("volume:\t\tnumber from 0 to 100;")
print("duration:\tnumber in seconds;")
print("samp_rate:\tsampling rate in Hertz.")
sys.exit()
if sys.argv[0].lower() not in func_map.keys():
print("type must be sine, square, or sawtooth")
# Give arguments names
signal_func = func_map[sys.argv[0].lower()]
freq = sys.argv[1]
vol = sys.argv[2]
duration = sys.argv[3]
samp_rate = sys.argv[4]
# Generate signal
sig = signal_func(freq=freq, samp_rate=samp_rate, duration=duration)
data = sig.data
wav.write(f"audio/{sig}", data, samp_rate=samp_rate, amp_perc=vol / 100)
print(f"File written to 'audio/{sig}.wav'")
#
# Computation
#
if not os.path.exists(audio_out_folder):
os.mkdir(audio_out_folder)
# Load stereo audios
source_root_abs = os.path.abspath(source_root)
source_file_dict = af.id_filename_dict(source_root_abs,
ids=ids_to_test,
regexes=[f"{i}-stereo.wav" for i in ids_to_test],
sort_function=lambda f: f)
# Load truth files
for idx in ids_to_test:
print("working on", idx)
# Load stereo recording
stereo_signal = librosa.load(f"{source_root}/{source_file_dict[idx][0]}", sr=samp_rate, mono=False)[0]
# Decompose with ICA
transformer = FastICA(n_components=2,
random_state=0)
est_signals = transformer.fit_transform(stereo_signal.T).T
# Write to file
for i in range(est_signals.shape[0]):
awav.write(f"ica_audio/{idx}-ica-{i}", est_signals[i], samp_rate=samp_rate)
recon_fig = FigData(xs=trec,
ys=xrec,
title="Reconstructed after filtering:\n"
f"{scale_name} minor triads w/ inversions",
plot_type="plot",
xlabel="Time (s)",
ylabel="Amplitude")
#
# Display results
#
aplot.single_subplots(
grid_size=(2, 2),
fig_data={
(0, 1): stft_fig,
(1, 1): stftf_fig,
(0, 0): signal_fig,
(1, 0): recon_fig
},
individual_figsize=(6, 4),
savefig_path=
f"ISTFT_{window_name}_{window_length}_{int(overlap_percent*100)}%_Pure_{scale_name}_minor_chord_sampr={samp_rate}"
)
wav.write(
f"audio/ISTFT_{window_name}_{window_length}_{int(overlap_percent * 100)}%_Pure_{scale_name}m",
np.concatenate((nst.data, xrec)),
samp_rate=samp_rate)
f"Separated source '{sep_names[i]}'\nat d={ds[i]} with width H={Hs[i]}",
yscale="linear",
ylim=ylim)
fig_data[(2, i + 1)] = recon_stft_figs[i]
# Plot null/peaks, src & source STFTs
aplot.single_subplots(
grid_size=(3, 4),
fig_data=fig_data,
individual_figsize=(6, 4),
# savefig_path=f"ADRess_{channel}_{tech_name}.png"
)
#
# Output audio
#
wav.write(f"audio/{name}_stereo", src.T, samp_rate=samp_rate)
wav.write(f"audio/{name}_left", left_signal, samp_rate=samp_rate)
wav.write(f"audio/{name}_right", right_signal, samp_rate=samp_rate)
for i, left_recon in enumerate(left_recons):
wav.write(f"audio/{tech_name}_{short_sep_names[i]}",
left_recon,
samp_rate=samp_rate)
for i, right_recon in enumerate(right_recons):
wav.write(f"audio/{tech_name}_{short_sep_names[i + len(left_recons)]}",
right_recon,
samp_rate=samp_rate)
# Out of range or index not in whitelist, move on
if n == -1 or idx not in whitelist:
continue
# Load the mono file
mono = librosa.load(f"{dir_name}/{mono_filename}", sr=samp_rate,
mono=True)[0]
# Decompose the spectrogram with NMF
S = np.abs(librosa.stft(mono))**2
comps, acts = librosa.decompose.decompose(S, n_components=n)
# Reconstruct using each basis
for i in range(n):
ind_comp = np.zeros(comps.shape)
ind_comp[i, :] = comps[i, :]
ind_act = np.zeros(acts.shape)
ind_act[:, i] = ind_act[:, i]
recon_spectrogram = comps @ acts
recon = librosa.istft(recon_spectrogram)
# Write file
wav.write(
f"nmf_audio/{idx}-{i}.wav",
recon,
samp_rate=samp_rate,
# auto_timestamp=True
)
# Plot the frequency-azimuth spectrogram at the specified time frame
aplot.single_subplots(grid_size=(2, 2),
fig_data={(0, 0): l_azi_null_fig,
(1, 0): l_azi_peak_fig,
(0, 1): l_stft_src_fig,
(1, 1): l_stft_rec_fig,
},
individual_figsize=(6, 4),
savefig_path=f"ADRess_left_{tech_name}.png")
aplot.single_subplots(grid_size=(2, 2),
fig_data={(0, 0): r_azi_null_fig,
(1, 0): r_azi_peak_fig,
(0, 1): r_stft_src_fig,
(1, 1): r_stft_rec_fig,
},
individual_figsize=(6, 4),
savefig_path=f"ADRess_right_{tech_name}.png")
# Output audio file
for i, chord in enumerate(chord_names):
wav.write(f"audio/true_{chord}M", true_sigs[i].data, samp_rate=samp_rate)
wav.write(f"audio/{name}_stereo", src.T, samp_rate=samp_rate)
wav.write(f"audio/{name}_left", lsig.data, samp_rate=samp_rate)
wav.write(f"audio/{name}_right", rsig.data, samp_rate=samp_rate)
wav.write(f"audio/{est_method}_{name}_{window}_{nperseg}_{noverlap}_sep1", left_recon, samp_rate=samp_rate)
wav.write(f"audio/{est_method}_{name}_{window}_{nperseg}_{noverlap}_sep2", right_recon, samp_rate=samp_rate)
#
# Generate data
#
signals = [
ss.SineSignal(freq=freqs[i],
duration=durations[i],
chop_range=chop_ranges[i]) for i in range(len(freqs))
]
nst = nsts.NonStationarySignal(signals)
#
# Encapsulate data in FigData
#
fig = FigData(xs=nst.samp_nums,
ys=nst.data,
title="Non stationary signal",
plot_type="plot",
xlabel="Time (s)",
ylabel="Amplitude")
#
# Plot
#
aplot.single_plot(fig)
#
# Output audio
#
wav.write("audio/nonst8", nst, dtype=np.uint8)
aplot.single_subplots(grid_size=(4, 2),
fig_data={(0, 0): fig_data[(3, 0)],
(0, 1): fig_data[(4, 0)],
(1, 0): fig_data[(3, 1)],
(1, 1): fig_data[(4, 1)],
(2, 0): fig_data[(3, 2)],
(2, 1): fig_data[(4, 2)]},
individual_figsize=(4.8, 3),
savefig_path=f"ADRess_{tech_name}_RECON.png",
show=False
)
#
# Output audio
#
wav.write(f"audio/{name}_stereo", src.T, samp_rate=samp_rate)
wav.write(f"audio/{name}_left", left_signal, samp_rate=samp_rate)
wav.write(f"audio/{name}_right", right_signal, samp_rate=samp_rate)
# True audio at each stereo position
for pos, audio in true_signals.items():
wav.write(f"audio/{name}_pos={pos}", audio, samp_rate=samp_rate)
# Reconstructed audio
for i, recon in enumerate(recons):
wav.write(f"audio/{tech_name}_{short_sep_names[i]}", recon, samp_rate=samp_rate,
auto_timestamp=True
)
_, recons, extra = adress.adress_stereo(left_signal=left_signal,
right_signal=right_signal,
samp_rate=samp_rate,
ds=ds,
Hs=Hs,
beta=beta,
window=window,
nperseg=nperseg,
noverlap=noverlap,
# print_options=["progress"],
more_outputs=["stfts", "stft_f", "stft_t"])
# Write audio:
# Pan-mixed audio
if write_pan_mix:
wav.write(f"audio/{idx}-stereo", src.T, samp_rate=samp_rate)
# Reconstructed audio
for i, recon in enumerate(recons):
wav.write(f"audio/{idx}-d={ds[i]}-H={Hs[i]}", recon, samp_rate=samp_rate,
# auto_timestamp=True
)
# Grab the other specified quantities
t = extra["stft_t"]
f = extra["stft_f"]
left_stft = extra["stfts"]["left"]
right_stft = extra["stfts"]["right"]
# Compute the null/peak spectrogram for each specified tau
nulls = []