cl_freq = 440
cl_pitch_label = "?"
cl_samp_rate = 44100
sine_samp_rate = 800
signal_params = {"duration": duration, "options": {"normalize": True}}
#
# Generate signals
#
# Clarinet signal using predefined amplitudes
cl_signal = sts.ClarinetApproxSignal(freq=cl_freq,
samp_rate=cl_samp_rate,
**signal_params)
# Superposition of two sines
sine_signal = sts.StationarySignal(sin_freqs=[25, 80],
sin_coeffs=[1, 0.5],
samp_rate=sine_samp_rate,
**signal_params)
# FFT of clarinet
cl_fft = fft(cl_signal.data)
cl_recon = ifft(cl_fft)
# Only consider positive frequencies (since the spectrum is symmetric)
cl_slice_num = cl_signal.samp_nums.shape[0] // 2
# Nyquist: max discernable freq is sampling rate / 2
short_right_sep_names = [n.split(" ")[0] for n in right_sep_names]
short_sep_names = short_left_sep_names + short_right_sep_names
# Generate signal
chords = [
m21.chord.Chord((soprano[i], alto[i], tenor[i], bass[i]))
for i in range(len(soprano))
]
# row = chord, column = melody
chord_freqs = np.array([[pitch.frequency for pitch in chord.pitches]
for chord in chords])
note_sigs = [[
sts.ClarinetApproxSignal(freq=freq,
duration=note_duration,
samp_rate=samp_rate) for freq in chord_freqs.T[i]
] for i in range(len(chord_freqs))]
melody_data = np.array([nst.NonStationarySignal(ns).data for ns in note_sigs])
left_signal = np.sum((left_intensities * melody_data.T).T, axis=0)
right_signal = np.sum(((1 - left_intensities) * melody_data.T).T, axis=0)
src = np.array((left_signal, right_signal))
_, left_recons, right_recons, extra = adress.adress(
left_signal=left_signal,
right_signal=right_signal,
samp_rate=samp_rate,
left_ds=left_ds,
left_Hs=left_Hs,
#
# Computation
#
# Format names
short_sep_names = [n.split(" ")[0] for n in sep_names]
# Generate signal
chords = [m21.chord.Chord((soprano[i], alto[i], tenor[i], bass[i])) for i in range(len(soprano))]
# row = chord, column = melody
chord_freqs = np.array([[pitch.frequency for pitch in chord.pitches] for chord in chords])
note_sigs = [[sts.ClarinetApproxSignal(freq=freq,
duration=note_duration,
samp_rate=samp_rate)
for freq in chord_freqs.T[i]]
for i in range(len(chord_freqs))]
melody_data = np.array([nst.NonStationarySignal(ns).data for ns in note_sigs])
left_signal = np.sum((left_intensities * melody_data.T).T, axis=0)
right_signal = np.sum(((1 - left_intensities) * melody_data.T).T, axis=0)
src = np.array((left_signal, right_signal))
# Compute the audio at each position
true_signals = {}
for li in set(left_intensities):
part_args = np.where(left_intensities == li) # Which parts have the same intensity?
true_signals[li] = np.sum(melody_data[part_args], axis=0)
signal_params = {
"duration": duration,
"samp_rate": samp_rate,
# "freq": cl_freq,
"options": {
"normalize": True
}
}
#
# Generate signals
#
# Generate clarinet signal using predefined amplitudes
cl_signal = sts.ClarinetApproxSignal(freq=cl_freq, **signal_params)
# Windowing the clarinet signal
w = window(cl_signal.samp_nums.shape[0], **window_params)
cl_window = cl_signal.data * w
# Generate clarinet Fourier Transform
cl_fft = fft(cl_signal.data)
cl_window_fft = fft(cl_window)
# Reconstructed signals from FFT
cl_ifft = ifft(cl_fft)
cl_window_ifft = ifft(cl_window_fft)
# Only consider positive frequencies (since the spectrum is symmetric)
cl_slice_num = cl_signal.samp_nums.shape[0] // 2
freqs = notes.major_scale(scale_name=scale_name,
n_notes=n_notes,
output="freq")
durations = [0.25] * n_notes
chop_ranges = [None] * n_notes
window = "blackmanharris"
#
# Generate data
#
# Generate signal
signals = [
sts.ClarinetApproxSignal(freq=freqs[i],
duration=durations[i],
samp_rate=samp_rate,
chop_range=chop_ranges[i])
for i in range(len(freqs))
]
nst = nsts.NonStationarySignal(signals)
# Compute FFT to demonstrate that FFT does not retain time information.
nst_fft = fft(nst.data)
nst_fft_x = fftfreq(nst.samp_nums.shape[0], 1 / nst.samp_rate)
# Only consider positive frequencies (since the spectrum is symmetric)
nst_slice_num = nst.samp_nums.shape[0] // 2
# Index of the max frequency present in the signal (less than some tolerance)
# cond = np.abs(nst_fft)[:nst_slice_num] / nst.samp_rate < freq_absence_tol
# fft_max_present_freq_idx = len(cond) - np.where(cond == False)[0][-1] - 1
duration = 0.01
samp_rate = 44100
cl_f = 466.16 / 2 # written pitch C4, actual pitch is Bb4
cl_pitch_label = "C4"
signal_params = {
"duration": duration,
"samp_rate": samp_rate,
"freq": cl_f,
"options": {
"normalize": True
}
}
# Generate clarinet signal using predefined amplitudes
cl_signal = pcs.ClarinetApproxSignal(**signal_params)
# Figure options
cl_fig = FigData(xs=cl_signal.samp_nums,
ys=[cl_signal.data],
line_options=[{"label": f"f0 = {cl_f}Hz"}],
xlabel="Time (seconds",
ylabel="Amplitude",
title=f"Synthesized clarinet (written pitch {cl_pitch_label})",
figsize=(5, 3)
)
# Plot signals
aplot.single_plot(cl_fig)