def gen_waveform_from_coords(coords):
'''
Takes x-y coordinates and genrates the waveform associated with them.
:param coords:
:return:
'''
params_normalised = decoder.predict([coords])
if params_normalised.shape[1] == 10:
params = denormalise_batch(params_normalised, y_norm, y_mean)
else:
dim4_params = denormalise_batch(params_normalised, x_batch_std,
x_batch_mean).numpy()
rounding_vals = np.array([0, 1, 2, 3])
diffs = abs(rounding_vals - dim4_params[0, 1])
closest_index = np.argmin(diffs)
dim4_params[0, 1] = rounding_vals[closest_index]
params = vec_to_freqpower(dim4_params)
waveform_length = 3200
# gen_sine = lambda x: 0.5*gen_sine_full(x, waveform_length,32000)
params_reshaped = np.reshape(params, (2, -1))
# complex_waveform = np.zeros((waveform_length,))
# for i in range(0,5):
# complex_waveform += gen_sine(params_reshaped[0,i])*params_reshaped[1,i]#*
# return complex_waveform
return gen_waveform_freq_and_power(params_reshaped, waveform_length)
def objective(**params):
params_normalised = decoder.predict([[params['X'], params['Y']]])
params_out = denormalise_batch(params_normalised, y_norm, y_mean)
outval = params_out[0][0].numpy()
if outval <= 0:
outval = 999999999
return outval
def get_synth_params(coords):
'''take [x,y] corrdinate and return synth parameters.'''
params_normalised = decoder.predict([coords])
dim4_params = denormalise_batch(params_normalised, x_batch_std,
x_batch_mean).numpy()
rounding_vals = np.array([0, 1, 2, 3])
diffs = abs(rounding_vals - dim4_params[0, 1])
closest_index = np.argmin(diffs)
dim4_params[0, 1] = rounding_vals[closest_index]
params = vec_to_freqpower(dim4_params).numpy()
return params
def objective_rms_to_peak_high_dimension_autoenc(**params):
fund_freq = params['fund_freq']
n_harmonics = params['n_harmonics']
detuning_hz = params['detuning_hz']
harmonics_power = params['harmonics_power']
params_1d = get_freqs_powers(
np.array([[
fund_freq, n_harmonics, detuning_hz, harmonics_power
]])) # get fundamental and powers of waveforms from the 4 params
params_1d = np.array(
[[fund_freq, n_harmonics, detuning_hz, harmonics_power]])
#normalise the parameters:
params_1d_normalised = (
params_1d - x_batch_mean) / x_batch_std #give this to the encoder
xy = encoder.predict(params_1d_normalised)
op_unnormalised = decoder.predict([xy[2]])
recovered_vals = denormalise_batch(op_unnormalised, x_batch_mean,
x_batch_std)
freq_powers_1d = get_freqs_powers(recovered_vals)
freq_powers_reshaped = np.reshape(freq_powers_1d, (2, -1))
waveform_length = 3200
waveform_out = gen_waveform_freq_and_power(freq_powers_reshaped,
waveform_length)
peak_to_rms = sound_metrics.peak_to_rms(waveform_out)
print(
"""The peak to RMS at fund_freq=%.6f n_harmonics=%d detuning=%d harmonic power=%.4f is %.6f"""
% (fund_freq, n_harmonics, detuning_hz, harmonics_power, peak_to_rms))
return -peak_to_rms
import tensorflow as tf
import numpy as np
from data_handling import denormalise_batch
y_mean = np.array([
1.5245759e+03, 2.2862642e+03, 2.2864233e+03, 1.5244318e+03, 0.0000000e+00,
1.0000000e+00, 3.7491098e-01, 2.5000253e-01, 1.2500882e-01, 0.0000000e+00
])
y_norm = np.array([
8.5161224e+02, 1.9803693e+03, 2.9146074e+03, 3.1424453e+03, 1.0000000e+00,
1.0000000e+00, 3.3071095e-01, 3.2275078e-01, 2.6020426e-01, 1.0000000e+00
])
decoder = tf.keras.models.load_model('../models/decoder')
prediction = decoder.predict([[-3, 3]])
params_out = denormalise_batch(prediction, y_norm, y_mean)