def get_diffuse_noise_psd(sensor_distances,
fft_size=1024,
sample_rate=16000,
sound_velocity=343):
"""This function does not calculate a PSD matrix.
Calculates a spatial coherence matrix for a spherically isotropic sound
field. This can be interpreted as a normalized PSD matrix.
The implementation is according to Bitzer, Joerg, and K. Uwe Simmer.
"Superdirective microphone arrays." In Microphone arrays, pp. 19-38.
Springer, Berlin, Heidelberg, 2001. In particular Equation (2.17).
Args:
sensor_distances: Shape (num_channels, num_channels) in meters.
fft_size:
sample_rate:
sound_velocity: Scalar in meters/ second.
Returns: spatial coherence matrix
with shape (num_frequency_bins, num_channels, num_channels)
"""
f = get_stft_center_frequencies(size=fft_size, sample_rate=sample_rate)
return np.sinc(2.0 * f[:, None, None] * sensor_distances[None, :, :] /
sound_velocity)
def get_steering_vector(
time_difference_of_arrival,
stft_size=1024,
sample_rate=16000,
normalize=False
):
"""Calculates analytic steering vector from TDoA.
Can be used when some oracle information is available. Often, DoA
estimation algorithms rely on steering vectors.
:param time_difference_of_arrival:
:param stft_size:
:param sample_rate:
:param normalize:
:return:
"""
center_frequencies = get_stft_center_frequencies(stft_size, sample_rate)
steering_vector = numpy.exp(
-2j * numpy.pi *
center_frequencies *
time_difference_of_arrival[..., numpy.newaxis]
)
if normalize:
steering_vector /= np.linalg.norm(
steering_vector, axis=-2, keepdims=True
)
return steering_vector
def get_steering_vector(time_difference_of_arrival,
stft_size=1024,
sample_rate=16000,
normalize=False):
center_frequencies = get_stft_center_frequencies(stft_size, sample_rate)
steering_vector = numpy.exp(-2j * numpy.pi * center_frequencies *
time_difference_of_arrival[..., numpy.newaxis])
if normalize:
steering_vector /= np.linalg.norm(steering_vector,
axis=-2,
keepdims=True)
return steering_vector