def get_undirected_interval_grade(self, chorale):
key = chorale.analyze('key')
chorale_distribution = histogram_to_distribution(
get_interval_histogram(chorale, directed=False))
dataset_distribution = self.distributions[
f'{key.mode}_undirected_interval_distribution']
u, v, u_weights, v_weights = distributions_to_wasserstein_inputs(
chorale_distribution, dataset_distribution)
return wasserstein_distance(u, v, u_weights, v_weights)
def get_harmonic_quality_grade(self, chorale):
key = chorale.analyze('key')
chorale_distribution = histogram_to_distribution(
get_harmonic_quality_histogram(chorale))
dataset_distribution = self.distributions[
f'{key.mode}_harmonic_quality_distribution']
u, v, u_weights, v_weights = distributions_to_wasserstein_inputs(
chorale_distribution, dataset_distribution)
return wasserstein_distance(u, v, u_weights, v_weights)
def get_B_directed_interval_grade(self, chorale):
key = chorale.analyze('key')
voice_ih = get_SATB_interval_histogram(chorale, voice=3, directed=True)
chorale_distribution = histogram_to_distribution(voice_ih)
dataset_distribution = self.distributions[
f'{key.mode}_B_directed_interval_distribution']
u, v, u_weights, v_weights = distributions_to_wasserstein_inputs(
chorale_distribution, dataset_distribution)
return wasserstein_distance(u, v, u_weights, v_weights)
def get_rhythm_grade(self, chorale):
"""
Arguments
chorale: music21.stream.Stream
Returns Wasserstein distance between normalized chorale rhythm distribution and normalized dataset rhythm distribution
"""
chorale_distribution = histogram_to_distribution(
get_rhythm_histogram(chorale))
dataset_distribution = self.distributions['rhythm_distribution']
u, v, u_weights, v_weights = distributions_to_wasserstein_inputs(
chorale_distribution, dataset_distribution)
return wasserstein_distance(u, v, u_weights, v_weights)
def get_note_grade(self, chorale):
"""
Arguments
chorale: music21.stream.Stream
Returns Wasserstein distance between normalized chorale note distribution and normalized dataset note distribution
"""
key = chorale.analyze('key')
chorale_distribution = histogram_to_distribution(
get_note_histogram(chorale, key))
dataset_distribution = self.distributions[
f'{key.mode}_note_distribution']
u, v, u_weights, v_weights = distributions_to_wasserstein_inputs(
chorale_distribution, dataset_distribution)
return wasserstein_distance(u, v, u_weights, v_weights)
def get_error_grade(self, chorale):
num_notes = len(chorale.flat.notes)
chorale_histogram = get_error_histogram(chorale, self.voice_ranges)
num_errors = sum(chorale_histogram.values())
# if num_errors == 0, the coefficient will be 0, so we can sidestep the calculation of wasserstein
if num_errors == 0:
return 0
chorale_distribution = histogram_to_distribution(chorale_histogram)
dataset_distribution = self.distributions['error_distribution']
error_note_ratio = num_errors / num_notes
u, v, u_weights, v_weights = distributions_to_wasserstein_inputs(
chorale_distribution, dataset_distribution)
return wasserstein_distance(u, v, u_weights, v_weights) * (
error_note_ratio / self.error_note_ratio)