def render(score, segmentation, hmm, f_discretization=10, e_discretization=30, subset=None, sensitivity=5):
# Extract scorefeatures
features = sf.vanDerWeijFeatures(score, segmentation)
# Discretize
features = normalize(preprocess(sf.vanDerWeijFeatures(score, segmentation), subset), hmm.normalizations)
observations = [tuple(discretize(f, f_discretization, 'linear')) for f in features]
print('Observations:\n{0}'.format(observations))
# Find the best expressive explanation for these features
print("Finding best fitting expression")
(p, states) = hmm.viterbi(observations)
print(p, states)
expression = []
for state in states:
expression.append(undiscretize_expression(state, e_discretization, sensitivity=sensitivity))
print('Expression states:\n{0}'.format(states))
return (p, expression)
def train(trainset):
expression = {}
features = {}
const = 0
Max = 0
Min = None
count = 0
print(">>> Loading scores and deviations, this will take hours and may eat all you memory")
for query in trainset:
print(">>> Loading: {0}".format(query))
score = db.getScore1(query)
deviations = db.getDeviation1(query)
alignment = Alignment(score, deviations)
melody = alignment.melody()
#segments = structure.newSegmentation(tools.parseScore(melody))
segments = structure.noteLevel(tools.parseScore(melody))
const += len(segments)
lengths = sum([len(s) for s in segments])
m = max([len(s) for s in segments])
mi = min([len(s) for s in segments])
if m > Max:
Max = m
if not Min:
Min = mi
if mi < Min:
Min = mi
print('>>> Extracting features')
expression[query] = performancefeatures.vanDerWeijExpression(alignment, segments)
features[query] = scorefeatures.vanDerWeijFeatures(melody, segments)
count += 1
print('{0}/{1} done'.format(count, len(trainset)))
print("Done, {0} segments found with an average length of: {1} (min: {2} max: {3})".format(const, lengths / float(const), Min, Max))
tools.saveFeatures(features, expression)
elif a[1] == 'corpora':
for x in tools.datasets():
print('Name:\t[{0}]\tInfo:\t{1}'.format(x, tools.corpusInfo(x)))
exit(0)
elif a[1] == 'corpusinfo':
choice = util.menu("Select corpus", tools.datasets())
print(tools.corpusInfo(tools.datasets()[choice]))
tools.extendedCorpusInfo(tools.datasets()[choice])
exit(0)
elif a[1] == 'features':
s = db.select()
score = Score(db.getScore1(s))
melodyscore = score.melody()
melody = tools.parseScore(melodyscore)
segmentation = structure.reasonableSegmentation(melody)
f = sf.vanDerWeijFeatures(melodyscore, segmentation)
for x in f:
#(dPitch, abs_dPitch, dDuration, abs_dDuration, silence, ddPitch, abs_ddPitch, pitch_direction)
print('dpitch: {0} abs_dPitch: {1} dDuration: {2} abs_dDuration: {3} silence: {4}'.format(x[0], x[1], x[2], x[3], x[4]))
print('ddPitch: {0} abs_ddPitch: {1} : pitch_direction: {2}'.format(x[5], x[6], x[7]))
exit(0)
elif a[1] == 'merge':
# Merge datasets
choice = -1
datasets = []
while True:
choice = util.menu('Choose datasets', ['Done'] + tools.datasets())
if choice == 0: break
datasets.append(tools.loadFeatures(tools.datasets()[choice-1]))
features = {}
expression = {}
import database as db from sequencer import * from alignment import * import structure, tools import scorefeatures as sf import performancefeatures as pf import perform s = db.select() a = Alignment(db.getScore1(s), db.getDeviation1(s)) melodyscore = a.melody() melody = tools.parseScore(melodyscore) onset = structure.groupings(structure.list_to_tree(structure.first_order_tree(structure.onset, melody, 0.1)), 1) score = sf.vanDerWeijFeatures(melodyscore, onset) performance = pf.vanDerWeijExpression(a, onset) print(score) print(performance) seq = Sequencer() seq.play(perform.vanDerWeijPerformSimple(a.score, melodyscore, onset, performance, bpm=a.deviations.bpm, converter=melody))
