def main(mf_name):
coll = mar_collection.MarCollection(mf_name)
filenames = coll.get_filenames()
bpm_mf_name = os.path.basename(mf_name)
bpm_mf_name = bpm_mf_name.replace(".mf", "-echonest_bpm.mf")
bpm_coll = mar_collection.MarCollection(bpm_mf_name)
for filename in filenames:
#print filename
#continue
bpm = echonest_upload(filename)
bpm_coll.set_item(filename, bpm)
bpm_coll.write()
def scheirer_collection(queue, datum, detected_mf_template):
mf_filename = datum[0]
output_dir = datum[1]
if not os.path.exists(output_dir):
os.makedirs(output_dir)
mf_basename = os.path.basename(
os.path.splitext(mf_filename)[0])
dirname = os.path.join(output_dir, mf_basename)
if not os.path.exists(dirname):
os.makedirs(dirname)
os.chdir(dirname)
detected_mf_filename = '%s-%s.mf' % (mf_basename, detected_mf_template)
out = open(detected_mf_filename, 'w')
coll = mar_collection.MarCollection(mf_filename)
problems = 0
for audio_filename in coll.get_filenames():
bpm = single_file_scheirer(
audio_filename)
if bpm < 0:
problems += 1
out.write("%s\t%s\n" % (audio_filename, bpm))
out.close()
### generate a wrong.mf for each collection
cmd = "tempo -pi %s -m PREDICTED %s" % (
detected_mf_filename, mf_filename)
subprocess.check_call(cmd, shell=True)
queue.put( (mf_filename, problems) )
def process_mfs(ground_mf, detected_mf, limit=None):
# load ground truth
ground_coll = mar_collection.MarCollection(ground_mf)
ground_bpms = {}
for dat in ground_coll.data:
filename = dat[0]
bpm_ground = float(dat[1])
ground_bpms[filename] = bpm_ground
user_coll = mar_collection.MarCollection(detected_mf)
good = 0
i = 0
for dat in user_coll.data:
#for dat in user_coll.data[:5]:
filename = dat[0]
cand_bpms = dat[1]
bpm_ground = ground_bpms[filename]
if "," in cand_bpms:
cand_bpms = [float(a) for a in cand_bpms.split(',')]
correct = False
if limit is not None and limit is not 0:
cand_bpms = cand_bpms[:limit]
if limit == 0:
for bpm_detected in cand_bpms[:1]:
corr = exact_accuracy(bpm_detected, bpm_ground)
if corr:
correct = True
if correct:
good += 1
i += 1
else:
for bpm_detected in cand_bpms:
corr = extended_harmonic_accuracy(bpm_detected, bpm_ground)
if corr:
correct = True
if correct:
good += 1
i += 1
#print cand_bpms
#print "Accuracy: %.2f (%i/%i)" % (100*float(good) / i, good, i)
accuracy = 100 * float(good) / len(user_coll.data)
return accuracy
def bpm_of_mf(defs, mf_filename, print_info=False):
coll = mar_collection.MarCollection(mf_filename)
new_filename = os.path.basename(mf_filename[:-3] + "-my-%i-%i.mf") % (
defs.OPTIONS_ONSET, defs.OPTIONS_BH)
newer = mar_collection.MarCollection(new_filename)
#out = open('detailed-cands.txt', 'w')
num_files = len(coll.data)
i = 0
good = 0
for dat in coll.data:
filename = dat[0]
bpm_ground = float(dat[1])
#print filename
bpm_detect, cands = bpm_of_file(defs, filename)
#out.write("%s\t%s\n" % (filename, cands))
bpm_label = str(bpm_detect)
#bpm_label = ','.join([str("%.3f") % f for f in cands])
#print filename
#print cands
#print bpm_label
newer.set_item(filename, bpm_label)
i += 1
#print "%i / %i" % (i, num_files)
#acc = evaluate_bpms.extended_harmonic_accuracy(bpm_detect, bpm_ground)
acc = evaluate_bpms.major_extended_harmonic_accuracy(
bpm_detect, bpm_ground)
if acc:
good += 1
accuracy = 100 * float(good) / i
if print_info:
print "Accuracy: %.2f (%i/%i)" % (accuracy, good, i),
if not acc:
print "\tground: %.2f\tdetected: %.2f" % (bpm_ground,
bpm_detect),
print filename
print
newer.write()
#out.close()
return accuracy
def main(mf_name):
try:
_ = os.environ["ECHO_NEST_API_KEY"]
except:
print "Need echonest key! you must run:"
print " export ECHO_NEST_API_KEY="
exit(1)
coll = mar_collection.MarCollection(mf_name)
if PROCESS_PROBLEMS:
filenames_z = coll.get_filenames_matching_label("0")
filenames_zp = coll.get_filenames_matching_label("0.0")
filenames = filenames_z + filenames_zp
else:
filenames = coll.get_filenames()
id_mf_name = os.path.basename(mf_name)
id_mf_name = id_mf_name.replace(".mf", "-echonest_id.mf")
id_coll = mar_collection.MarCollection(id_mf_name)
bpm_mf_name = os.path.basename(mf_name)
bpm_mf_name = bpm_mf_name.replace(".mf", "-echonest_bpm.mf")
bpm_coll = mar_collection.MarCollection(bpm_mf_name)
problems = []
for filename in filenames:
#print filename
#continue
echonest_id, bpm = echonest_upload(filename)
id_coll.set_item(filename, echonest_id)
bpm_coll.set_item(filename, bpm)
if bpm is 0:
problems.append(filename)
id_coll.write()
bpm_coll.write()
if len(problems) > 0:
print "*** problems:", len(problems)
out = open("problems.mf", 'w')
for p in problems:
out.write("%s\n" % p)
out.close()
def vamp_qm_default(queue, datum):
mf_filename = datum[0]
output_dir = datum[1]
if not os.path.exists(output_dir):
os.makedirs(output_dir)
mf_basename = os.path.basename(os.path.splitext(mf_filename)[0])
dirname = os.path.join(output_dir, mf_basename)
if not os.path.exists(dirname):
os.makedirs(dirname)
os.chdir(dirname)
detected_mf_filename_mean = '%s-qm_default_mean.mf' % (mf_basename)
detected_mf_filename_median = '%s-qm_default_median.mf' % (mf_basename)
detected_mf_filename_mode = '%s-qm_default_mode.mf' % (mf_basename)
out_mean = open(detected_mf_filename_mean, 'w')
out_median = open(detected_mf_filename_median, 'w')
out_mode = open(detected_mf_filename_mode, 'w')
coll = mar_collection.MarCollection(mf_filename)
problems = 0
for audio_filename in coll.get_filenames():
result_filename = single_file_generate(audio_filename, dirname)
bpm_mean, bpm_median, bpm_mode = single_file_evaluate_mean_median(
audio_filename, result_filename)
if bpm_mean < 0 or bpm_median < 0:
problems += 1
out_mean.write("%s\t%s\n" % (audio_filename, bpm_mean))
out_median.write("%s\t%s\n" % (audio_filename, bpm_median))
out_mode.write("%s\t%s\n" % (audio_filename, bpm_mode))
out_mean.close()
out_median.close()
out_mode.close()
### generate a wrong.mf for each collection
cmd = "tempo -pi %s -m PREDICTED %s" % (detected_mf_filename_mean,
mf_filename)
subprocess.check_call(cmd, shell=True)
#cmd = "mv %s %s" % ("wrong.mf", "wrong-mean.mf")
#subprocess.check_call(cmd, shell=True)
cmd = "tempo -pi %s -m PREDICTED %s" % (detected_mf_filename_median,
mf_filename)
subprocess.check_call(cmd, shell=True)
#cmd = "mv %s %s" % ("wrong.mf", "wrong-median.mf")
#subprocess.check_call(cmd, shell=True)
cmd = "tempo -pi %s -m PREDICTED %s" % (detected_mf_filename_mode,
mf_filename)
subprocess.check_call(cmd, shell=True)
#cmd = "mv %s %s" % ("wrong.mf", "wrong-mode.mf")
#subprocess.check_call(cmd, shell=True)
queue.put((mf_filename, problems))
#!/usr/bin/env python
import os
import sys
import scipy.io.wavfile
import numpy
import mar_collection
def rewrite_file(filename):
print "give attention to:\t", filename
tmp = "tmp_wavfile.wav"
cmd = "sox \"%s\" -t wavpcm \"%s\"" % (filename, tmp)
os.system(cmd)
os.system("cp \"%s\" \"%s\"" % (tmp, filename))
if __name__ == "__main__":
user_filename = sys.argv[1]
if user_filename[-3:] == ".mf":
coll = mar_collection.MarCollection(user_filename)
for filename in coll.get_filenames():
try:
sr, data_unnormalized = scipy.io.wavfile.read(filename)
maxval = numpy.iinfo(data_unnormalized.dtype).max+1
except:
rewrite_file(filename)
pass
bpm = late_heuristic.late_heuristic(bpm1, bpm2, candidate_bpms[-1][0])
#bpm_i = summed_beat_histograms.argmax()
#bpm = histogram_bpms[bpm_i]
if plot:
pylab.show()
#print bpm
#return bpm, candidate_bpms
return bpm
if __name__ == "__main__":
user_filename = sys.argv[1]
if user_filename[-3:] == ".mf":
coll = mar_collection.MarCollection(user_filename)
new_filename = os.path.basename(
user_filename[:-3] + "-my.mf")
newer = mar_collection.MarCollection(new_filename)
#out = open('detailed-cands.txt', 'w')
num_files = len(coll.data)
i = 0
for dat in coll.data:
filename = dat[0]
bpm_ground = dat[1]
#print filename
bpm_detect = main(filename)
#out.write("%s\t%s\n" % (filename, cands))
newer.set_item( filename, bpm_detect)
i += 1
print "%i / %i" % (i, num_files)
#!/usr/bin/env python import sys import mar_collection base = sys.argv[1] new = sys.argv[2] base_coll = mar_collection.MarCollection(base) base_coll.merge_mf(new) base_coll.write()
