def first_downbeat_cost_all(x_t_array,fs,low_freq,high_freq,hop_sec,rms_thresh,offset,
actual_first_downbeat_array):
# Compute the percent of tracks in a set that have greater first downbeat error than
# a given tolerance, where first downbeat error is the error in seconds between the estimated
# and ground truth tempos
# Parameters:
# x_t_array: array of audio samples, where x_t_array[0] holds all
# samples of the first track, and so on
# fs: sample rate of audio files
# low_freq, high_freq, hop_sec, rms_thresh, offset:
# parameters for get_first_downbeat
# actual_tempo_array: array of ground truth tempos
# Returns:
# cost: decimal percent of tracks in a set that have greater than "tolerance" first downbeat error
cost_accum = 0.;
for i in range(len(x_t_array)):
first_downbeat = core.get_first_downbeat(x_t_array[i],fs,low_freq,high_freq,hop_sec,
rms_thresh,offset)[1]
if (abs(first_downbeat-actual_first_downbeat_array[i]) > 0.5): # within 0.5 seconds
cost_accum = cost_accum + 1
print actual_first_downbeat_array[i],first_downbeat
cost = cost_accum/len(x_t_array)
return cost
def histogram_segment_lengths(jamspath,track_names,x_t_array,fs,win_size_nov,hop_size_nov,win_size_tpo,
hop_size_tpo,min_bpm,max_bpm,low_freq,high_freq,hop_sec,rms_thresh,offset,
fix_or_not,tol,beat_or_meas,ref_or_est):
# From a set of (estimated or ground truth) segment boundary annotations and a set of tracks,
# compute two histograms: a histogram of the lengths of all segments in all tracks, in beats,
# as computed by our PLP beat tracker, and a histogram of the placement of each segment,
# in beats or measures, relative to the first downbeat
# Parameters:
# jamspath: folder in which the .jams boundary annotations or estimations reside
# track_names: names of the tracks, including audio extensions
# x_t_array: array of audio samples, where x_t_array[0] holds all
# samples of the first track, and so on
# fs: sample rate of audio files
# win_size_nov, hop_size_nov, win_size_tpo, hop_size_tpo, min_bpm, max_bpm:
# parameters for get_beats
# low_freq, high_freq, hop_sec, rms_thresh, offset:
# parameters for get_first_downbeat
# fix_or_not: boolean, 1 = perform length fixing on the segment lengths, 0 = don't
# tol: tolerance for length fixing, in beats/measures
# beats_or_meas: boolean, 1 = histogram lengths/placements in beats, 0 = in measures
# ref_or_est: boolean, 1 = annotated references, 0 = MSAF estimations (the jams files are different)
# Returns:
# length_histogram: histogram of segment lengths in beats/measures, where
# length_histogram[n] = number of n-long segments
# placement_histogram: histogram of segment boundary placements in beats/measures, where
# placement_histogram[n] = number of boundaries falling on nth beat/meas after first downbeat
length_histogram = np.zeros(400)
placement_histogram = np.zeros(2000)
for i in range(len(track_names)):
# read .jams file to obtain estimated times
print track_names[i]
if track_names[i][-4:] == 'flac':
f = open(jamspath+track_names[i][0:-4]+'jams','r')
else:
f = open(jamspath+track_names[i][0:-3]+'jams','r')
lines = f.readlines()
f.close()
est_times = []
if ref_or_est:
val_line = 9
while (lines[val_line][13:18] == 'value'):
est_times.append(lines[val_line][22:27])
val_line += 12
else:
val_line = 43
while (lines[val_line][13:18] == 'value'):
est_times.append(lines[val_line][21:27])
val_line += 15
val_line -= 11
est_times.append(lines[val_line][21:27])
est_times_string = np.array(est_times, dtype='|S6')
print est_times_string
est_times = est_times_string.astype(np.float)
# compute placements and lengths of segments
beats_idx,beat_times,tempo = core.get_beats(x_t_array[i],fs,win_size_nov,hop_size_nov,
win_size_tpo,hop_size_tpo,min_bpm,max_bpm)
first_downbeat_idx, first_downbeat_time = core.get_first_downbeat(x_t_array[i],fs,low_freq,high_freq,
hop_sec,rms_thresh,offset)
if not (beat_or_meas):
beats_idx, beat_times = core.get_downbeats(onset_idx,first_downbeat_idx,fs,tempo)
placements,lengths = core.compute_boundary_placements(est_times[0:-2],beat_times,first_downbeat_time)
# compute histograms
for l in lengths:
length_histogram[l] += 1
for p in placements:
placement_histogram[p] += 1
return length_histogram, placement_histogram
