AI/wekws
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

remove defaultdict

Browse Source
This commit is contained in:
blessyyyu 2022-03-24 14:19:14 +08:00
parent 00c5acf31a
commit bb1fc50d5a
2 changed files with 22 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
kws/bin/compute_det.py
View File

@ -17,15 +17,20 @@ import argparse
import json import json
from collections import defaultdict from collections import defaultdict
def load_label_and_score(keyword, label_file, score_file): def load_label_and_score(keyword, label_file, score_file):
score_table = defaultdict(list) # score_table: {uttid: [keywordlist]}
score_table = {}
with open(score_file, 'r', encoding='utf8') as fin: with open(score_file, 'r', encoding='utf8') as fin:
for line in fin: for line in fin:
arr = line.strip().split() arr = line.strip().split()
key = arr[0] key = arr[0]
str_list = arr[1:] current_keyword = arr[1]
scores = list(map(float, str_list)) str_list = arr[2:]
score_table[key].append(scores) if int(current_keyword) == keyword:
scores = list(map(float, str_list))
if key not in score_table:
score_table.update({key: scores})
keyword_table = {} keyword_table = {}
filler_table = {} filler_table = {}
filler_duration = 0.0 filler_duration = 0.0
@ -46,12 +51,14 @@ def load_label_and_score(keyword, label_file, score_file):
filler_duration += duration filler_duration += duration
return keyword_table, filler_table, filler_duration return keyword_table, filler_table, filler_duration
if __name__ == '__main__': if __name__ == '__main__':
parser = argparse.ArgumentParser(description='compute det curve') parser = argparse.ArgumentParser(description='compute det curve')
parser.add_argument('--test_data', required=True, help='label file') parser.add_argument('--test_data', required=True, help='label file')
parser.add_argument('--keyword', type=int, default=0, help='score file') parser.add_argument('--keyword', type=int, default=0, help='keyword label')
parser.add_argument('--score_file', required=True, help='score file') parser.add_argument('--score_file', required=True, help='score file')
parser.add_argument('--step', type=float, default=0.01, help='score file') parser.add_argument('--step', type=float, default=0.01,
help='threshold step')
parser.add_argument('--window_shift', type=int, default=50, parser.add_argument('--window_shift', type=int, default=50,
help='window_shift is used to skip the frames after triggered') help='window_shift is used to skip the frames after triggered')
parser.add_argument('--stats_file', parser.add_argument('--stats_file',
@ -63,32 +70,32 @@ if __name__ == '__main__':
args.keyword, args.test_data, args.score_file) args.keyword, args.test_data, args.score_file)
print('Filler total duration Hours: {}'.format(filler_duration / 3600.0)) print('Filler total duration Hours: {}'.format(filler_duration / 3600.0))
with open(args.stats_file, 'w', encoding='utf8') as fout: with open(args.stats_file, 'w', encoding='utf8') as fout:
keyword_index = int(args.stats_file.split('/')[-1].split('.')[1]) keyword_index = int(args.keyword)
threshold = 0.0 threshold = 0.0
while threshold <= 1.0: while threshold <= 1.0:
num_false_reject = 0 num_false_reject = 0
# transverse the all keyword_table # transverse the all keyword_table
for key, scores_list in keyword_table.items(): for key, score_list in keyword_table.items():
# computer positive test sample, use the max score of list. # computer positive test sample, use the max score of list.
score = max(scores_list[keyword_index]) score = max(score_list)
if float(score) < threshold: if float(score) < threshold:
num_false_reject += 1 num_false_reject += 1
num_false_alarm = 0 num_false_alarm = 0
# transverse the all filler_table # transverse the all filler_table
for key, scores_list in filler_table.items(): for key, score_list in filler_table.items():
i = 0 i = 0
score_list = scores_list[keyword_index]
while i < len(score_list): while i < len(score_list):
if score_list[i] >= threshold: if score_list[i] >= threshold:
num_false_alarm += 1 num_false_alarm += 1
i += window_shift i += window_shift
else: else:
i += 1 i += 1
if len(keyword_table) != 0 : if len(keyword_table) != 0:
false_reject_rate = num_false_reject / len(keyword_table) false_reject_rate = num_false_reject / len(keyword_table)
num_false_alarm = max(num_false_alarm, 1e-6) num_false_alarm = max(num_false_alarm, 1e-6)
if filler_duration != 0: if filler_duration != 0:
false_alarm_per_hour = num_false_alarm / (filler_duration / 3600.0) false_alarm_per_hour = num_false_alarm / \
(filler_duration / 3600.0)
fout.write('{:.6f} {:.6f} {:.6f}\n'.format(threshold, fout.write('{:.6f} {:.6f} {:.6f}\n'.format(threshold,
false_alarm_per_hour, false_alarm_per_hour,
false_reject_rate)) false_reject_rate))

3
kws/bin/score.py
View File

@ -119,7 +119,8 @@ def main():
keyword_scores = score[:, keyword_i] keyword_scores = score[:, keyword_i]
score_frames = ' '.join(['{:.6f}'.format(x) score_frames = ' '.join(['{:.6f}'.format(x)
for x in keyword_scores.tolist()]) for x in keyword_scores.tolist()])
fout.write('{} {}\n'.format(key, score_frames)) fout.write('{} {} {}\n'.format(
key, keyword_i, score_frames))
if batch_idx % 10 == 0: if batch_idx % 10 == 0:
print('Progress batch {}'.format(batch_idx)) print('Progress batch {}'.format(batch_idx))
sys.stdout.flush() sys.stdout.flush()