def calculate_report(results_tuple):
r'''
This routine will calculate a WER report.
It'll compute the `mean` WER and create ``Sample`` objects of the ``report_count`` top lowest
loss items from the provided WER results tuple (only items with WER!=0 and ordered by their WER).
'''
samples = []
items = list(zip(*results_tuple))
total_levenshtein = 0.0
total_label_length = 0.0
for label, decoding, distance, loss in items:
sample_wer = wer(label, decoding)
sample = Sample(label, decoding, loss, distance, sample_wer)
samples.append(sample)
total_levenshtein += levenshtein(label.split(), decoding.split())
total_label_length += float(len(label.split()))
# Getting the WER from the accumulated levenshteins and lengths
samples_wer = total_levenshtein / total_label_length
# Filter out all items with WER=0
samples = [s for s in samples if s.wer > 0]
# Order the remaining items by their loss (lowest loss on top)
samples.sort(key=lambda s: s.loss)
# Take only the first report_count items
samples = samples[:FLAGS.report_count]
# Order this top FLAGS.report_count items by their WER (lowest WER on top)
samples.sort(key=lambda s: s.wer)
return samples_wer, samples
def process_decode_result(item):
label, decoding, distance, loss = item
sample_wer = wer(label, decoding)
return AttrDict({
'src': label,
'res': decoding,
'loss': loss,
'distance': distance,
'wer': sample_wer,
'levenshtein': levenshtein(label.split(), decoding.split()),
'label_length': float(len(label.split())),
})
def get_bestmatch_keywords_using_wer(str):
global allwords
if str in allwords:
return str
r = []
str1 = ' '.join(list(str))
for o in allwords:
o1 = ' '.join(list(o))
# print (type(o), type(str), o1, str1)
r.append(wer(o1, str1))
idx = np.argmin(np.array(r), axis=0)
# print (idx)
# print(str, allwords[idx])
return allwords[idx]
