forked from igormq/speech2text
-
Notifications
You must be signed in to change notification settings - Fork 0
/
api.py
141 lines (107 loc) · 4.34 KB
/
api.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
import argparse
import json
import os
import re
import pandas as pd
from num2words import num2words
from tqdm import tqdm
from asr.apis import SpeechRecognitionAPI
from asr.data.transforms import ToLabel
from asr.utils.error_rate import cer, wer
DEFAULT_APIS = ['gcp', 'ibm', 'bing']
def transcribe(args):
data_dir = args.data_dir
apis = args.apis
lang = args.lang
manifest = args.manifest
file = args.file
output = args.output
if manifest:
audio_files = list(
map(
lambda x: (os.path.join(data_dir, x[0]),
open(os.path.join(data_dir, x[1]), 'r', encoding='utf8').read().strip()),
pd.read_csv(manifest).values))
if file:
if not os.path.isfile(file):
raise ValueError('{} not found.'.format(file))
audio_files = [(file, '')]
# preload saved results
if os.path.exists(output):
with open(output, 'r', encoding='utf8') as f:
out_data = json.load(f)
else:
out_data = {}
# construct the api objects
apis = {api: SpeechRecognitionAPI(api, lang=lang) for api in apis}
for audio_file, transcription in tqdm(audio_files):
out_data.setdefault(audio_file, {})
out_data[audio_file]['ref'] = transcription
for api_name, api in apis.items():
if api_name in out_data[audio_file]:
continue
out_data[audio_file][api_name] = api.recognize(audio_file)
# saving results
with open(output, 'w', encoding='utf8') as f:
json.dump(out_data, f, indent=4, ensure_ascii=False)
def _parse_numbers(x, lang):
def cast_number(x):
try:
return int(x)
except ValueError:
return float(x)
x = re.sub(r'r\$\s+([0-9]+)', r'\1 reais', x)
x = re.sub(r'([0-9]+)r\$', r'\1 reais', x)
if lang == 'pt_BR':
x = x.replace('por cento', 'porcento')
x = re.sub(r'\+([0-9]+)', r'mais \1', x)
x = re.sub(r'([0-9]+)\%', r'\1 porcento', x)
x = re.sub(r'([0-9]+)(?:o|a)s?.?',
lambda x: num2words(int(x.group(1).replace(',', '.')), ordinal=True, lang=lang), x)
return re.sub(ToLabel.GET_NUMBERS_PATTERN,
lambda x: num2words(cast_number(x.group().replace(',', '.')), lang=lang), x)
def evaluate(args):
filepath = args.file
lang = args.lang
with open(filepath, 'r') as f:
data = json.load(f)
total_wer = {}
total_cer = {}
num_tokens = {}
num_chars = {}
for utterance in tqdm(data):
ref = _parse_numbers(data[utterance]['ref'].lower(), lang)
for api in set(data[utterance].keys()).difference(set(['ref'])):
total_wer.setdefault(api, 0)
total_cer.setdefault(api, 0)
num_tokens.setdefault(api, 0)
num_chars.setdefault(api, 0)
transcript = _parse_numbers(data[utterance][api].lower(), lang)
wer_inst = wer(transcript, ref)
cer_inst = cer(transcript, ref)
total_wer[api] += wer_inst
total_cer[api] += cer_inst
num_tokens[api] += len(ref.split())
num_chars[api] += len(ref)
for api in set(total_wer.keys()):
print("{} - WER: {:.02f}% CER: {:.02f}%".format(
api, (float(total_wer[api]) / num_tokens[api]) * 100,
(float(total_cer[api]) / num_chars[api]) * 100))
if __name__ == '__main__':
parser = argparse.ArgumentParser()
subparsers = parser.add_subparsers()
trans_parser = subparsers.add_parser('transcribe')
trans_parser.add_argument('--api', default=DEFAULT_APIS, nargs='+')
trans_parser.add_argument('--lang', default='pt_BR', choices=['pt_BR', 'en'])
trans_parser.add_argument('--data-dir', default='data', type=str)
trans_file_group = trans_parser.add_mutually_exclusive_group(required=True)
trans_file_group.add_argument('--manifest', '-m', type=str)
trans_file_group.add_argument('--file', '-f', type=str)
trans_parser.add_argument('--output', default='api.results.json', type=str)
trans_parser.set_defaults(func=transcribe)
eval_parser = subparsers.add_parser('eval')
eval_parser.add_argument('--lang', default='pt_BR', choices=['pt_BR', 'en'])
eval_parser.add_argument('file', type=str)
eval_parser.set_defaults(func=evaluate)
args = parser.parse_args()
args.func(args)