File size: 5,284 Bytes
7132467
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7693786
7132467
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6c5f6a4
7132467
 
5a444dc
7132467
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
142
143
144
145
146
147
148
149
150
151
152
153
#!/usr/bin/env python3
import argparse
import re
import unicodedata
from typing import Dict

import torch
from datasets import Audio, Dataset, load_dataset, load_metric

from transformers import AutoFeatureExtractor, pipeline


def log_results(result: Dataset, args: Dict[str, str]):
    """DO NOT CHANGE. This function computes and logs the result metrics."""

    log_outputs = args.log_outputs
    dataset_id = "_".join(args.dataset.split("/") + [args.config, args.split])

    # load metric
    wer = load_metric("wer")
    cer = load_metric("cer")

    # compute metrics
    wer_result = wer.compute(references=result["target"], predictions=result["prediction"])
    cer_result = cer.compute(references=result["target"], predictions=result["prediction"])

    # print & log results
    result_str = f"WER: {wer_result}\n" f"CER: {cer_result}"
    print(result_str)

    with open(f"{dataset_id}_eval_results.txt", "w") as f:
        f.write(result_str)

    # log all results in text file. Possibly interesting for analysis
    if log_outputs is not None:
        pred_file = f"log_{dataset_id}_predictions.txt"
        target_file = f"log_{dataset_id}_targets.txt"

        with open(pred_file, "w") as p, open(target_file, "w") as t:

            # mapping function to write output
            def write_to_file(batch, i):
                p.write(f"{i}" + "\n")
                p.write(batch["prediction"] + "\n")
                t.write(f"{i}" + "\n")
                t.write(batch["target"] + "\n")

            result.map(write_to_file, with_indices=True)


chars_to_remove_regex = r'[\,\?\.\!\-\_\;\:\"\“\%\‘\”\�\^]'

def remove_accents(text):
    nfkd_form = unicodedata.normalize('NFKD', text)
    return u"".join([c for c in nfkd_form if not unicodedata.combining(c)])

def remove_special_characters(text):
    text = re.sub(chars_to_remove_regex, '', text).lower()
    text = re.sub("ç", r'[cedille]', text)
    text = re.sub("&", r'et', text)    
    text = re.sub("%", r' pourcents', text)   
    text = re.sub("([0-9]+)(,|.)([0-9+])", r'\1 virgule \3', text) 
    text = re.sub("\$", r'dollar', text)
    text = re.sub("\£", r'livre', text)
    text = re.sub("\€", r'euro', text)
    text = remove_accents(text)
    text = re.sub(r"\[cedille\]", 'ç', text) + " "
    return text
    
def normalize_text(text: str) -> str:
    text = remove_special_characters(text)

    # In addition, we can normalize the target text, e.g. removing new lines characters etc...
    # note that order is important here!
    token_sequences_to_ignore = ["\n\n", "\n", "   ", "  "]

    for t in token_sequences_to_ignore:
        text = " ".join(text.split(t))

    return text


def main(args):
    # load dataset
    dataset = load_dataset(args.dataset, args.config, split=args.split, use_auth_token=True)

    # for testing: only process the first two examples as a test
    # dataset = dataset.select(range(2))

    # load processor
    feature_extractor = AutoFeatureExtractor.from_pretrained(args.model_id)
    sampling_rate = feature_extractor.sampling_rate

    # resample audio
    dataset = dataset.cast_column("audio", Audio(sampling_rate=sampling_rate))

    # load eval pipeline
    if args.device is None:
        args.device = 0 if torch.cuda.is_available() else -1
    asr = pipeline("automatic-speech-recognition", model=args.model_id, device=args.device)

    # map function to decode audio
    def map_to_pred(batch):
        prediction = asr(
            batch["audio"]["array"], chunk_length_s=args.chunk_length_s, stride_length_s=args.stride_length_s
        )

        batch["prediction"] = prediction["text"]  # "".join(prediction["text"].split("<s>"))
        batch["target"] = normalize_text(batch["sentence"])
        return batch

    # run inference on all examples
    result = dataset.map(map_to_pred, remove_columns=dataset.column_names)

    # compute and log_results
    # do not change function below
    log_results(result, args)


if __name__ == "__main__":
    parser = argparse.ArgumentParser()

    parser.add_argument(
        "--model_id", type=str, required=True, help="Model identifier. Should be loadable with 🤗 Transformers"
    )
    parser.add_argument(
        "--dataset",
        type=str,
        required=True,
        help="Dataset name to evaluate the `model_id`. Should be loadable with 🤗 Datasets",
    )
    parser.add_argument(
        "--config", type=str, required=True, help="Config of the dataset. *E.g.* `'en'`  for Common Voice"
    )
    parser.add_argument("--split", type=str, required=True, help="Split of the dataset. *E.g.* `'test'`")
    parser.add_argument(
        "--chunk_length_s", type=float, default=None, help="Chunk length in seconds. Defaults to 5 seconds."
    )
    parser.add_argument(
        "--stride_length_s", type=float, default=None, help="Stride of the audio chunks. Defaults to 1 second."
    )
    parser.add_argument(
        "--log_outputs", action="store_true", help="If defined, write outputs to log file for analysis."
    )
    parser.add_argument(
        "--device",
        type=int,
        default=None,
        help="The device to run the pipeline on. -1 for CPU (default), 0 for the first GPU and so on.",
    )
    args = parser.parse_args()

    main(args)