PereLluis13 commited on
Commit
8b50196
1 Parent(s): 6324d57

Upload eval.py

Browse files
Files changed (1) hide show
  1. eval.py +129 -0
eval.py ADDED
@@ -0,0 +1,129 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ from datasets import load_dataset, load_metric, Audio, Dataset, set_caching_enabled
3
+ from transformers import pipeline, AutoFeatureExtractor
4
+ import re
5
+ import argparse
6
+ import unicodedata
7
+ from typing import Dict
8
+ from text.numbers_ca import normalize_numbers_ca
9
+
10
+ def log_results(result: Dataset, args: Dict[str, str]):
11
+ """ DO NOT CHANGE. This function computes and logs the result metrics. """
12
+
13
+ log_outputs = args.log_outputs
14
+ dataset_id = "_".join(args.dataset.split("/") + [args.config, args.split])
15
+
16
+ # load metric
17
+ wer = load_metric("wer")
18
+ cer = load_metric("cer")
19
+
20
+ # compute metrics
21
+ wer_result = wer.compute(references=result["target"], predictions=result["prediction"])
22
+ cer_result = cer.compute(references=result["target"], predictions=result["prediction"])
23
+
24
+ # print & log results
25
+ result_str = (
26
+ f"WER: {wer_result}\n"
27
+ f"CER: {cer_result}"
28
+ )
29
+ print(result_str)
30
+ model_name = args.model_id.split("/")[-1]
31
+ with open(f"{dataset_id}_{model_name}_eval_results.txt", "w") as f:
32
+ f.write(result_str)
33
+
34
+ # log all results in text file. Possibly interesting for analysis
35
+ if log_outputs is not None:
36
+ pred_file = f"log_{dataset_id}_{model_name}_predictions.txt"
37
+ target_file = f"log_{dataset_id}_{model_name}_targets.txt"
38
+
39
+ with open(pred_file, "w") as p, open(target_file, "w") as t:
40
+
41
+ # mapping function to write output
42
+ def write_to_file(batch, i):
43
+ p.write(f"{i}" + "\n")
44
+ p.write(batch["prediction"] + "\n")
45
+ t.write(f"{i}" + "\n")
46
+ t.write(batch["target"] + "\n")
47
+
48
+ result.map(write_to_file, with_indices=True)
49
+
50
+ def normalize_text(text: str) -> str:
51
+ """ DO ADAPT FOR YOUR USE CASE. this function normalizes the target text. """
52
+ text = normalize_numbers_ca(text)
53
+
54
+ chars_to_ignore_regex = '[,?.!\;\:"“%”�—…–]' # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
55
+
56
+ text = text.lower()
57
+ # normalize non-standard (stylized) unicode characters
58
+ text = unicodedata.normalize('NFKC', text)
59
+ # remove punctuation
60
+ text = re.sub(chars_to_ignore_regex, "", text)
61
+ text = re.sub("á", "a", text)
62
+ text = re.sub("ñ", "ny", text)
63
+ # Let's also make sure we split on all kinds of newlines, spaces, etc...
64
+ text = " ".join(text.split())
65
+
66
+ return text
67
+
68
+ def main(args):
69
+ # load dataset
70
+ dataset = load_dataset(args.dataset, args.config, split=args.split, use_auth_token=True)
71
+ # for testing: only process the first two examples as a test
72
+ # dataset = dataset.select(range(10))
73
+
74
+ # load processor
75
+ feature_extractor = AutoFeatureExtractor.from_pretrained(args.model_id)
76
+ sampling_rate = feature_extractor.sampling_rate
77
+
78
+ # resample audio
79
+ dataset = dataset.cast_column("audio", Audio(sampling_rate=sampling_rate))
80
+
81
+ # load eval pipeline
82
+ asr = pipeline("automatic-speech-recognition", model=args.model_id, device=0)
83
+
84
+ # map function to decode audio
85
+ def map_to_pred(batch):
86
+ prediction = asr(batch["audio"]["array"], chunk_length_s=args.chunk_length_s, stride_length_s=args.stride_length_s)
87
+
88
+ batch["prediction"] = prediction["text"]
89
+ batch["target"] = normalize_text(batch[args.text_column])
90
+ return batch
91
+
92
+ # run inference on all examples
93
+ result = dataset.map(map_to_pred, remove_columns=dataset.column_names)
94
+
95
+ # compute and log_results
96
+ # do not change function below
97
+ log_results(result, args)
98
+
99
+
100
+ if __name__ == "__main__":
101
+ parser = argparse.ArgumentParser()
102
+
103
+ parser.add_argument(
104
+ "--model_id", type=str, required=True, help="Model identifier. Should be loadable with 🤗 Transformers"
105
+ )
106
+ parser.add_argument(
107
+ "--dataset", type=str, required=True, help="Dataset name to evaluate the `model_id`. Should be loadable with 🤗 Datasets"
108
+ )
109
+ parser.add_argument(
110
+ "--config", type=str, required=True, help="Config of the dataset. *E.g.* `'en'` for Common Voice"
111
+ )
112
+ parser.add_argument(
113
+ "--split", type=str, required=True, help="Split of the dataset. *E.g.* `'test'`"
114
+ )
115
+ parser.add_argument(
116
+ "--text_column", type=str, default="sentence", help="The name of the dataset column containing the text data. Defaults to 'text'"
117
+ )
118
+ parser.add_argument(
119
+ "--chunk_length_s", type=float, default=None, help="Chunk length in seconds. Defaults to None. For long audio files a good value would be 5.0 seconds."
120
+ )
121
+ parser.add_argument(
122
+ "--stride_length_s", type=float, default=None, help="Stride of the audio chunks. Defaults to None. For long audio files a good value would be 1.0 seconds."
123
+ )
124
+ parser.add_argument(
125
+ "--log_outputs", action='store_true', help="If defined, write outputs to log file for analysis."
126
+ )
127
+ args = parser.parse_args()
128
+
129
+ main(args)