AndrewMcDowell commited on
Commit
ab72f16
β€’
1 Parent(s): 887461a

Upload eval_arabic.py

Browse files
Files changed (1) hide show
  1. eval_arabic.py +127 -0
eval_arabic.py ADDED
@@ -0,0 +1,127 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ from datasets import load_dataset, load_metric, Audio, Dataset
3
+ from transformers import pipeline, AutoFeatureExtractor
4
+ import re
5
+ import argparse
6
+ import unicodedata
7
+ from typing import Dict
8
+
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
+
31
+ with open(f"{dataset_id}_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}_predictions.txt"
37
+ target_file = f"log_{dataset_id}_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
+
51
+ def normalize_text(text: str) -> str:
52
+ """ DO ADAPT FOR YOUR USE CASE. this function normalizes the target text. """
53
+
54
+ chars_to_ignore_regex = '[,?.!\-\;\:\"β€œ%β€˜β€οΏ½β€”β€™β€¦β€“]' # 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
+
62
+ # Let's also make sure we split on all kinds of newlines, spaces, etc...
63
+ text = " ".join(text.split())
64
+
65
+ return text
66
+
67
+
68
+ def main(args):
69
+ # load dataset
70
+ dataset = load_dataset(args.dataset, args.config, split=args.split, use_auth_token=True)
71
+
72
+ # for testing: only process the first two examples as a test
73
+ # dataset = dataset.select(range(10))
74
+
75
+ # load processor
76
+ feature_extractor = AutoFeatureExtractor.from_pretrained(args.model_id)
77
+ sampling_rate = feature_extractor.sampling_rate
78
+
79
+ # resample audio
80
+ dataset = dataset.cast_column("audio", Audio(sampling_rate=sampling_rate))
81
+
82
+ # load eval pipeline
83
+ asr = pipeline("automatic-speech-recognition", model=args.model_id)
84
+
85
+ # map function to decode audio
86
+ def map_to_pred(batch):
87
+ prediction = asr(batch["audio"]["array"], chunk_length_s=args.chunk_length_s, stride_length_s=args.stride_length_s)
88
+
89
+ batch["prediction"] = prediction["text"]
90
+ batch["target"] = normalize_text(batch["sentence"])
91
+ return batch
92
+
93
+ # run inference on all examples
94
+ result = dataset.map(map_to_pred, remove_columns=dataset.column_names)
95
+
96
+ # compute and log_results
97
+ # do not change function below
98
+ log_results(result, args)
99
+
100
+
101
+ if __name__ == "__main__":
102
+ parser = argparse.ArgumentParser()
103
+
104
+ parser.add_argument(
105
+ "--model_id", type=str, required=True, help="Model identifier. Should be loadable with πŸ€— Transformers"
106
+ )
107
+ parser.add_argument(
108
+ "--dataset", type=str, required=True, help="Dataset name to evaluate the `model_id`. Should be loadable with πŸ€— Datasets"
109
+ )
110
+ parser.add_argument(
111
+ "--config", type=str, required=True, help="Config of the dataset. *E.g.* `'en'` for Common Voice"
112
+ )
113
+ parser.add_argument(
114
+ "--split", type=str, required=True, help="Split of the dataset. *E.g.* `'test'`"
115
+ )
116
+ parser.add_argument(
117
+ "--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."
118
+ )
119
+ parser.add_argument(
120
+ "--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."
121
+ )
122
+ parser.add_argument(
123
+ "--log_outputs", action='store_true', help="If defined, write outputs to log file for analysis."
124
+ )
125
+ args = parser.parse_args()
126
+
127
+ main(args)