aapot commited on
Commit
310c4e5
1 Parent(s): 586894d

Upload lm-boosted decoder

Browse files
.gitattributes CHANGED
@@ -25,3 +25,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
25
  *.zip filter=lfs diff=lfs merge=lfs -text
26
  *.zstandard filter=lfs diff=lfs merge=lfs -text
27
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
25
  *.zip filter=lfs diff=lfs merge=lfs -text
26
  *.zstandard filter=lfs diff=lfs merge=lfs -text
27
  *tfevents* filter=lfs diff=lfs merge=lfs -text
28
+ language_model/unigrams.txt filter=lfs diff=lfs merge=lfs -text
alphabet.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"labels": [" ", "a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z", "\u00e4", "\u00e5", "\u00f6", "\u2047", "", "<s>", "</s>"], "is_bpe": false}
eval.py ADDED
@@ -0,0 +1,144 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ import argparse
3
+ import re
4
+ from typing import Dict
5
+
6
+ import torch
7
+ from datasets import Audio, Dataset, load_dataset, load_metric
8
+
9
+ from transformers import AutoFeatureExtractor, pipeline
10
+
11
+
12
+ def log_results(result: Dataset, args: Dict[str, str]):
13
+ """DO NOT CHANGE. This function computes and logs the result metrics."""
14
+
15
+ log_outputs = args.log_outputs
16
+ dataset_id = "_".join(args.dataset.split("/") + [args.config, args.split])
17
+
18
+ # load metric
19
+ wer = load_metric("wer")
20
+ cer = load_metric("cer")
21
+
22
+ # compute metrics
23
+ wer_result = wer.compute(references=result["target"], predictions=result["prediction"])
24
+ cer_result = cer.compute(references=result["target"], predictions=result["prediction"])
25
+
26
+ # print & log results
27
+ result_str = f"WER: {wer_result}\n" f"CER: {cer_result}"
28
+ print(result_str)
29
+
30
+ with open(f"{dataset_id}_eval_results.txt", "w") as f:
31
+ f.write(result_str)
32
+
33
+ # log all results in text file. Possibly interesting for analysis
34
+ if log_outputs is not None:
35
+ pred_file = f"log_{dataset_id}_predictions.txt"
36
+ target_file = f"log_{dataset_id}_targets.txt"
37
+
38
+ with open(pred_file, "w") as p, open(target_file, "w") as t:
39
+
40
+ # mapping function to write output
41
+ def write_to_file(batch, i):
42
+ p.write(f"{i}" + "\n")
43
+ p.write(batch["prediction"] + "\n")
44
+ t.write(f"{i}" + "\n")
45
+ t.write(batch["target"] + "\n")
46
+
47
+ result.map(write_to_file, with_indices=True)
48
+
49
+
50
+ def normalize_text(text: str) -> str:
51
+ """DO ADAPT FOR YOUR USE CASE. this function normalizes the target text."""
52
+
53
+ CHARS_TO_IGNORE = [",", "?", "¿", ".", "!", "¡", ";", ";", ":", '""', "%", '"', "�", "ʿ", "·", "჻", "~", "՞",
54
+ "؟", "،", "।", "॥", "«", "»", "„", "“", "”", "「", "」", "‘", "’", "《", "》", "(", ")", "[", "]",
55
+ "{", "}", "=", "`", "_", "+", "<", ">", "…", "–", "°", "´", "ʾ", "‹", "›", "©", "®", "—", "→", "。",
56
+ "、", "﹂", "﹁", "‧", "~", "﹏", ",", "{", "}", "(", ")", "[", "]", "【", "】", "‥", "〽",
57
+ "『", "』", "〝", "〟", "⟨", "⟩", "〜", ":", "!", "?", "♪", "؛", "/", "\\", "º", "−", "^", "ʻ", "ˆ", "'"]
58
+
59
+ chars_to_remove_regex = f"[{re.escape(''.join(CHARS_TO_IGNORE))}]"
60
+
61
+ text = re.sub(chars_to_remove_regex, "", text.lower())
62
+ text = re.sub("[-]", " ", text)
63
+
64
+ # In addition, we can normalize the target text, e.g. removing new lines characters etc...
65
+ # note that order is important here!
66
+ token_sequences_to_ignore = ["\n\n", "\n", " ", " "]
67
+
68
+ for t in token_sequences_to_ignore:
69
+ text = " ".join(text.split(t))
70
+
71
+ return text
72
+
73
+
74
+ def main(args):
75
+ # load dataset
76
+ dataset = load_dataset(args.dataset, args.config, split=args.split, use_auth_token=True)
77
+
78
+ # for testing: only process the first two examples as a test
79
+ # dataset = dataset.select(range(10))
80
+
81
+ # load processor
82
+ feature_extractor = AutoFeatureExtractor.from_pretrained(args.model_id)
83
+ sampling_rate = feature_extractor.sampling_rate
84
+
85
+ # resample audio
86
+ dataset = dataset.cast_column("audio", Audio(sampling_rate=sampling_rate))
87
+
88
+ # load eval pipeline
89
+ if args.device is None:
90
+ args.device = 0 if torch.cuda.is_available() else -1
91
+ asr = pipeline("automatic-speech-recognition", model=args.model_id, device=args.device)
92
+
93
+ # map function to decode audio
94
+ def map_to_pred(batch):
95
+ prediction = asr(
96
+ batch["audio"]["array"], chunk_length_s=args.chunk_length_s, stride_length_s=args.stride_length_s
97
+ )
98
+
99
+ batch["prediction"] = prediction["text"]
100
+ batch["target"] = normalize_text(batch["sentence"])
101
+ return batch
102
+
103
+ # run inference on all examples
104
+ result = dataset.map(map_to_pred, remove_columns=dataset.column_names)
105
+
106
+ # compute and log_results
107
+ # do not change function below
108
+ log_results(result, args)
109
+
110
+
111
+ if __name__ == "__main__":
112
+ parser = argparse.ArgumentParser()
113
+
114
+ parser.add_argument(
115
+ "--model_id", type=str, required=True, help="Model identifier. Should be loadable with 🤗 Transformers"
116
+ )
117
+ parser.add_argument(
118
+ "--dataset",
119
+ type=str,
120
+ required=True,
121
+ help="Dataset name to evaluate the `model_id`. Should be loadable with 🤗 Datasets",
122
+ )
123
+ parser.add_argument(
124
+ "--config", type=str, required=True, help="Config of the dataset. *E.g.* `'en'` for Common Voice"
125
+ )
126
+ parser.add_argument("--split", type=str, required=True, help="Split of the dataset. *E.g.* `'test'`")
127
+ parser.add_argument(
128
+ "--chunk_length_s", type=float, default=None, help="Chunk length in seconds. Defaults to 5 seconds."
129
+ )
130
+ parser.add_argument(
131
+ "--stride_length_s", type=float, default=None, help="Stride of the audio chunks. Defaults to 1 second."
132
+ )
133
+ parser.add_argument(
134
+ "--log_outputs", action="store_true", help="If defined, write outputs to log file for analysis."
135
+ )
136
+ parser.add_argument(
137
+ "--device",
138
+ type=int,
139
+ default=None,
140
+ help="The device to run the pipeline on. -1 for CPU (default), 0 for the first GPU and so on.",
141
+ )
142
+ args = parser.parse_args()
143
+
144
+ main(args)
language_model/attrs.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"alpha": 0.5, "beta": 1.5, "unk_score_offset": -10.0, "score_boundary": true}
language_model/kenlm_finnish.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:79ca983c89af32b5c38560d822324bf2023c3aa2b32ed17fd1ea67aac68b5166
3
+ size 1021613027
language_model/unigrams.txt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c0f65403ece43fcc9eb95148060e402d7af46692f7a41a94b75ae23f40fa93d7
3
+ size 14815049
preprocessor_config.json CHANGED
@@ -4,6 +4,7 @@
4
  "feature_size": 1,
5
  "padding_side": "right",
6
  "padding_value": 0.0,
 
7
  "return_attention_mask": true,
8
  "sampling_rate": 16000
9
  }
 
4
  "feature_size": 1,
5
  "padding_side": "right",
6
  "padding_value": 0.0,
7
+ "processor_class": "Wav2Vec2ProcessorWithLM",
8
  "return_attention_mask": true,
9
  "sampling_rate": 16000
10
  }
special_tokens_map.json CHANGED
@@ -1 +1 @@
1
- {"bos_token": "<s>", "eos_token": "</s>", "unk_token": "[UNK]", "pad_token": "[PAD]", "additional_special_tokens": [{"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}]}
 
1
+ {"bos_token": "<s>", "eos_token": "</s>", "unk_token": "[UNK]", "pad_token": "[PAD]", "additional_special_tokens": [{"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, {"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}]}
tokenizer_config.json CHANGED
@@ -1 +1 @@
1
- {"unk_token": "[UNK]", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "[PAD]", "do_lower_case": false, "word_delimiter_token": "|", "replace_word_delimiter_char": " ", "special_tokens_map_file": null, "name_or_path": "./", "tokenizer_class": "Wav2Vec2CTCTokenizer"}
 
1
+ {"unk_token": "[UNK]", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "[PAD]", "do_lower_case": false, "word_delimiter_token": "|", "replace_word_delimiter_char": " ", "special_tokens_map_file": null, "name_or_path": "/content/wav2vec2-base-fi-voxpopuli-v2-finetuned", "tokenizer_class": "Wav2Vec2CTCTokenizer", "processor_class": "Wav2Vec2ProcessorWithLM"}