versae commited on
Commit
f625488
1 Parent(s): 854e92c

Add 5gram lang model

Browse files
.gitattributes CHANGED
@@ -26,3 +26,4 @@ saved_model/**/* 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
  wandb/run-20220201_153024-1w85vsuu/run-1w85vsuu.wandb 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
  wandb/run-20220201_153024-1w85vsuu/run-1w85vsuu.wandb filter=lfs diff=lfs merge=lfs -text
29
+ language_model/unigrams.txt filter=lfs diff=lfs merge=lfs -text
add_kenlm.py ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import argparse
2
+ from transformers import AutoProcessor
3
+ from transformers import Wav2Vec2ProcessorWithLM
4
+ from pyctcdecode import build_ctcdecoder
5
+
6
+
7
+ def main(args):
8
+ processor = AutoProcessor.from_pretrained(args.model_name_or_path)
9
+ vocab_dict = processor.tokenizer.get_vocab()
10
+ sorted_vocab_dict = {
11
+ k.lower(): v for k, v in sorted(vocab_dict.items(), key=lambda item: item[1])
12
+ }
13
+ decoder = build_ctcdecoder(
14
+ labels=list(sorted_vocab_dict.keys()),
15
+ kenlm_model_path=args.kenlm_model_path,
16
+ )
17
+ processor_with_lm = Wav2Vec2ProcessorWithLM(
18
+ feature_extractor=processor.feature_extractor,
19
+ tokenizer=processor.tokenizer,
20
+ decoder=decoder,
21
+ )
22
+ processor_with_lm.save_pretrained(args.model_name_or_path)
23
+ print(f"Run: ~/bin/build_binary language_model/*.arpa language_model/5gram.bin -T $(pwd) && rm language_model/*.arpa")
24
+
25
+ def parse_args():
26
+ parser = argparse.ArgumentParser()
27
+ parser.add_argument('--model_name_or_path', default="./", help='Model name or path. Defaults to ./')
28
+ parser.add_argument('--kenlm_model_path', required=True, help='Path to KenLM arpa file.')
29
+ args = parser.parse_args()
30
+ return args
31
+
32
+ if __name__ == "__main__":
33
+ args = parse_args()
34
+ main(args)
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", "\u00e5", "\u00e6", "\u00f8", "\u2047", "", "<s>", "</s>"], "is_bpe": false}
eval.py CHANGED
@@ -6,7 +6,7 @@ from typing import Dict
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]):
@@ -96,7 +96,12 @@ def main(args):
96
  # load eval pipeline
97
  if args.device is None:
98
  args.device = 0 if torch.cuda.is_available() else -1
99
- asr = pipeline("automatic-speech-recognition", model=args.model_id, device=args.device)
 
 
 
 
 
100
 
101
  # map function to decode audio
102
  def map_to_pred(batch):
 
6
  import torch
7
  from datasets import Audio, Dataset, load_dataset, load_metric
8
 
9
+ from transformers import AutoFeatureExtractor, AutoModel, AutoTokenizer, pipeline
10
 
11
 
12
  def log_results(result: Dataset, args: Dict[str, str]):
 
96
  # load eval pipeline
97
  if args.device is None:
98
  args.device = 0 if torch.cuda.is_available() else -1
99
+ asr = pipeline("automatic-speech-recognition",
100
+ model=AutoModel.from_pretrained(args.model_id),
101
+ tokenizer=AutoTokenizer.from_pretrained(args.model_id),
102
+ feature_extractor=feature_extractor,
103
+ device=args.device
104
+ )
105
 
106
  # map function to decode audio
107
  def map_to_pred(batch):
language_model/5gram.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7b41c24c63f2f0585bea83666369593f3b3e6d047f327a90f36ebca2c35ef0ff
3
+ size 4243671427
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/unigrams.txt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ac3e71ca49838ca355df6fdcb8d89344a5a9bf9e1a76587cdf5df1367c19b9a9
3
+ size 16759269
preprocessor_config.json CHANGED
@@ -4,6 +4,7 @@
4
  "feature_size": 1,
5
  "padding_side": "right",
6
  "padding_value": 0,
 
7
  "return_attention_mask": true,
8
  "sampling_rate": 16000
9
  }
 
4
  "feature_size": 1,
5
  "padding_side": "right",
6
  "padding_value": 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}, {"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}, {"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": "|", "special_tokens_map_file": null, "tokenizer_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": "|", "special_tokens_map_file": null, "tokenizer_file": null, "name_or_path": "./", "tokenizer_class": "Wav2Vec2CTCTokenizer", "processor_class": "Wav2Vec2ProcessorWithLM"}