Plim commited on
Commit
4898777
1 Parent(s): 66bd9dd

add results on dev audio with step 24000

Browse files
README.md CHANGED
@@ -20,10 +20,10 @@ model-index:
20
  metrics:
21
  - name: Test WER
22
  type: wer
23
- value: 21.65
24
  - name: Test CER
25
  type: cer
26
- value: 6.52
27
  - task:
28
  name: Automatic Speech Recognition
29
  type: automatic-speech-recognition
@@ -34,10 +34,11 @@ model-index:
34
  metrics:
35
  - name: Test WER
36
  type: wer
37
- value: 61.72
38
  - name: Test CER
39
  type: cer
40
- value: 16.43
 
41
  ---
42
 
43
  ## Model description
@@ -83,8 +84,18 @@ The following hyperparameters were used during training:
83
  | 0.8488 | 4.59 | 16000 | inf | 0.2187 |
84
  | 0.8359 | 4.87 | 17000 | inf | 0.2172 |
85
 
86
- It achieves the best result on the validation set on Step 17000:
87
- - Wer: 0.2172
 
 
 
 
 
 
 
 
 
 
88
 
89
  Got some issue with validation loss calculation.
90
 
 
20
  metrics:
21
  - name: Test WER
22
  type: wer
23
+ value: to recompute with STEP 24000
24
  - name: Test CER
25
  type: cer
26
+ value: to recompute with STEP 24000
27
  - task:
28
  name: Automatic Speech Recognition
29
  type: automatic-speech-recognition
 
34
  metrics:
35
  - name: Test WER
36
  type: wer
37
+ value: 35.29
38
  - name: Test CER
39
  type: cer
40
+ value: 13.94
41
+
42
  ---
43
 
44
  ## Model description
 
84
  | 0.8488 | 4.59 | 16000 | inf | 0.2187 |
85
  | 0.8359 | 4.87 | 17000 | inf | 0.2172 |
86
 
87
+ Training continued with checkpoint from STEP 17000:
88
+ | / | 5.16 | 18000 | inf | 0.2176 |
89
+ | / | 5.45 | 19000 | inf | 0.2181 |
90
+ | / | 5.73 | 20000 | inf | 0.2155 |
91
+ | / | 6.02 | 21000 | inf | 0.2140 |
92
+ | / | 6.31 | 22000 | inf | 0.2124 |
93
+ | / | 6.59 | 23000 | inf | 0.2117 |
94
+ | / | 6.88 | 24000 | inf | 0.2116 |
95
+
96
+
97
+ It achieves the best result on the validation set on Step 24000:
98
+ - Wer: 0.2116
99
 
100
  Got some issue with validation loss calculation.
101
 
eval.py CHANGED
@@ -48,18 +48,15 @@ def log_results(result: Dataset, args: Dict[str, str]):
48
 
49
  def normalize_text(text: str) -> str:
50
  """DO ADAPT FOR YOUR USE CASE. this function normalizes the target text."""
51
-
52
- chars_to_ignore_regex = '[^a-zàâäçéèêëîïôöùûüÿ\'’ ]' # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
53
-
54
- text = re.sub(chars_to_ignore_regex, "", text.lower()).replace('’', "'")
55
-
56
  # In addition, we can normalize the target text, e.g. removing new lines characters etc...
57
  # note that order is important here!
58
  token_sequences_to_ignore = ["\n\n", "\n", " ", " "]
59
-
60
  for t in token_sequences_to_ignore:
61
  text = " ".join(text.split(t))
62
 
 
 
 
63
  return text
64
 
65
 
@@ -68,7 +65,7 @@ def main(args):
68
  dataset = load_dataset(args.dataset, args.config, split=args.split, use_auth_token=True)
69
 
70
  # for testing: only process the first two examples as a test
71
- # dataset = dataset.select(range(10))
72
 
73
  # load processor
74
  feature_extractor = AutoFeatureExtractor.from_pretrained(args.model_id)
 
48
 
49
  def normalize_text(text: str) -> str:
50
  """DO ADAPT FOR YOUR USE CASE. this function normalizes the target text."""
 
 
 
 
 
51
  # In addition, we can normalize the target text, e.g. removing new lines characters etc...
52
  # note that order is important here!
53
  token_sequences_to_ignore = ["\n\n", "\n", " ", " "]
 
54
  for t in token_sequences_to_ignore:
55
  text = " ".join(text.split(t))
56
 
57
+ chars_to_ignore_regex = '[^a-zàâäçéèêëîïôöùûüÿ\'’ ]' # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
58
+ text = re.sub(chars_to_ignore_regex, "", text.lower()).replace('’', "'")
59
+
60
  return text
61
 
62
 
 
65
  dataset = load_dataset(args.dataset, args.config, split=args.split, use_auth_token=True)
66
 
67
  # for testing: only process the first two examples as a test
68
+ # dataset = dataset.select(range(2))
69
 
70
  # load processor
71
  feature_extractor = AutoFeatureExtractor.from_pretrained(args.model_id)
log_speech-recognition-community-v2_dev_data_fr_validation_predictions.txt CHANGED
The diff for this file is too large to render. See raw diff
 
log_speech-recognition-community-v2_dev_data_fr_validation_targets.txt CHANGED
The diff for this file is too large to render. See raw diff
 
speech-recognition-community-v2_dev_data_fr_validation_eval_results.txt CHANGED
@@ -1,2 +1,2 @@
1
- WER: 0.617242860210436
2
- CER: 0.16435482455790507
 
1
+ WER: 0.35289081159028435
2
+ CER: 0.1394068190984395