ozcangundes commited on
Commit
4f39b79
1 Parent(s): e1fd5f6

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +7 -7
README.md CHANGED
@@ -24,7 +24,7 @@ model-index:
24
  metrics:
25
  - name: Test WER
26
  type: wer
27
- value: 30.24
28
  ---
29
 
30
  # Wav2Vec2-Large-XLSR-53-Turkish
@@ -51,15 +51,15 @@ resampler = torchaudio.transforms.Resample(48_000, 16_000)
51
  # Preprocessing the datasets.
52
  # We need to read the aduio files as arrays
53
  def speech_file_to_array_fn(batch):
54
- \tspeech_array, sampling_rate = torchaudio.load(batch["path"])
55
- \tbatch["speech"] = resampler(speech_array).squeeze().numpy()
56
- \treturn batch
57
 
58
  test_dataset = test_dataset.map(speech_file_to_array_fn)
59
  inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
60
 
61
  with torch.no_grad():
62
- \tlogits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
63
 
64
  predicted_ids = torch.argmax(logits, dim=-1)
65
 
@@ -88,7 +88,7 @@ model = Wav2Vec2ForCTC.from_pretrained("ozcangundes/wav2vec2-large-xlsr-53-turki
88
 
89
  model.to("cuda")
90
 
91
- chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“\%\‘\”\�\’\']'
92
  resampler = torchaudio.transforms.Resample(48_000, 16_000)
93
 
94
  # Preprocessing the datasets.
@@ -118,7 +118,7 @@ result = test_dataset.map(evaluate, batched=True, batch_size=8)
118
  print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
119
  ```
120
 
121
- **Test Result**: 30.24 %
122
 
123
  ## Training
124
 
 
24
  metrics:
25
  - name: Test WER
26
  type: wer
27
+ value: 29.62
28
  ---
29
 
30
  # Wav2Vec2-Large-XLSR-53-Turkish
 
51
  # Preprocessing the datasets.
52
  # We need to read the aduio files as arrays
53
  def speech_file_to_array_fn(batch):
54
+ \\tspeech_array, sampling_rate = torchaudio.load(batch["path"])
55
+ \\tbatch["speech"] = resampler(speech_array).squeeze().numpy()
56
+ \\treturn batch
57
 
58
  test_dataset = test_dataset.map(speech_file_to_array_fn)
59
  inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
60
 
61
  with torch.no_grad():
62
+ \\tlogits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
63
 
64
  predicted_ids = torch.argmax(logits, dim=-1)
65
 
 
88
 
89
  model.to("cuda")
90
 
91
+ chars_to_ignore_regex = '[\\,\\?\\.\\!\\-\\;\\:\\"\\“\\%\\‘\\”\\�\\’\\']'
92
  resampler = torchaudio.transforms.Resample(48_000, 16_000)
93
 
94
  # Preprocessing the datasets.
 
118
  print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
119
  ```
120
 
121
+ **Test Result**: 29.62 %
122
 
123
  ## Training
124