Rubens commited on
Commit
9471b89
1 Parent(s): f14e791

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +15 -15
README.md CHANGED
@@ -57,15 +57,15 @@ resampler = torchaudio.transforms.Resample(48_000, 16_000)
57
  # Preprocessing the datasets.
58
  # We need to read the audio files as arrays
59
  def speech_file_to_array_fn(batch):
60
- speech_array, sampling_rate = torchaudio.load(batch["path"])
61
- batch["speech"] = resampler(speech_array).squeeze().numpy()
62
- return batch
63
 
64
  test_dataset = test_dataset.map(speech_file_to_array_fn)
65
  inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
66
 
67
  with torch.no_grad():
68
- logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
69
 
70
  predicted_ids = torch.argmax(logits, dim=-1)
71
 
@@ -93,37 +93,37 @@ processor = Wav2Vec2Processor.from_pretrained("Rubens/Wav2Vec2-Large-XLSR-53-a-P
93
  model = Wav2Vec2ForCTC.from_pretrained("Rubens/Wav2Vec2-Large-XLSR-53-a-Portuguese")
94
  model.to("cuda")
95
 
96
- chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“]' # TODO: adapt this list to include all special characters you removed from the data
97
  resampler = torchaudio.transforms.Resample(48_000, 16_000)
98
 
99
  # Preprocessing the datasets.
100
  # We need to read the aduio files as arrays
101
  def speech_file_to_array_fn(batch):
102
- batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
103
- speech_array, sampling_rate = torchaudio.load(batch["path"])
104
- batch["speech"] = resampler(speech_array).squeeze().numpy()
105
- return batch
106
 
107
  test_dataset = test_dataset.map(speech_file_to_array_fn)
108
 
109
  # Preprocessing the datasets.
110
  # We need to read the aduio files as arrays
111
  def evaluate(batch):
112
- inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
113
 
114
- with torch.no_grad():
115
- logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
116
 
117
  pred_ids = torch.argmax(logits, dim=-1)
118
- batch["pred_strings"] = processor.batch_decode(pred_ids)
119
- return batch
120
 
121
  result = test_dataset.map(evaluate, batched=True, batch_size=8)
122
 
123
  print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
124
  ```
125
 
126
- **Test Result (wer) **: 19.30 %
127
 
128
 
129
  ## Training
 
57
  # Preprocessing the datasets.
58
  # We need to read the audio files as arrays
59
  def speech_file_to_array_fn(batch):
60
+ \tspeech_array, sampling_rate = torchaudio.load(batch["path"])
61
+ \tbatch["speech"] = resampler(speech_array).squeeze().numpy()
62
+ \treturn batch
63
 
64
  test_dataset = test_dataset.map(speech_file_to_array_fn)
65
  inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
66
 
67
  with torch.no_grad():
68
+ \tlogits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
69
 
70
  predicted_ids = torch.argmax(logits, dim=-1)
71
 
 
93
  model = Wav2Vec2ForCTC.from_pretrained("Rubens/Wav2Vec2-Large-XLSR-53-a-Portuguese")
94
  model.to("cuda")
95
 
96
+ chars_to_ignore_regex = '[\\,\\?\\.\\!\\-\\;\\:\\"\\“]' # TODO: adapt this list to include all special characters you removed from the data
97
  resampler = torchaudio.transforms.Resample(48_000, 16_000)
98
 
99
  # Preprocessing the datasets.
100
  # We need to read the aduio files as arrays
101
  def speech_file_to_array_fn(batch):
102
+ \tbatch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
103
+ \tspeech_array, sampling_rate = torchaudio.load(batch["path"])
104
+ \tbatch["speech"] = resampler(speech_array).squeeze().numpy()
105
+ \treturn batch
106
 
107
  test_dataset = test_dataset.map(speech_file_to_array_fn)
108
 
109
  # Preprocessing the datasets.
110
  # We need to read the aduio files as arrays
111
  def evaluate(batch):
112
+ \tinputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
113
 
114
+ \twith torch.no_grad():
115
+ \t\tlogits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
116
 
117
  pred_ids = torch.argmax(logits, dim=-1)
118
+ \tbatch["pred_strings"] = processor.batch_decode(pred_ids)
119
+ \treturn batch
120
 
121
  result = test_dataset.map(evaluate, batched=True, batch_size=8)
122
 
123
  print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
124
  ```
125
 
126
+ **Test Result(wer)**: 19.30 %
127
 
128
 
129
  ## Training