anuragshas commited on
Commit
93d20e7
β€’
1 Parent(s): b897e94

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +16 -16
README.md CHANGED
@@ -42,19 +42,19 @@ resampler = torchaudio.transforms.Resample(48_000, 16_000)
42
  # Preprocessing the datasets.
43
  # We need to read the aduio files as arrays
44
  def speech_file_to_array_fn(batch):
45
- \tspeech_array, sampling_rate = torchaudio.load(batch["path"])
46
- \tbatch["speech"] = resampler(speech_array).squeeze().numpy()
47
- \treturn batch
48
  test_dataset = test_dataset.map(speech_file_to_array_fn)
49
  inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
50
  with torch.no_grad():
51
- \tlogits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
52
  predicted_ids = torch.argmax(logits, dim=-1)
53
  print("Prediction:", processor.batch_decode(predicted_ids))
54
  print("Reference:", test_dataset["sentence"][:2])
55
  ```
56
  ## Evaluation
57
- The model can be evaluated as follows on the Odia test data of Common Voice.
58
  ```python
59
  import torch
60
  import torchaudio
@@ -66,25 +66,25 @@ wer = load_metric("wer")
66
  processor = Wav2Vec2Processor.from_pretrained("anuragshas/wav2vec2-xlsr-53-pa-in")
67
  model = Wav2Vec2ForCTC.from_pretrained("anuragshas/wav2vec2-xlsr-53-pa-in")
68
  model.to("cuda")
69
- chars_to_ignore_regex = '[\\,\\?\\.\\!\\-\\;\\:\\\"\\β€œ\\%\\β€˜\\”\\ΰ₯€\\’\'\\…]'
70
  resampler = torchaudio.transforms.Resample(48_000, 16_000)
71
  # Preprocessing the datasets.
72
  # We need to read the aduio files as arrays
73
  def speech_file_to_array_fn(batch):
74
- \tbatch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
75
- \tspeech_array, sampling_rate = torchaudio.load(batch["path"])
76
- \tbatch["speech"] = resampler(speech_array).squeeze().numpy()
77
- \treturn batch
78
  test_dataset = test_dataset.map(speech_file_to_array_fn)
79
  # Preprocessing the datasets.
80
  # We need to read the aduio files as arrays
81
  def evaluate(batch):
82
- \tinputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
83
- \twith torch.no_grad():
84
- \t\tlogits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
85
- \tpred_ids = torch.argmax(logits, dim=-1)
86
- \tbatch["pred_strings"] = processor.batch_decode(pred_ids)
87
- \treturn batch
88
  result = test_dataset.map(evaluate, batched=True, batch_size=8)
89
  print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
90
  ```
 
42
  # Preprocessing the datasets.
43
  # We need to read the aduio files as arrays
44
  def speech_file_to_array_fn(batch):
45
+ speech_array, sampling_rate = torchaudio.load(batch["path"])
46
+ batch["speech"] = resampler(speech_array).squeeze().numpy()
47
+ return batch
48
  test_dataset = test_dataset.map(speech_file_to_array_fn)
49
  inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
50
  with torch.no_grad():
51
+ logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
52
  predicted_ids = torch.argmax(logits, dim=-1)
53
  print("Prediction:", processor.batch_decode(predicted_ids))
54
  print("Reference:", test_dataset["sentence"][:2])
55
  ```
56
  ## Evaluation
57
+ The model can be evaluated as follows on the Punjabi test data of Common Voice.
58
  ```python
59
  import torch
60
  import torchaudio
 
66
  processor = Wav2Vec2Processor.from_pretrained("anuragshas/wav2vec2-xlsr-53-pa-in")
67
  model = Wav2Vec2ForCTC.from_pretrained("anuragshas/wav2vec2-xlsr-53-pa-in")
68
  model.to("cuda")
69
+ chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\β€œ\%\β€˜\”\ΰ₯€\’\'\…]'
70
  resampler = torchaudio.transforms.Resample(48_000, 16_000)
71
  # Preprocessing the datasets.
72
  # We need to read the aduio files as arrays
73
  def speech_file_to_array_fn(batch):
74
+ batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
75
+ speech_array, sampling_rate = torchaudio.load(batch["path"])
76
+ batch["speech"] = resampler(speech_array).squeeze().numpy()
77
+ return batch
78
  test_dataset = test_dataset.map(speech_file_to_array_fn)
79
  # Preprocessing the datasets.
80
  # We need to read the aduio files as arrays
81
  def evaluate(batch):
82
+ inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
83
+ with torch.no_grad():
84
+ logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
85
+ pred_ids = torch.argmax(logits, dim=-1)
86
+ batch["pred_strings"] = processor.batch_decode(pred_ids)
87
+ return batch
88
  result = test_dataset.map(evaluate, batched=True, batch_size=8)
89
  print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
90
  ```