birgermoell commited on
Commit
f0dc14d
1 Parent(s): 3fcd3d1

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +19 -18
README.md CHANGED
@@ -24,7 +24,7 @@ model-index:
24
  value: ???
25
  ---
26
 
27
- # Wav2Vec2-Large-XLSR-53-Hungarian
28
 
29
  Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) in Finnish using the [Common Voice](https://huggingface.co/datasets/common_voice)
30
  When using this model, make sure that your speech input is sampled at 16kHz.
@@ -49,15 +49,15 @@ resampler = torchaudio.transforms.Resample(48_000, 16_000)
49
  # Preprocessing the datasets.
50
  # We need to read the aduio files as arrays
51
  def speech_file_to_array_fn(batch):
52
- speech_array, sampling_rate = torchaudio.load(batch["path"])
53
- batch["speech"] = resampler(speech_array).squeeze().numpy()
54
- return batch
55
 
56
  test_dataset = test_dataset.map(speech_file_to_array_fn)
57
  inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
58
 
59
  with torch.no_grad():
60
- logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
61
 
62
  predicted_ids = torch.argmax(logits, dim=-1)
63
 
@@ -85,41 +85,42 @@ processor = Wav2Vec2Processor.from_pretrained("birgermoell/wav2vec2-large-xlsr-f
85
  model = Wav2Vec2ForCTC.from_pretrained("birgermoell/wav2vec2-large-xlsr-finnish")
86
  model.to("cuda")
87
 
88
- chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“]'
89
  resampler = torchaudio.transforms.Resample(48_000, 16_000)
90
 
91
  # Preprocessing the datasets.
92
  # We need to read the aduio files as arrays
93
  def speech_file_to_array_fn(batch):
94
- batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
95
- speech_array, sampling_rate = torchaudio.load(batch["path"])
96
- batch["speech"] = resampler(speech_array).squeeze().numpy()
97
- return batch
98
 
99
  test_dataset = test_dataset.map(speech_file_to_array_fn)
100
 
101
  # Preprocessing the datasets.
102
  # We need to read the aduio files as arrays
103
  def evaluate(batch):
104
- inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
105
 
106
- with torch.no_grad():
107
- logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
108
  pred_ids = torch.argmax(logits, dim=-1)
109
 
110
- batch["pred_strings"] = processor.batch_decode(pred_ids)
111
- return batch
112
 
113
  result = test_dataset.map(evaluate, batched=True, batch_size=8)
114
 
115
  print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
116
  ```
117
 
118
- **Test Result**: ???
119
-
120
 
121
  ## Training
122
 
123
  The Common Voice `train` and `validation` datasets were used for training.
124
- The script used for training can be found ???
 
125
 
 
24
  value: ???
25
  ---
26
 
27
+ # Wav2Vec2-Large-XLSR-53-Finnish
28
 
29
  Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) in Finnish using the [Common Voice](https://huggingface.co/datasets/common_voice)
30
  When using this model, make sure that your speech input is sampled at 16kHz.
 
49
  # Preprocessing the datasets.
50
  # We need to read the aduio files as arrays
51
  def speech_file_to_array_fn(batch):
52
+ \tspeech_array, sampling_rate = torchaudio.load(batch["path"])
53
+ \tbatch["speech"] = resampler(speech_array).squeeze().numpy()
54
+ \treturn batch
55
 
56
  test_dataset = test_dataset.map(speech_file_to_array_fn)
57
  inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
58
 
59
  with torch.no_grad():
60
+ \tlogits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
61
 
62
  predicted_ids = torch.argmax(logits, dim=-1)
63
 
 
85
  model = Wav2Vec2ForCTC.from_pretrained("birgermoell/wav2vec2-large-xlsr-finnish")
86
  model.to("cuda")
87
 
88
+ chars_to_ignore_regex = '[\\,\\?\\.\\!\\-\\;\\:\\"\\“]'
89
  resampler = torchaudio.transforms.Resample(48_000, 16_000)
90
 
91
  # Preprocessing the datasets.
92
  # We need to read the aduio files as arrays
93
  def speech_file_to_array_fn(batch):
94
+ \tbatch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
95
+ \tspeech_array, sampling_rate = torchaudio.load(batch["path"])
96
+ \tbatch["speech"] = resampler(speech_array).squeeze().numpy()
97
+ \treturn batch
98
 
99
  test_dataset = test_dataset.map(speech_file_to_array_fn)
100
 
101
  # Preprocessing the datasets.
102
  # We need to read the aduio files as arrays
103
  def evaluate(batch):
104
+ \tinputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
105
 
106
+ \twith torch.no_grad():
107
+ \t\tlogits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
108
  pred_ids = torch.argmax(logits, dim=-1)
109
 
110
+ \tbatch["pred_strings"] = processor.batch_decode(pred_ids)
111
+ \treturn batch
112
 
113
  result = test_dataset.map(evaluate, batched=True, batch_size=8)
114
 
115
  print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
116
  ```
117
 
118
+ **Test Result**:
119
+ The WER is 55.097365
120
 
121
  ## Training
122
 
123
  The Common Voice `train` and `validation` datasets were used for training.
124
+ The script used for training can be found here
125
+ https://colab.research.google.com/drive/16AyzqMWU_aWNe3IA-NxrhskB1WLPHG-Q?usp=sharing
126