Update README.md
Browse files
README.md
CHANGED
@@ -74,7 +74,7 @@ The model can be evaluated as follows on the zh-CN test data of Common Voice.
|
|
74 |
Original CER calculation refer to https://huggingface.co/ctl/wav2vec2-large-xlsr-cantonese
|
75 |
|
76 |
```python
|
77 |
-
|
78 |
|
79 |
import torch
|
80 |
import torchaudio
|
@@ -114,7 +114,7 @@ processor = Wav2Vec2Processor.from_pretrained("ydshieh/wav2vec2-large-xlsr-53-ch
|
|
114 |
model = Wav2Vec2ForCTC.from_pretrained("ydshieh/wav2vec2-large-xlsr-53-chinese-zh-cn-gpt")
|
115 |
model.to("cuda")
|
116 |
|
117 |
-
chars_to_ignore_regex = '[
|
118 |
|
119 |
resampler = torchaudio.transforms.Resample(48_000, 16_000)
|
120 |
|
|
|
74 |
Original CER calculation refer to https://huggingface.co/ctl/wav2vec2-large-xlsr-cantonese
|
75 |
|
76 |
```python
|
77 |
+
# pip install jiwer
|
78 |
|
79 |
import torch
|
80 |
import torchaudio
|
|
|
114 |
model = Wav2Vec2ForCTC.from_pretrained("ydshieh/wav2vec2-large-xlsr-53-chinese-zh-cn-gpt")
|
115 |
model.to("cuda")
|
116 |
|
117 |
+
chars_to_ignore_regex = '[\\\\\\\\,\\\\\\\\?\\\\\\\\.\\\\\\\\!\\\\\\\\-\\\\\\\\;\\\\\\\\:"\\\\\\\\“\\\\\\\\%\\\\\\\\‘\\\\\\\\”\\\\\\\\�\\\\\\\\.\\\\\\\\⋯\\\\\\\\!\\\\\\\\-\\\\\\\\:\\\\\\\\–\\\\\\\\。\\\\\\\\》\\\\\\\\,\\\\\\\\)\\\\\\\\,\\\\\\\\?\\\\\\\\;\\\\\\\\~\\\\\\\\~\\\\\\\\…\\\\\\\\︰\\\\\\\\,\\\\\\\\(\\\\\\\\」\\\\\\\\‧\\\\\\\\《\\\\\\\\﹔\\\\\\\\、\\\\\\\\—\\\\\\\\/\\\\\\\\,\\\\\\\\「\\\\\\\\﹖\\\\\\\\·\\\\\\\\×\\\\\\\\̃\\\\\\\\̌\\\\\\\\ε\\\\\\\\λ\\\\\\\\μ\\\\\\\\и\\\\\\\\т\\\\\\\\─\\\\\\\\□\\\\\\\\〈\\\\\\\\〉\\\\\\\\『\\\\\\\\』\\\\\\\\ア\\\\\\\\オ\\\\\\\\カ\\\\\\\\チ\\\\\\\\ド\\\\\\\\ベ\\\\\\\\ャ\\\\\\\\ヤ\\\\\\\\ン\\\\\\\\・\\\\\\\\丶\\\\\\\\a\\\\\\\\b\\\\\\\\f\\\\\\\\g\\\\\\\\i\\\\\\\\n\\\\\\\\p\\\\\\\\t' + "\\\\\\\\']"
|
118 |
|
119 |
resampler = torchaudio.transforms.Resample(48_000, 16_000)
|
120 |
|