Iskaj commited on
Commit
e4f4d3d
1 Parent(s): 49bb107

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +1 -63
README.md CHANGED
@@ -1,63 +1 @@
1
- ---
2
- language: nl
3
- datasets:
4
- - common_voice
5
- tags:
6
- - speech
7
- - audio
8
- - automatic-speech-recognition
9
- license: apache-2.0
10
- ---
11
-
12
- ## Evaluation on Common Voice NL Test
13
-
14
- ```python
15
- import torchaudio
16
- from datasets import load_dataset, load_metric
17
- from transformers import (
18
- Wav2Vec2ForCTC,
19
- Wav2Vec2Processor,
20
- )
21
- import torch
22
- import re
23
- import sys
24
-
25
- model_name = "facebook/wav2vec2-large-xlsr-53-dutch"
26
- device = "cuda"
27
- chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"]' # noqa: W605
28
-
29
- model = Wav2Vec2ForCTC.from_pretrained(model_name).to(device)
30
- processor = Wav2Vec2Processor.from_pretrained(model_name)
31
-
32
- ds = load_dataset("common_voice", "nl", split="test", data_dir="./cv-corpus-6.1-2020-12-11")
33
-
34
- resampler = torchaudio.transforms.Resample(orig_freq=48_000, new_freq=16_000)
35
-
36
- def map_to_array(batch):
37
- speech, _ = torchaudio.load(batch["path"])
38
- batch["speech"] = resampler.forward(speech.squeeze(0)).numpy()
39
- batch["sampling_rate"] = resampler.new_freq
40
- batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower().replace("’", "'")
41
- return batch
42
-
43
- ds = ds.map(map_to_array)
44
-
45
-
46
- def map_to_pred(batch):
47
- features = processor(batch["speech"], sampling_rate=batch["sampling_rate"][0], padding=True, return_tensors="pt")
48
- input_values = features.input_values.to(device)
49
- attention_mask = features.attention_mask.to(device)
50
- with torch.no_grad():
51
- logits = model(input_values, attention_mask=attention_mask).logits
52
- pred_ids = torch.argmax(logits, dim=-1)
53
- batch["predicted"] = processor.batch_decode(pred_ids)
54
- batch["target"] = batch["sentence"]
55
- return batch
56
-
57
- result = ds.map(map_to_pred, batched=True, batch_size=16, remove_columns=list(ds.features.keys()))
58
-
59
- wer = load_metric("wer")
60
- print(wer.compute(predictions=result["predicted"], references=result["target"]))
61
- ```
62
-
63
- **Result**: 21.1 %
 
1
+ Copy of "facebook/wav2vec2-large-xlsr-53-dutch"