1
---
2
language: ga
3
datasets:
4
- common_voice
5
tags:
6
- audio
7
- automatic-speech-recognition
8
- speech
9
- xlsr-fine-tuning-week
10
license: apache-2.0
11
model-index:
12
- name: XLSR Wav2Vec2 Irish by Manan Dey
13
  results:
14
  - task: 
15
      name: Speech Recognition
16
      type: automatic-speech-recognition
17
    dataset:
18
      name: Common Voice ga-IE
19
      type: common_voice
20
      args: ga-IE
21
    metrics:
22
       - name: Test WER
23
         type: wer
24
         value: 42.34
25
---
26
# Wav2Vec2-Large-XLSR-53-Irish
27
Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) in Irish using the [Common Voice](https://huggingface.co/datasets/common_voice)
28
When using this model, make sure that your speech input is sampled at 16kHz.
29
## Usage
30
The model can be used directly (without a language model) as follows:
31
```python
32
import torch
33
import torchaudio
34
from datasets import load_dataset
35
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
36
test_dataset = load_dataset("common_voice", "ga-IE", split="test[:2%]").
37
processor = Wav2Vec2Processor.from_pretrained("manandey/wav2vec2-large-xlsr-_irish")
38
model = Wav2Vec2ForCTC.from_pretrained("manandey/wav2vec2-large-xlsr-_irish")
39
resampler = torchaudio.transforms.Resample(48_000, 16_000)
40
# Preprocessing the datasets.
41
# We need to read the aduio files as arrays
42
def speech_file_to_array_fn(batch):
43
    speech_array, sampling_rate = torchaudio.load(batch["path"])
44
    batch["speech"] = resampler(speech_array).squeeze().numpy()
45
    return batch
46
test_dataset = test_dataset.map(speech_file_to_array_fn)
47
inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
48
with torch.no_grad():
49
    logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
50
predicted_ids = torch.argmax(logits, dim=-1)
51
print("Prediction:", processor.batch_decode(predicted_ids))
52
print("Reference:", test_dataset["sentence"][:2])
53
```
54
## Evaluation
55
The model can be evaluated as follows on the {language} test data of Common Voice.
56
```python
57
import torch
58
import torchaudio
59
from datasets import load_dataset, load_metric
60
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
61
import re
62
test_dataset = load_dataset("common_voice", "ga-IE", split="test")
63
wer = load_metric("wer")
64
processor = Wav2Vec2Processor.from_pretrained("manandey/wav2vec2-large-xlsr-_irish")
65
model = Wav2Vec2ForCTC.from_pretrained("manandey/wav2vec2-large-xlsr-_irish")
66
model.to("cuda")
67
chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“\%\‘\”\�\’\–\(\)]'
68
resampler = torchaudio.transforms.Resample(48_000, 16_000)
69
# Preprocessing the datasets.
70
# We need to read the aduio files as arrays
71
def speech_file_to_array_fn(batch):
72
    batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
73
    speech_array, sampling_rate = torchaudio.load(batch["path"])
74
    batch["speech"] = resampler(speech_array).squeeze().numpy()
75
    return batch
76
test_dataset = test_dataset.map(speech_file_to_array_fn)
77
# Preprocessing the datasets.
78
# We need to read the aduio files as arrays
79
def evaluate(batch):
80
    inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
81
    with torch.no_grad():
82
        logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
83
    pred_ids = torch.argmax(logits, dim=-1)
84
    batch["pred_strings"] = processor.batch_decode(pred_ids)
85
    return batch
86
result = test_dataset.map(evaluate, batched=True, batch_size=8)
87
print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
88
```
89
**Test Result**: 42.34%
90
## Training
91
The Common Voice `train`, `validation` datasets were used for training.