1 |
--- |
2 |
language: kk |
3 |
datasets: |
4 |
- kazakh_speech_corpus |
5 |
metrics: |
6 |
- wer |
7 |
tags: |
8 |
- audio |
9 |
- automatic-speech-recognition |
10 |
- speech |
11 |
- xlsr-fine-tuning-week |
12 |
license: apache-2.0 |
13 |
model-index: |
14 |
- name: Wav2Vec2-XLSR-53 Kazakh by adilism |
15 |
results: |
16 |
- task: |
17 |
name: Speech Recognition |
18 |
type: automatic-speech-recognition |
19 |
dataset: |
20 |
name: Kazakh Speech Corpus v1.1 |
21 |
type: kazakh_speech_corpus |
22 |
args: kk |
23 |
metrics: |
24 |
- name: Test WER |
25 |
type: wer |
26 |
value: 19.65 |
27 |
--- |
28 |
|
29 |
# Wav2Vec2-Large-XLSR-53-Kazakh |
30 |
|
31 |
Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) for Kazakh ASR using the [Kazakh Speech Corpus v1.1](https://issai.nu.edu.kz/kz-speech-corpus/?version=1.1) |
32 |
|
33 |
When using this model, make sure that your speech input is sampled at 16kHz. |
34 |
|
35 |
## Usage |
36 |
|
37 |
The model can be used directly (without a language model) as follows: |
38 |
|
39 |
```python |
40 |
import torch |
41 |
import torchaudio |
42 |
from datasets import load_dataset |
43 |
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor |
44 |
|
45 |
from utils import get_test_dataset |
46 |
|
47 |
test_dataset = get_test_dataset("ISSAI_KSC_335RS_v1.1") |
48 |
|
49 |
processor = Wav2Vec2Processor.from_pretrained("wav2vec2-large-xlsr-kazakh") |
50 |
model = Wav2Vec2ForCTC.from_pretrained("wav2vec2-large-xlsr-kazakh") |
51 |
|
52 |
|
53 |
# Preprocessing the datasets. |
54 |
# We need to read the audio files as arrays |
55 |
def speech_file_to_array_fn(batch): |
56 |
speech_array, sampling_rate = torchaudio.load(batch["path"]) |
57 |
batch["speech"] = torchaudio.transforms.Resample(sampling_rate, 16_000)(speech_array).squeeze().numpy() |
58 |
return batch |
59 |
|
60 |
test_dataset = test_dataset.map(speech_file_to_array_fn) |
61 |
inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True) |
62 |
|
63 |
with torch.no_grad(): |
64 |
logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits |
65 |
|
66 |
predicted_ids = torch.argmax(logits, dim=-1) |
67 |
|
68 |
print("Prediction:", processor.batch_decode(predicted_ids)) |
69 |
print("Reference:", test_dataset["sentence"][:2]) |
70 |
``` |
71 |
|
72 |
|
73 |
## Evaluation |
74 |
|
75 |
The model can be evaluated as follows on the test set of [Kazakh Speech Corpus v1.1](https://issai.nu.edu.kz/kz-speech-corpus/?version=1.1). To evaluate, download the [archive](https://www.openslr.org/resources/102/ISSAI_KSC_335RS_v1.1_flac.tar.gz), untar and pass the path to data to `get_test_dataset` as below: |
76 |
|
77 |
```python |
78 |
import torch |
79 |
import torchaudio |
80 |
from datasets import load_dataset, load_metric |
81 |
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor |
82 |
import re |
83 |
|
84 |
from utils import get_test_dataset |
85 |
|
86 |
test_dataset = get_test_dataset("ISSAI_KSC_335RS_v1.1") |
87 |
wer = load_metric("wer") |
88 |
|
89 |
processor = Wav2Vec2Processor.from_pretrained("adilism/wav2vec2-large-xlsr-kazakh") |
90 |
model = Wav2Vec2ForCTC.from_pretrained("adilism/wav2vec2-large-xlsr-kazakh") |
91 |
model.to("cuda") |
92 |
|
93 |
|
94 |
# Preprocessing the datasets. |
95 |
# We need to read the audio files as arrays |
96 |
def speech_file_to_array_fn(batch): |
97 |
batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower() |
98 |
speech_array, sampling_rate = torchaudio.load(batch["path"]) |
99 |
batch["speech"] = torchaudio.transforms.Resample(sampling_rate, 16_000)(speech_array).squeeze().numpy() |
100 |
return batch |
101 |
|
102 |
test_dataset = test_dataset.map(speech_file_to_array_fn) |
103 |
|
104 |
def evaluate(batch): |
105 |
inputs = processor(batch["text"], sampling_rate=16_000, return_tensors="pt", padding=True) |
106 |
|
107 |
with torch.no_grad(): |
108 |
logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits |
109 |
|
110 |
pred_ids = torch.argmax(logits, dim=-1) |
111 |
batch["pred_strings"] = processor.batch_decode(pred_ids) |
112 |
return batch |
113 |
|
114 |
result = test_dataset.map(evaluate, batched=True, batch_size=8) |
115 |
|
116 |
print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"]))) |
117 |
``` |
118 |
|
119 |
**Test Result**: 19.65% |
120 |
|
121 |
|
122 |
## Training |
123 |
|
124 |
The Kazakh Speech Corpus v1.1 `train` dataset was used for training. |
