bayartsogt commited on
Commit
0b1358f
2 Parent(s): 1f3aea1 648b654

merge readme

Browse files
Files changed (1) hide show
  1. README.md +115 -0
README.md ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: mn
3
+ datasets:
4
+ - common_voice mn
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: XLSR Wav2Vec2 Mongolian V1 by Bayartsogt
15
+ results:
16
+ - task:
17
+ name: Speech Recognition
18
+ type: automatic-speech-recognition
19
+ dataset:
20
+ name: Common Voice mn
21
+ type: common_voice
22
+ args: mn
23
+ metrics:
24
+ - name: Test WER
25
+ type: wer
26
+ value: 35.33
27
+ ---
28
+
29
+ # Wav2Vec2-Large-XLSR-53-Mongolian-v1
30
+
31
+ Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on Mongolian using the [Common Voice](https://huggingface.co/datasets/common_voice).
32
+
33
+ When using this model, make sure that your speech input is sampled at 16kHz.
34
+
35
+ ## Usage
36
+ The model can be used directly (without a language model) as follows:
37
+ ```python
38
+ import torch
39
+ import torchaudio
40
+ from datasets import load_dataset
41
+ from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
42
+ test_dataset = load_dataset("common_voice", "mn", split="test[:2%]")
43
+ processor = Wav2Vec2Processor.from_pretrained("bayartsogt/wav2vec2-large-xlsr-mongolian-v1")
44
+ model = Wav2Vec2ForCTC.from_pretrained("bayartsogt/wav2vec2-large-xlsr-mongolian-v1")
45
+ resampler = torchaudio.transforms.Resample(48_000, 16_000)
46
+ # Preprocessing the datasets.
47
+ # We need to read the aduio files as arrays
48
+ def speech_file_to_array_fn(batch):
49
+ speech_array, sampling_rate = torchaudio.load(batch["path"])
50
+ batch["speech"] = resampler(speech_array).squeeze().numpy()
51
+ return batch
52
+ test_dataset = test_dataset.map(speech_file_to_array_fn)
53
+ inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
54
+ with torch.no_grad():
55
+ logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
56
+ predicted_ids = torch.argmax(logits, dim=-1)
57
+ print("Prediction:", processor.batch_decode(predicted_ids))
58
+ print("Reference:", test_dataset["sentence"][:2])
59
+ ```
60
+
61
+ ## Evaluation
62
+
63
+ The model can be evaluated as follows on the Mongolian test data of Common Voice.
64
+
65
+
66
+ ```python
67
+ import torch
68
+ import torchaudio
69
+ from datasets import load_dataset, load_metric
70
+ from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
71
+ import re
72
+
73
+ test_dataset = load_dataset("common_voice", "mn", split="test")
74
+ wer = load_metric("wer")
75
+
76
+ processor = Wav2Vec2Processor.from_pretrained("bayartsogt/wav2vec2-large-xlsr-mongolian-v1")
77
+ model = Wav2Vec2ForCTC.from_pretrained("bayartsogt/wav2vec2-large-xlsr-mongolian-v1")
78
+ model.to("cuda")
79
+
80
+ chars_to_ignore_regex = '[\!\"\'\,\.\«\»\?\-]'
81
+ resampler = torchaudio.transforms.Resample(48_000, 16_000)
82
+
83
+ # Preprocessing the datasets.
84
+ # We need to read the aduio files as arrays
85
+ def speech_file_to_array_fn(batch):
86
+ batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
87
+ speech_array, sampling_rate = torchaudio.load(batch["path"])
88
+ batch["speech"] = resampler(speech_array).squeeze().numpy()
89
+ return batch
90
+
91
+ test_dataset = test_dataset.map(speech_file_to_array_fn)
92
+
93
+ # Preprocessing the datasets.
94
+ # We need to read the aduio files as arrays
95
+ def evaluate(batch):
96
+ inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
97
+
98
+ with torch.no_grad():
99
+ logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
100
+
101
+ pred_ids = torch.argmax(logits, dim=-1)
102
+ batch["pred_strings"] = processor.batch_decode(pred_ids)
103
+ return batch
104
+
105
+ result = test_dataset.map(evaluate, batched=True, batch_size=8)
106
+
107
+ print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
108
+ ```
109
+
110
+ **Test Result**: 35.33 %
111
+
112
+
113
+ ## Training
114
+
115
+ The Common Voice `train` dataset was used for training as well as ... and ...