marma commited on
Commit
20b7036
1 Parent(s): 61a8286
README.md ADDED
@@ -0,0 +1,123 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: sv
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 Swedish by Marma
13
+ results:
14
+ - task:
15
+ name: Speech Recognition
16
+ type: automatic-speech-recognition
17
+ dataset:
18
+ name: Common Voice sv-SE
19
+ type: common_voice
20
+ args: sv
21
+ metrics:
22
+ - name: Test WER
23
+ type: wer
24
+ value: 23.33
25
+ ---
26
+
27
+ # Wav2Vec2-Large-XLSR-53-Swedish
28
+
29
+ Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) in Swedish using the [Common Voice](https://huggingface.co/datasets/common_voice)
30
+ When using this model, make sure that your speech input is sampled at 16kHz.
31
+
32
+ ## Usage
33
+
34
+ The model can be used directly (without a language model) as follows:
35
+
36
+ ```python
37
+ import torch
38
+ import torchaudio
39
+ from datasets import load_dataset
40
+ from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
41
+
42
+ test_dataset = load_dataset("common_voice", "sv-SE", split="test[:2%]").
43
+
44
+ processor = Wav2Vec2Processor.from_pretrained("marma/wav2vec2-large-xlsr-swedish")
45
+ model = Wav2Vec2ForCTC.from_pretrained("marma/wav2vec2-large-xlsr-swedish")
46
+
47
+ resampler = torchaudio.transforms.Resample(48_000, 16_000)
48
+
49
+ # Preprocessing the datasets.
50
+ # We need to read the aduio files as arrays
51
+ def speech_file_to_array_fn(batch):
52
+ speech_array, sampling_rate = torchaudio.load(batch["path"])
53
+ batch["speech"] = resampler(speech_array).squeeze().numpy()
54
+ return batch
55
+
56
+ test_dataset = test_dataset.map(speech_file_to_array_fn)
57
+ inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
58
+
59
+ with torch.no_grad():
60
+ logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
61
+
62
+ predicted_ids = torch.argmax(logits, dim=-1)
63
+
64
+ print("Prediction:", processor.batch_decode(predicted_ids))
65
+ print("Reference:", test_dataset["sentence"][:2])
66
+ ```
67
+
68
+
69
+ ## Evaluation
70
+
71
+ The model can be evaluated as follows on the Swedish test data of Common Voice.
72
+
73
+
74
+ ```python
75
+ import torch
76
+ import torchaudio
77
+ from datasets import load_dataset, load_metric
78
+ from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
79
+ import re
80
+
81
+ test_dataset = load_dataset("common_voice", "sv-SE", split="test")
82
+ wer = load_metric("wer")
83
+
84
+ processor = Wav2Vec2Processor.from_pretrained("marma/wav2vec2-large-xlsr-swedish")
85
+ model = Wav2Vec2ForCTC.from_pretrained("marma/wav2vec2-large-xlsr-swedish")
86
+ model.to("cuda")
87
+
88
+ chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“]'
89
+ resampler = torchaudio.transforms.Resample(48_000, 16_000)
90
+
91
+ # Preprocessing the datasets.
92
+ # We need to read the aduio files as arrays
93
+ def speech_file_to_array_fn(batch):
94
+ batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
95
+ speech_array, sampling_rate = torchaudio.load(batch["path"])
96
+ batch["speech"] = resampler(speech_array).squeeze().numpy()
97
+ return batch
98
+
99
+ test_dataset = test_dataset.map(speech_file_to_array_fn)
100
+
101
+ # Preprocessing the datasets.
102
+ # We need to read the aduio files as arrays
103
+ def evaluate(batch):
104
+ inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
105
+
106
+ with torch.no_grad():
107
+ logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
108
+
109
+ pred_ids = torch.argmax(logits, dim=-1)
110
+ batch["pred_strings"] = processor.batch_decode(pred_ids)
111
+ return batch
112
+
113
+ result = test_dataset.map(evaluate, batched=True, batch_size=8)
114
+
115
+ print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
116
+ ```
117
+
118
+ **Test Result**: 23.33 %
119
+
120
+
121
+ ## Training
122
+
123
+ The [NST Swedish Dictation](https://www.nb.no/sprakbanken/en/resource-catalogue/oai-nb-no-sbr-17/) was used for training.
config.json ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "activation_dropout": 0.1,
3
+ "apply_spec_augment": true,
4
+ "architectures": [
5
+ "Wav2Vec2ForCTC"
6
+ ],
7
+ "attention_dropout": 0.1,
8
+ "bos_token_id": 1,
9
+ "conv_bias": true,
10
+ "conv_dim": [
11
+ 512,
12
+ 512,
13
+ 512,
14
+ 512,
15
+ 512,
16
+ 512,
17
+ 512
18
+ ],
19
+ "conv_kernel": [
20
+ 10,
21
+ 3,
22
+ 3,
23
+ 3,
24
+ 3,
25
+ 2,
26
+ 2
27
+ ],
28
+ "conv_stride": [
29
+ 5,
30
+ 2,
31
+ 2,
32
+ 2,
33
+ 2,
34
+ 2,
35
+ 2
36
+ ],
37
+ "ctc_loss_reduction": "sum",
38
+ "ctc_zero_infinity": false,
39
+ "do_stable_layer_norm": true,
40
+ "eos_token_id": 2,
41
+ "feat_extract_activation": "gelu",
42
+ "feat_extract_dropout": 0.0,
43
+ "feat_extract_norm": "layer",
44
+ "feat_proj_dropout": 0.1,
45
+ "final_dropout": 0.1,
46
+ "gradient_checkpointing": false,
47
+ "hidden_act": "gelu",
48
+ "hidden_dropout": 0.1,
49
+ "hidden_dropout_prob": 0.1,
50
+ "hidden_size": 1024,
51
+ "initializer_range": 0.02,
52
+ "intermediate_size": 4096,
53
+ "layer_norm_eps": 1e-05,
54
+ "layerdrop": 0.1,
55
+ "mask_feature_length": 10,
56
+ "mask_feature_prob": 0.0,
57
+ "mask_time_length": 10,
58
+ "mask_time_prob": 0.05,
59
+ "model_type": "wav2vec2",
60
+ "num_attention_heads": 16,
61
+ "num_conv_pos_embedding_groups": 16,
62
+ "num_conv_pos_embeddings": 128,
63
+ "num_feat_extract_layers": 7,
64
+ "num_hidden_layers": 24,
65
+ "pad_token_id": 0,
66
+ "transformers_version": "4.4.0.dev0",
67
+ "vocab_size": 35
68
+ }
preprocessor_config.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_normalize": true,
3
+ "feature_size": 1,
4
+ "padding_side": "right",
5
+ "padding_value": 0.0,
6
+ "return_attention_mask": true,
7
+ "sampling_rate": 16000
8
+ }
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dc24d4db3a895f9c8042cee59749fee054d594f135c35ad58d4e2cc9a5578a17
3
+ size 1262071447
special_tokens_map.json ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ {"bos_token": "<s>", "eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>"}
2
+
tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "<pad>", "do_lower_case": true, "return_attention_mask": false, "do_normalize": true}
vocab.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"<pad>": 0, "<s>": 1, "</s>": 2, "<unk>": 3, "|": 4, "T": 5, "E": 6, "A": 7, "R": 8, "N": 9, "S": 10, "I": 11, "L": 12, "D": 13, "O": 14, "M": 15, "G": 16, "K": 17, "F": 18, "U": 19, "V": 20, "H": 21, "Ä": 22, "Å": 23, "P": 24, "Ö": 25, "B": 26, "J": 27, "C": 28, "Y": 29, "X": 30, "É": 31, "Z": 32, "W": 33 }