mrshu commited on
Commit
770b582
1 Parent(s): 8c4b9e7

First version

Browse files
README.md ADDED
@@ -0,0 +1,99 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: sl
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 Slovene
13
+ results:
14
+ - task:
15
+ name: Speech Recognition
16
+ type: automatic-speech-recognition
17
+ dataset:
18
+ name: Common Voice sl
19
+ type: common_voice
20
+ args: sl
21
+ metrics:
22
+ - name: Test WER
23
+ type: wer
24
+ value: 36.97
25
+ ---
26
+ # Wav2Vec2-Large-XLSR-53-Slovene
27
+ Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) in Slovene 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
+
30
+ ## Usage
31
+ The model can be used directly (without a language model) as follows:
32
+
33
+ ```python
34
+ import torch
35
+ import torchaudio
36
+ from datasets import load_dataset
37
+ from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
38
+ test_dataset = load_dataset("common_voice", "sl", split="test[:2%]").
39
+ processor = Wav2Vec2Processor.from_pretrained("mrshu/wav2vec2-large-xlsr-slovene")
40
+ model = Wav2Vec2ForCTC.from_pretrained("mrshu/wav2vec2-large-xlsr-slovene")
41
+ resampler = torchaudio.transforms.Resample(48_000, 16_000)
42
+ # Preprocessing the datasets.
43
+ # We need to read the aduio files as arrays
44
+ def speech_file_to_array_fn(batch):
45
+ speech_array, sampling_rate = torchaudio.load(batch["path"])
46
+ batch["speech"] = resampler(speech_array).squeeze().numpy()
47
+ return batch
48
+ test_dataset = test_dataset.map(speech_file_to_array_fn)
49
+ inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
50
+ with torch.no_grad():
51
+ logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
52
+ predicted_ids = torch.argmax(logits, dim=-1)
53
+ print("Prediction:", processor.batch_decode(predicted_ids))
54
+ print("Reference:", test_dataset["sentence"][:2])
55
+ ```
56
+
57
+
58
+ ## Evaluation
59
+ The model can be evaluated as follows on the Slovene test data of Common Voice.
60
+ ```python
61
+ import torch
62
+ import torchaudio
63
+ from datasets import load_dataset, load_metric
64
+ from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
65
+ import re
66
+ test_dataset = load_dataset("common_voice", "sl", split="test")
67
+ wer = load_metric("wer")
68
+ processor = Wav2Vec2Processor.from_pretrained("mrshu/wav2vec2-large-xlsr-slovene")
69
+ model = Wav2Vec2ForCTC.from_pretrained("mrshu/wav2vec2-large-xlsr-slovene")
70
+ model.to("cuda")
71
+ chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“\%\‘\”\�\«\»\)\(\„\'\–\’\—]'
72
+ resampler = torchaudio.transforms.Resample(48_000, 16_000)
73
+ # Preprocessing the datasets.
74
+ # We need to read the aduio files as arrays
75
+ def speech_file_to_array_fn(batch):
76
+ batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
77
+ speech_array, sampling_rate = torchaudio.load(batch["path"])
78
+ batch["speech"] = resampler(speech_array).squeeze().numpy()
79
+ return batch
80
+ test_dataset = test_dataset.map(speech_file_to_array_fn)
81
+ # Preprocessing the datasets.
82
+ # We need to read the aduio files as arrays
83
+ def evaluate(batch):
84
+ inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
85
+ with torch.no_grad():
86
+ logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
87
+ pred_ids = torch.argmax(logits, dim=-1)
88
+ batch["pred_strings"] = processor.batch_decode(pred_ids)
89
+ return batch
90
+ result = test_dataset.map(evaluate, batched=True, batch_size=8)
91
+ print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
92
+ ```
93
+
94
+ **Test Result**: 36.97 %
95
+
96
+ ## Training
97
+ The Common Voice `train`, `validation` datasets were used for training.
98
+ The script used for training can be found [here](https://colab.research.google.com/drive/14uahdilysnFsiYniHxY9fyKjFGuYQe7p)
99
+
config.json ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "/content/drive/MyDrive/wav2vec2-large-xlsr-slovene-demo/checkpoint-2400",
3
+ "activation_dropout": 0.0,
4
+ "apply_spec_augment": true,
5
+ "architectures": [
6
+ "Wav2Vec2ForCTC"
7
+ ],
8
+ "attention_dropout": 0.1,
9
+ "bos_token_id": 1,
10
+ "conv_bias": true,
11
+ "conv_dim": [
12
+ 512,
13
+ 512,
14
+ 512,
15
+ 512,
16
+ 512,
17
+ 512,
18
+ 512
19
+ ],
20
+ "conv_kernel": [
21
+ 10,
22
+ 3,
23
+ 3,
24
+ 3,
25
+ 3,
26
+ 2,
27
+ 2
28
+ ],
29
+ "conv_stride": [
30
+ 5,
31
+ 2,
32
+ 2,
33
+ 2,
34
+ 2,
35
+ 2,
36
+ 2
37
+ ],
38
+ "ctc_loss_reduction": "mean",
39
+ "ctc_zero_infinity": false,
40
+ "do_stable_layer_norm": true,
41
+ "eos_token_id": 2,
42
+ "feat_extract_activation": "gelu",
43
+ "feat_extract_dropout": 0.0,
44
+ "feat_extract_norm": "layer",
45
+ "feat_proj_dropout": 0.0,
46
+ "final_dropout": 0.0,
47
+ "gradient_checkpointing": true,
48
+ "hidden_act": "gelu",
49
+ "hidden_dropout": 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_channel_length": 10,
56
+ "mask_channel_min_space": 1,
57
+ "mask_channel_other": 0.0,
58
+ "mask_channel_prob": 0.0,
59
+ "mask_channel_selection": "static",
60
+ "mask_feature_length": 10,
61
+ "mask_feature_prob": 0.0,
62
+ "mask_time_length": 10,
63
+ "mask_time_min_space": 1,
64
+ "mask_time_other": 0.0,
65
+ "mask_time_prob": 0.05,
66
+ "mask_time_selection": "static",
67
+ "model_type": "wav2vec2",
68
+ "num_attention_heads": 16,
69
+ "num_conv_pos_embedding_groups": 16,
70
+ "num_conv_pos_embeddings": 128,
71
+ "num_feat_extract_layers": 7,
72
+ "num_hidden_layers": 24,
73
+ "pad_token_id": 30,
74
+ "transformers_version": "4.4.0",
75
+ "vocab_size": 31
76
+ }
optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d6412a776d986b7778ed3bc3e5aeebce7da3725634501e8152750a644f8856c5
3
+ size 2524030271
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 sha256:1f5876d15d270501f95c5b012a92adcc065d2e36cdcd9203dcdc815d3e22a4b6
3
+ size 1262060951
scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d925a061b55ae2fd0dc674f239e8a7c89ec7351e753745cdfa70b1eb43df83e5
3
+ size 623
special_tokens_map.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"bos_token": "<s>", "eos_token": "</s>", "unk_token": "[UNK]", "pad_token": "[PAD]"}
tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"unk_token": "[UNK]", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "[PAD]", "do_lower_case": false, "word_delimiter_token": "|"}
trainer_state.json ADDED
@@ -0,0 +1,100 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 29.625766871165645,
5
+ "global_step": 2400,
6
+ "is_hyper_param_search": false,
7
+ "is_local_process_zero": true,
8
+ "is_world_process_zero": true,
9
+ "log_history": [
10
+ {
11
+ "epoch": 4.93,
12
+ "learning_rate": 6.400000000000001e-05,
13
+ "loss": 0.0363,
14
+ "step": 400
15
+ },
16
+ {
17
+ "epoch": 4.93,
18
+ "eval_loss": 0.6197775602340698,
19
+ "eval_runtime": 84.7989,
20
+ "eval_samples_per_second": 10.389,
21
+ "eval_wer": 0.5268051434223541,
22
+ "step": 400
23
+ },
24
+ {
25
+ "epoch": 9.87,
26
+ "learning_rate": 6.756476683937824e-05,
27
+ "loss": 0.0221,
28
+ "step": 800
29
+ },
30
+ {
31
+ "epoch": 9.87,
32
+ "eval_loss": 0.7040311694145203,
33
+ "eval_runtime": 86.0378,
34
+ "eval_samples_per_second": 10.24,
35
+ "eval_wer": 0.5272007912957468,
36
+ "step": 800
37
+ },
38
+ {
39
+ "epoch": 14.81,
40
+ "learning_rate": 5.098445595854923e-05,
41
+ "loss": 0.0174,
42
+ "step": 1200
43
+ },
44
+ {
45
+ "epoch": 14.81,
46
+ "eval_loss": 0.7597708702087402,
47
+ "eval_runtime": 85.9482,
48
+ "eval_samples_per_second": 10.25,
49
+ "eval_wer": 0.5339268051434224,
50
+ "step": 1200
51
+ },
52
+ {
53
+ "epoch": 19.75,
54
+ "learning_rate": 3.440414507772021e-05,
55
+ "loss": 0.0126,
56
+ "step": 1600
57
+ },
58
+ {
59
+ "epoch": 19.75,
60
+ "eval_loss": 0.7824531197547913,
61
+ "eval_runtime": 85.9896,
62
+ "eval_samples_per_second": 10.245,
63
+ "eval_wer": 0.5224530168150346,
64
+ "step": 1600
65
+ },
66
+ {
67
+ "epoch": 24.69,
68
+ "learning_rate": 1.7823834196891192e-05,
69
+ "loss": 0.0125,
70
+ "step": 2000
71
+ },
72
+ {
73
+ "epoch": 24.69,
74
+ "eval_loss": 0.769702672958374,
75
+ "eval_runtime": 86.2655,
76
+ "eval_samples_per_second": 10.213,
77
+ "eval_wer": 0.5224530168150346,
78
+ "step": 2000
79
+ },
80
+ {
81
+ "epoch": 29.63,
82
+ "learning_rate": 1.2435233160621762e-06,
83
+ "loss": 0.0167,
84
+ "step": 2400
85
+ },
86
+ {
87
+ "epoch": 29.63,
88
+ "eval_loss": 0.7395063638687134,
89
+ "eval_runtime": 83.5861,
90
+ "eval_samples_per_second": 10.54,
91
+ "eval_wer": 0.5198813056379822,
92
+ "step": 2400
93
+ }
94
+ ],
95
+ "max_steps": 2430,
96
+ "num_train_epochs": 30,
97
+ "total_flos": 9.370015200295258e+18,
98
+ "trial_name": null,
99
+ "trial_params": null
100
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0937970b40987b25b7283a4c90e818b7ccd2993130e0236b66f7b493b31db24f
3
+ size 2287
vocab.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"r": 0, "o": 1, "x": 2, "e": 3, "u": 4, "š": 5, "v": 6, "ž": 7, "c": 8, "h": 9, "i": 10, "p": 11, "č": 12, "w": 13, "b": 14, "y": 15, "s": 16, "d": 17, "m": 18, "z": 19, "l": 20, "f": 21, "n": 23, "j": 24, "a": 25, "k": 26, "t": 27, "g": 28, "|": 22, "[UNK]": 29, "[PAD]": 30}