File size: 3,852 Bytes
193880e 20d85b6 193880e 20d85b6 193880e 20d85b6 193880e 20d85b6 193880e 0a873ad 193880e 0a873ad 193880e 0a873ad 193880e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 |
---
license: apache-2.0
language:
- vi
tags:
- automatic-speech-recognition
- robust-speech-event
- common-voice
model-index:
- name: xls-asr-vi-40h-1B
results:
- task:
name: Speech Recognition
type: automatic-speech-recognition
dataset:
name: Common Voice 7.0 vi
type: common_voice
args: vi
metrics:
- name: Test WER
type: wer
value: 34.21
- name: Test CER
type: cer
value: 19.94
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# xls-asr-vi-40h-1B
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-1b](https://huggingface.co/facebook/wav2vec2-xls-r-1b) on the common voice 7.0 vi & private dataset.
### Benchmark WER result:
| | [VIVOS](https://huggingface.co/datasets/vivos) | [COMMON VOICE 7.0 VI](https://huggingface.co/datasets/mozilla-foundation/common_voice_7_0) |
|---|---|---|
|without LM| 25.92 | 34.21 |
### Benchmark CER result:
| | [VIVOS](https://huggingface.co/datasets/vivos) | [COMMON VOICE 7.0 VI](https://huggingface.co/datasets/mozilla-foundation/common_voice_7_0) |
|---|---|---|
|without LM| 9.24 | 19.94 |
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1500
- num_epochs: 50.0
- mixed_precision_training: Native AMP
- attention_dropout: 0.2
- activation_dropout: 0.1
- warmup_steps: 1500
- mask_time_prob: .15
- mask_time_length: 10
- mask_feature_prob: 0.25
- mask_feature_length: 64
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|:-------------:|:-----:|:-----:|:---------------:|:------:|
| 4.6222 | 1.85 | 1500 | 5.9479 | 0.5474 |
| 1.1362 | 3.7 | 3000 | 7.9799 | 0.5094 |
| 0.7814 | 5.56 | 4500 | 5.0330 | 0.4724 |
| 0.6281 | 7.41 | 6000 | 2.3484 | 0.5020 |
| 0.5472 | 9.26 | 7500 | 2.2495 | 0.4793 |
| 0.4827 | 11.11 | 9000 | 1.1530 | 0.4768 |
| 0.4327 | 12.96 | 10500 | 1.6160 | 0.4646 |
| 0.3989 | 14.81 | 12000 | 3.2633 | 0.4703 |
| 0.3522 | 16.67 | 13500 | 2.2337 | 0.4708 |
| 0.3201 | 18.52 | 15000 | 3.6879 | 0.4565 |
| 0.2899 | 20.37 | 16500 | 5.4389 | 0.4599 |
| 0.2776 | 22.22 | 18000 | 3.5284 | 0.4537 |
| 0.2574 | 24.07 | 19500 | 2.1759 | 0.4649 |
| 0.2378 | 25.93 | 21000 | 3.3901 | 0.4448 |
| 0.217 | 27.78 | 22500 | 1.1632 | 0.4565 |
| 0.2115 | 29.63 | 24000 | 1.7441 | 0.4232 |
| 0.1959 | 31.48 | 25500 | 3.4992 | 0.4304 |
| 0.187 | 33.33 | 27000 | 3.6163 | 0.4369 |
| 0.1748 | 35.19 | 28500 | 3.6038 | 0.4467 |
| 0.17 | 37.04 | 30000 | 2.9708 | 0.4362 |
| 0.159 | 38.89 | 31500 | 3.2045 | 0.4279 |
| 0.153 | 40.74 | 33000 | 3.2427 | 0.4287 |
| 0.1463 | 42.59 | 34500 | 3.5439 | 0.4270 |
| 0.139 | 44.44 | 36000 | 3.9381 | 0.4150 |
| 0.1352 | 46.3 | 37500 | 4.1744 | 0.4092 |
| 0.1369 | 48.15 | 39000 | 4.2279 | 0.4154 |
| 0.1273 | 50.0 | 40500 | 4.1691 | 0.4133 |
### Framework versions
- Transformers 4.16.0.dev0
- Pytorch 1.10.1+cu102
- Datasets 1.17.1.dev0
- Tokenizers 0.11.0
|