wav2vec2-xlsr-hre-v1
This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the hre-audio-dataset8 dataset. It achieves the following results on the evaluation set:
- Loss: 0.9722
- Cer Ortho: 57.1608
- Cer: 48.9420
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0003
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 100
- num_epochs: 20
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer Ortho | Cer |
|---|---|---|---|---|---|
| 0.3674 | 0.4608 | 100 | 0.9867 | 61.0553 | 54.7010 |
| 0.3941 | 0.9217 | 200 | 0.9548 | 63.8729 | 54.5722 |
| 0.3239 | 1.3825 | 300 | 0.9726 | 60.3733 | 53.8178 |
| 0.4434 | 1.8433 | 400 | 0.8731 | 59.9605 | 53.5787 |
| 0.3304 | 2.3041 | 500 | 1.0734 | 64.0883 | 54.7930 |
| 0.3552 | 2.7650 | 600 | 0.8890 | 62.6525 | 53.1555 |
| 0.3390 | 3.2258 | 700 | 0.9412 | 58.5786 | 53.7994 |
| 0.3741 | 3.6866 | 800 | 0.9152 | 56.8737 | 52.6035 |
| 0.2740 | 4.1475 | 900 | 0.9503 | 57.8428 | 52.1803 |
| 0.2942 | 4.6083 | 1000 | 0.9740 | 57.8607 | 52.2355 |
| 0.3040 | 5.0691 | 1100 | 0.8920 | 58.9555 | 51.3707 |
| 0.3025 | 5.5300 | 1200 | 0.9718 | 58.5068 | 52.1067 |
| 0.2609 | 5.9908 | 1300 | 0.9868 | 59.3144 | 52.2171 |
| 0.2534 | 6.4516 | 1400 | 1.0018 | 56.7839 | 52.4011 |
| 0.3303 | 6.9124 | 1500 | 0.9764 | 58.3632 | 51.3707 |
| 0.2839 | 7.3733 | 1600 | 0.9399 | 60.0503 | 51.4811 |
| 0.2668 | 7.8341 | 1700 | 0.9931 | 60.9476 | 52.2355 |
| 0.3022 | 8.2949 | 1800 | 0.9181 | 60.4810 | 51.3891 |
| 0.3641 | 8.7558 | 1900 | 0.8288 | 58.0043 | 50.4876 |
| 0.3030 | 9.2166 | 2000 | 0.9127 | 58.8299 | 51.2052 |
| 0.3124 | 9.6774 | 2100 | 0.9390 | 56.6762 | 51.3707 |
| 0.2614 | 10.1382 | 2200 | 0.9390 | 58.8658 | 51.3339 |
| 0.2920 | 10.5991 | 2300 | 0.8887 | 58.0402 | 50.0644 |
| 0.2787 | 11.0599 | 2400 | 0.9710 | 61.3783 | 50.8372 |
| 0.2660 | 11.5207 | 2500 | 1.0166 | 62.5090 | 51.8675 |
| 0.2955 | 11.9816 | 2600 | 0.9085 | 60.4630 | 50.3404 |
| 0.2702 | 12.4424 | 2700 | 0.9840 | 60.2656 | 50.2300 |
| 0.2810 | 12.9032 | 2800 | 0.9933 | 59.5477 | 50.0092 |
| 0.2654 | 13.3641 | 2900 | 0.9433 | 59.5298 | 50.1932 |
| 0.2132 | 13.8249 | 3000 | 0.9613 | 57.7710 | 49.6964 |
| 0.1992 | 14.2857 | 3100 | 0.9687 | 58.7940 | 49.6596 |
| 0.2426 | 14.7465 | 3200 | 0.9831 | 59.8169 | 50.6716 |
| 0.2276 | 15.2074 | 3300 | 0.9570 | 58.5248 | 49.3284 |
| 0.1987 | 15.6682 | 3400 | 1.0591 | 58.5068 | 50.4692 |
| 0.2157 | 16.1290 | 3500 | 1.0039 | 57.9505 | 50.1012 |
| 0.1643 | 16.5899 | 3600 | 0.9512 | 56.8378 | 48.9604 |
| 0.2012 | 17.0507 | 3700 | 0.9912 | 57.3941 | 49.6412 |
| 0.1917 | 17.5115 | 3800 | 1.0092 | 57.5915 | 49.5860 |
| 0.1653 | 17.9724 | 3900 | 0.9984 | 57.5556 | 49.4572 |
| 0.1890 | 18.4332 | 4000 | 0.9601 | 57.1967 | 49.1812 |
| 0.1715 | 18.8940 | 4100 | 1.0008 | 57.5556 | 49.5676 |
| 0.1785 | 19.3548 | 4200 | 0.9669 | 57.0172 | 48.9052 |
| 0.1341 | 19.8157 | 4300 | 0.9722 | 57.2146 | 48.9604 |
| 0.1606 | 20.0 | 4340 | 0.9722 | 57.1608 | 48.9420 |
Framework versions
- Transformers 5.13.1
- Pytorch 2.11.0+cu128
- Datasets 2.18.0
- Tokenizers 0.22.2