speaker-segmentation-fine-tuned-ami-speaker-diarization
This model is a fine-tuned version of openai/whisper-small on the diarizers-community/ami_speaker_diarization_dataset dataset. It achieves the following results on the evaluation set:
- Loss: 0.4425
- Der: 0.1760
- False Alarm: 0.0627
- Missed Detection: 0.0634
- Confusion: 0.0499
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: 0.001
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- num_epochs: 5
Training results
| Training Loss | Epoch | Step | Validation Loss | Der | False Alarm | Missed Detection | Confusion |
|---|---|---|---|---|---|---|---|
| 0.3957 | 1.0 | 1809 | 0.4538 | 0.1799 | 0.0605 | 0.0656 | 0.0537 |
| 0.4027 | 2.0 | 3618 | 0.4446 | 0.1780 | 0.0645 | 0.0627 | 0.0508 |
| 0.3639 | 3.0 | 5427 | 0.4504 | 0.1798 | 0.0669 | 0.0604 | 0.0524 |
| 0.3764 | 4.0 | 7236 | 0.4431 | 0.1762 | 0.0632 | 0.0623 | 0.0508 |
| 0.3916 | 5.0 | 9045 | 0.4425 | 0.1760 | 0.0627 | 0.0634 | 0.0499 |
Framework versions
- Transformers 4.40.1
- Pytorch 2.1.2
- Datasets 2.18.0
- Tokenizers 0.19.1
- Downloads last month
- 5
Model tree for Ataullha/speaker-segmentation-fine-tuned-ami-speaker-diarization-eng
Base model
openai/whisper-small