metadata

library_name: transformers
language:
  - id
license: mit
base_model: pyannote/speaker-diarization-3.1
tags:
  - speaker-diarization
  - speaker-segmentation
  - generated_from_trainer
datasets:
  - speaker-segmentation
model-index:
  - name: speaker-segmentation-fine-tuned-datasetID-hugging_2_4_updated_02
    results: []

speaker-segmentation-fine-tuned-datasetID-hugging_2_4_updated_02

This model is a fine-tuned version of pyannote/speaker-diarization-3.1 on the speaker-segmentation dataset. It achieves the following results on the evaluation set:

Loss: 0.4274
Model Preparation Time: 0.0065
Der: 0.1421
False Alarm: 0.0214
Missed Detection: 0.0105
Confusion: 0.1102

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: 64
eval_batch_size: 64
seed: 100
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.15
num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Der	False Alarm	Missed Detection	Confusion
0.5722	1.0	285	0.5763	0.0065	0.1903	0.0275	0.0158	0.1470
0.5001	2.0	570	0.5077	0.0065	0.1685	0.0240	0.0125	0.1319
0.4846	3.0	855	0.4810	0.0065	0.1604	0.0221	0.0121	0.1261
0.4441	4.0	1140	0.4686	0.0065	0.1577	0.0220	0.0111	0.1245
0.447	5.0	1425	0.4477	0.0065	0.1494	0.0216	0.0110	0.1168
0.4265	6.0	1710	0.4408	0.0065	0.1458	0.0217	0.0106	0.1135
0.428	7.0	1995	0.4311	0.0065	0.1433	0.0215	0.0105	0.1114
0.4047	8.0	2280	0.4291	0.0065	0.1429	0.0214	0.0105	0.1109
0.4069	9.0	2565	0.4280	0.0065	0.1422	0.0214	0.0105	0.1103
0.4133	10.0	2850	0.4274	0.0065	0.1421	0.0214	0.0105	0.1102

Framework versions

Transformers 4.50.3
Pytorch 2.6.0+cu124
Datasets 3.5.0
Tokenizers 0.21.1