metadata

license: apache-2.0
base_model: facebook/hubert-large-ll60k
tags:
  - generated_from_trainer
metrics:
  - wer
model-index:
  - name: speech_ocean_hubert_mdd
    results: []

speech_ocean_hubert_mdd

This model is a fine-tuned version of facebook/hubert-large-ll60k on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 2.3987
Wer: 0.5798
Cer: 0.6474

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: 8
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: 500
num_epochs: 20
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
94.265	0.9873	39	85.1429	0.9995	0.9840
52.6364	2.0	79	32.7798	1.0	1.0
27.2435	2.9873	118	14.8924	1.0	1.0
10.122	4.0	158	7.1662	1.0	1.0
5.6778	4.9873	197	5.3731	1.0	1.0
5.3058	6.0	237	5.2800	1.0	1.0
5.3233	6.9873	276	5.2647	1.0	1.0
5.2846	8.0	316	5.2266	1.0	1.0
5.2555	8.9873	355	5.1857	1.0	1.0
5.172	10.0	395	5.1649	1.0	1.0
5.1551	10.9873	434	5.0741	1.0	1.0
5.1312	12.0	474	5.0101	1.0	1.0
4.9727	12.9873	513	4.7469	1.0	1.0
4.6317	14.0	553	4.3717	0.9541	0.9580
4.1657	14.9873	592	3.8313	0.8953	0.9418
3.7754	16.0	632	3.3712	0.8003	0.8418
3.3744	16.9873	671	2.9940	0.7360	0.8192
3.0452	18.0	711	2.6717	0.6829	0.7675
2.7951	18.9873	750	2.4711	0.6175	0.6972
2.7407	19.7468	780	2.3987	0.5798	0.6474

Framework versions

Transformers 4.40.0
Pytorch 2.2.1+cu121
Datasets 2.19.0
Tokenizers 0.19.1