michaelszhu/wav2vec2-small-finetuned-radio-ASR

BANG WAV2VEC v1 (EN)

This model is a fine-tuned version of openai/wav2vec2 on the Radio-Modified Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 261.3464
• Wer: 99.5704

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: 0.0001
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 6000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
206.0696	0.1667	1000	368.4753	98.1649
164.236	0.3333	2000	310.1212	98.1649
168.343	0.5	3000	299.5959	98.1649
165.9723	0.6667	4000	293.7289	98.1635
155.8457	0.8333	5000	265.2044	98.1635
155.6296	1.0	6000	261.3464	99.5704

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1