arampacha/wav2vec2-xls-r-1b-uk-cv

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - UK dataset. It achieves the following results on the evaluation set:

• Loss: 0.1747
• Wer: 0.2107
• Cer: 0.0408

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: 8e-05
• train_batch_size: 16
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.3719	4.35	500	0.3389	0.4236	0.0833
1.1361	8.7	1000	0.2309	0.3162	0.0630
1.0517	13.04	1500	0.2166	0.3056	0.0597
1.0118	17.39	2000	0.2141	0.2784	0.0557
0.9922	21.74	2500	0.2231	0.2941	0.0594
0.9929	26.09	3000	0.2171	0.2892	0.0587
0.9485	30.43	3500	0.2236	0.2956	0.0599
0.9573	34.78	4000	0.2314	0.3043	0.0616
0.9195	39.13	4500	0.2169	0.2812	0.0580
0.8915	43.48	5000	0.2109	0.2780	0.0560
0.8449	47.83	5500	0.2050	0.2534	0.0514
0.8028	52.17	6000	0.2032	0.2456	0.0492
0.7881	56.52	6500	0.1890	0.2380	0.0469
0.7423	60.87	7000	0.1816	0.2245	0.0442
0.7248	65.22	7500	0.1789	0.2165	0.0422
0.6993	69.57	8000	0.1747	0.2107	0.0408

Framework versions

• Transformers 4.17.0.dev0
• Pytorch 1.10.2+cu102
• Datasets 1.18.2.dev0
• Tokenizers 0.11.0