wav2vec2-large-xls-r-300m-hsb-v1

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - HSB dataset. It achieves the following results on the evaluation set:

• Loss: 0.5684
• Wer: 0.4402

Evaluation Commands

1. To evaluate on mozilla-foundation/common_voice_8_0 with test split

python eval.py --model_id DrishtiSharma/wav2vec2-large-xls-r-300m-hsb-v1 --dataset mozilla-foundation/common_voice_8_0 --config hsb --split test --log_outputs

2. To evaluate on speech-recognition-community-v2/dev_data

Upper Sorbian language isn't available in speech-recognition-community-v2/dev_data

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.00045
• 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: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
8.972	3.23	100	3.7498	1.0
3.3401	6.45	200	3.2320	1.0
3.2046	9.68	300	3.1741	0.9806
2.4031	12.9	400	1.0579	0.8996
1.0427	16.13	500	0.7989	0.7557
0.741	19.35	600	0.6405	0.6299
0.5699	22.58	700	0.6129	0.5928
0.4607	25.81	800	0.6548	0.5695
0.3827	29.03	900	0.6268	0.5190
0.3282	32.26	1000	0.5919	0.5016
0.2764	35.48	1100	0.5953	0.4805
0.2335	38.71	1200	0.5717	0.4728
0.2106	41.94	1300	0.5674	0.4569
0.1859	45.16	1400	0.5685	0.4502
0.1592	48.39	1500	0.5684	0.4402

Framework versions

• Transformers 4.16.1
• Pytorch 1.10.0+cu111
• Datasets 1.18.2
• Tokenizers 0.11.0