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

Loss: 0.3421
Wer: 0.2860

Evaluation Commands

To evaluate on mozilla-foundation/common_voice_8_0 with test split

python eval.py --model_id DrishtiSharma/wav2vec2-large-xls-r-300m-bg-d2 --dataset mozilla-foundation/common_voice_8_0 --config bg --split test --log_outputs

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

python eval.py --model_id DrishtiSharma/wav2vec2-large-xls-r-300m-bg-d2 --dataset speech-recognition-community-v2/dev_data --config bg --split validation --chunk_length_s 10 --stride_length_s 1

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.00025
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: 700
num_epochs: 35
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.8791	1.74	200	3.1902	1.0
3.0441	3.48	400	2.8098	0.9864
1.1499	5.22	600	0.4668	0.5014
0.4968	6.96	800	0.4162	0.4472
0.3553	8.7	1000	0.3580	0.3777
0.3027	10.43	1200	0.3422	0.3506
0.2562	12.17	1400	0.3556	0.3639
0.2272	13.91	1600	0.3621	0.3583
0.2125	15.65	1800	0.3436	0.3358
0.1904	17.39	2000	0.3650	0.3545
0.1695	19.13	2200	0.3366	0.3241
0.1532	20.87	2400	0.3550	0.3311
0.1453	22.61	2600	0.3582	0.3131
0.1359	24.35	2800	0.3524	0.3084
0.1233	26.09	3000	0.3503	0.2973
0.1114	27.83	3200	0.3434	0.2946
0.1051	29.57	3400	0.3474	0.2956
0.0965	31.3	3600	0.3426	0.2907
0.0923	33.04	3800	0.3478	0.2894
0.0894	34.78	4000	0.3421	0.2860

Framework versions

Transformers 4.16.2
Pytorch 1.10.0+cu111
Datasets 1.18.3
Tokenizers 0.11.0