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: 0.5197
Wer: 0.4689

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-v1 --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-v1 --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: 7e-05
train_batch_size: 32
eval_batch_size: 16
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 2000
num_epochs: 50.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.3711	2.61	300	4.3122	1.0
3.1653	5.22	600	3.1156	1.0
2.8904	7.83	900	2.8421	0.9918
0.9207	10.43	1200	0.9895	0.8689
0.6384	13.04	1500	0.6994	0.7700
0.5215	15.65	1800	0.5628	0.6443
0.4573	18.26	2100	0.5316	0.6174
0.3875	20.87	2400	0.4932	0.5779
0.3562	23.48	2700	0.4972	0.5475
0.3218	26.09	3000	0.4895	0.5219
0.2954	28.7	3300	0.5226	0.5192
0.287	31.3	3600	0.4957	0.5146
0.2587	33.91	3900	0.4944	0.4893
0.2496	36.52	4200	0.4976	0.4895
0.2365	39.13	4500	0.5185	0.4819
0.2264	41.74	4800	0.5152	0.4776
0.2224	44.35	5100	0.5031	0.4746
0.2096	46.96	5400	0.5062	0.4708
0.2038	49.57	5700	0.5217	0.4698

Framework versions

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