samitizerxu/wav2vec2-xls-r-300m-es

wav2vec2-cls-r-300m-es

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

• Loss: 0.5160
• Wer: 0.4016

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 8.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.1277	1.14	500	2.0259	0.9999
1.4111	2.28	1000	1.1251	0.8894
0.8461	3.42	1500	0.8205	0.7244
0.5042	4.57	2000	0.6116	0.5463
0.3072	5.71	2500	0.5507	0.4506
0.2181	6.85	3000	0.5213	0.4177
0.1608	7.99	3500	0.5161	0.4019

Framework versions

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

Evaluation Commands

1. To evaluate on mozilla-foundation/common_voice_7_0 with split test

python eval.py --model_id samitizerxu/wav2vec2-xls-r-300m-es --dataset mozilla-foundation/common_voice_7_0 --config es --split test

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

python eval.py --model_id samitizerxu/wav2vec2-xls-r-300m-es --dataset speech-recognition-community-v2/dev_data --config es --split validation --chunk_length_s 5.0 --stride_length_s 1.0