This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_7_0 - SV-SE dataset. It achieves the following results on the evaluation set:

Loss: 0.2604
Wer: 0.2334

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: 7.5e-05
train_batch_size: 4
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 1
total_train_batch_size: 32
total_eval_batch_size: 32
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

See Tensorboard

Evaluation Commands

To evaluate on mozilla-foundation/common_voice_7_0 with split test

python eval.py --model_id patrickvonplaten/xls-r-300-sv-cv7 --dataset mozilla-foundation/common_voice_7_0 --config sv-SE --split test

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

python eval.py --model_id patrickvonplaten/xls-r-300-sv-cv7 --dataset speech-recognition-community-v2/dev_data --config sv --split validation --chunk_length_s 5.0 --stride_length_s 1.0

Framework versions

Transformers 4.17.0.dev0
Pytorch 1.9.0+cu111
Datasets 1.18.4.dev0
Tokenizers 0.10.3