You can deploy this model and start earning money today!

AlhagAli/
wav2vec2-xls-r-300m-poor-data-german-colab12

Name: wav2vec2-xls-r-300m-poor-data-german-colab12
Availability: InStock

automatic-speech-recognition

generated_from_trainer

apache-2.0

endpoints_compatible

Views

No views yet

wav2vec2-xls-r-300m-poor-data-german-colab12

This model is a part of my bachelor thesis for built a towrds robuster ASR with Wav2Vec2.0 with german noise data. This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the modified common voice dataset. This model has been fine tuned with 10000 DP, 7500 for training and 2500 für test. It achieves the following results on the evaluation set:

Loss: 1.6421
Wer: 0.9630

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0003
train_batch_size: 10
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 20
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 7
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
7.3484	1.0	132	3.1272	1.0
2.9727	2.0	264	2.9679	1.0
2.9202	3.0	396	3.2757	1.0
2.898	4.0	528	2.9306	1.0000
2.8612	5.0	660	2.8673	0.9983
2.5811	6.0	792	2.1479	0.9999
1.7869	7.0	924	1.6421	0.9630

Framework versions

Transformers 4.28.0
Pytorch 2.1.0+cu121
Datasets 2.15.0
Tokenizers 0.13.3

AlhagAli/wav2vec2-xls-r-300m-poor-data-german-colab12

wav2vec2-xls-r-300m-poor-data-german-colab12

Training hyperparameters

Training results

Framework versions

AlhagAli/
wav2vec2-xls-r-300m-poor-data-german-colab12