sgangireddy/whisper-largev2-mls-german

Whisper largeV2 German MLS

This model is a fine-tuned version of openai/whisper-large-v2 on the facebook/multilingual_librispeech german dataset. It achieves the following results on the evaluation set:

• Loss: 0.1370
• Wer: 6.0483

Model description

The model is fine-tuned for 4000 updates/steps on multilingual librispeech German train data.

• Zero-shot - 5.5 (MLS German test)
• Fine-tune MLS German train - 6.04 (MLS German test)

Even after fine-tuning the model is doing slightly worse than the zero-shot.

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: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1755	0.25	1000	0.1844	7.7118
0.1185	0.5	2000	0.1636	7.0659
0.1081	0.75	3000	0.1396	6.0844
0.1222	1.0	4000	0.1370	6.0483

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2