luigisaetta/whisper-medium-it

This model is a fine-tuned version of openai/whisper-medium on the common_voice_11_0 dataset.

It achieves the following results on the evaluation set:

• Loss: 0.1452
• Wer: 5.7191

Model description

This model is a fine-tuning of the OpenAI Whisper Medium model, on the specified dataset.

Intended uses & limitations

This model has been developed as part of the Hugging Face Whisper Fine Tuning sprint, December 2022.

It is meant to spread the knowledge on how these models are built and can be used to develop solutions where it is needed ASR on the Italian Language.

It has not been extensively tested. It is possible that on other datasets the accuracy will be lower.

Please, test it before using it.

Training and evaluation data

Trained and tested on Mozilla Common Voice, vers. 11

Training procedure

The script run.sh, and the Python file, used for the training are saved in the repository.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-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: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1216	0.2	1000	0.2289	10.0594
0.1801	0.4	2000	0.1851	7.6593
0.1763	0.6	3000	0.1615	6.5258
0.1337	0.8	4000	0.1506	6.0427
0.0742	1.05	5000	0.1452	5.7191

Framework versions

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