qanastek/whisper-tiny-french-cased

Whisper Tiny French Cased

This model is a fine-tuned version of openai/whisper-tiny on the mozilla-foundation/common_voice_11_0 fr dataset. It achieves the following results on the evaluation set:

• Loss: 0.6509
• Wer on mozilla-foundation/common_voice_11_0 fr: 33.0655
• Wer on google/fleurs fr_fr: 36.69
• Wer on facebook/voxpopuli fr: 32.71

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: 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
0.7185	0.2	1000	0.7608	38.1636
0.6052	1.2	2000	0.6949	34.9513
0.4467	2.2	3000	0.6708	34.3393
0.4773	3.2	4000	0.6536	33.2102
0.4479	4.2	5000	0.6509	33.0655

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.11.0+cu102
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2