jlondonobo/whisper-large-v2-pt

Whisper Large V2 Portuguese 🇧🇷🇵🇹

Bem-vindo ao whisper large-v2 para transcrição em português 👋🏻

Transcribe Portuguese audio to text with the highest precision.

• Loss: 0.282
• Wer: 5.590

This model is a fine-tuned version of openai/whisper-large-v2 on the mozilla-foundation/common_voice_11 dataset. If you want a lighter model, you may be interested in jlondonobo/whisper-medium-pt. It achieves faster inference with almost no difference in WER.

Comparable models

Reported WER is based on the evaluation subset of Common Voice.

Model	WER	# Parameters
jlondonobo/whisper-large-v2-pt	5.590 🤗	1550M
openai/whisper-large-v2	6.300	1550M
jlondonobo/whisper-medium-pt	6.579	769M
jonatasgrosman/wav2vec2-large-xlsr-53-portuguese	11.310	317M
Edresson/wav2vec2-large-xlsr-coraa-portuguese	20.080	317M

Training hyperparameters

We used the following hyperparameters for training:

• learning_rate: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• 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
0.0828	1.09	1000	0.1868	6.778
0.0241	3.07	2000	0.2057	6.109
0.0084	5.06	3000	0.2367	6.029
0.0015	7.04	4000	0.2469	5.709
0.0009	9.02	5000	0.2821	5.590 🤗

Framework versions

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