language:
- pt
license: apache-2.0
tags:
- whisper-event
- generated_from_trainer
datasets:
- mozilla-foundation/common_voice_11_0
metrics:
- wer
- cer
base_model: openai/whisper-large-v2
model-index:
- name: Whisper Large Portuguese
results:
- task:
type: automatic-speech-recognition
name: Automatic Speech Recognition
dataset:
name: mozilla-foundation/common_voice_11_0 pt
type: mozilla-foundation/common_voice_11_0
config: pt
split: test
args: pt
metrics:
- type: wer
value: 4.816664144852979
name: WER
- type: cer
value: 1.6052355927195898
name: CER
- task:
type: automatic-speech-recognition
name: Automatic Speech Recognition
dataset:
name: google/fleurs pt_br
type: google/fleurs
config: pt_br
split: test
args: pt_br
metrics:
- type: wer
value: 8.56762285333714
name: WER
- type: cer
value: 5.462965196208485
name: CER
Whisper Large Portuguese
This model is a fine-tuned version of openai/whisper-large-v2 on Portuguese using the train and validation splits of Common Voice 11. Not all validation split data were used during training, I extracted 1k samples from the validation split to be used for evaluation during fine-tuning.
Usage
from transformers import pipeline
transcriber = pipeline(
"automatic-speech-recognition",
model="jonatasgrosman/whisper-large-pt-cv11"
)
transcriber.model.config.forced_decoder_ids = (
transcriber.tokenizer.get_decoder_prompt_ids(
language="pt",
task="transcribe"
)
)
transcription = transcriber("path/to/my_audio.wav")
Evaluation
I've performed the evaluation of the model using the test split of two datasets, the Common Voice 11 (same dataset used for the fine-tuning) and the Fleurs (dataset not seen during the fine-tuning). As Whisper can transcribe casing and punctuation, I've performed the model evaluation in 2 different scenarios, one using the raw text and the other using the normalized text (lowercase + removal of punctuations). Additionally, for the Fleurs dataset, I've evaluated the model in a scenario where there are no transcriptions of numerical values since the way these values are described in this dataset is different from how they are described in the dataset used in fine-tuning (Common Voice), so it is expected that this difference in the way of describing numerical values will affect the performance of the model for this type of transcription in Fleurs.
Common Voice 11
| CER | WER | |
|---|---|---|
| jonatasgrosman/whisper-large-pt-cv11 | 2.52 | 9.56 |
| jonatasgrosman/whisper-large-pt-cv11 + text normalization | 1.60 | 4.82 |
| openai/whisper-large-v2 | 4.32 | 13.92 |
| openai/whisper-large-v2 + text normalization | 2.84 | 7.02 |
Fleurs
| CER | WER | |
|---|---|---|
| jonatasgrosman/whisper-large-pt-cv11 | 4.88 | 12.08 |
| jonatasgrosman/whisper-large-pt-cv11 + text normalization | 5.46 | 8.57 |
| jonatasgrosman/whisper-large-pt-cv11 + keep only non-numeric samples | 2.35 | 9.00 |
| jonatasgrosman/whisper-large-pt-cv11 + text normalization + keep only non-numeric samples | 3.36 | 6.05 |
| openai/whisper-large-v2 | 3.52 | 10.55 |
| openai/whisper-large-v2 + text normalization | 4.19 | 7.04 |
| openai/whisper-large-v2 + keep only non-numeric samples | 2.61 | 9.29 |
| openai/whisper-large-v2 + text normalization + keep only non-numeric samples | 3.56 | 6.15 |