Audio
Whisper Small Portuguese (CV13)
Fine-tuned Whisper Small for Portuguese speech recognition, trained on Common Voice 13.0 achieving 10.3 WER.
About this model
This model is a fine-tune of OpenAI's Whisper Small specifically optimized for Portuguese automatic speech recognition (ASR). It was trained on the Mozilla Common Voice 13.0 Portuguese dataset using Hugging Face's Whisper training pipeline. The model achieves a Word Error Rate (WER) of 10.30 on the CV13.0 Portuguese test split, representing a meaningful improvement over the base Whisper Small for Portuguese transcription tasks.
Training used a learning rate of 1e-05 with batch size 64 over 5,000 steps, completed in approximately 8 hours on a single A100 80GB GPU. The optimizer was Adam with betas (0.9, 0.999) and a linear learning rate schedule with 500 warmup steps. The model is distributed in safetensors format and is compatible with the transformers library for inference via the automatic-speech-recognition pipeline.
For usage, load the model with `AutoModelForSpeechSeq2Seq.from_pretrained("artificialguybr/whisper-small-pt-cv13")` and pair it with a Whisper processor. Recommended for Portuguese audio transcription; performance on other languages may degrade. The model expects 16kHz audio input and works best with clear speech without heavy background noise.
Learn more
Read the in-depth guide about Whisper Small Portuguese (CV13)
Project signals
- 7 Hugging Face downloads
- 1 Hugging Face likes
Topics
transformers · pytorch · safetensors · whisper · automatic-speech-recognition · model-index · endpoints_compatible