speech-accent-pt-br-classifier

Speech Portuguese (Brazilian) Accent Classifier

🎙️🤖🇧🇷

This project is a speech accent classifier that distinguishes between Portuguese (Brazilian) and other accents.

Project Overview

This application uses a trained model to classify speech accents into two categories:

1. Portuguese (Brazilian)
2. Other

The model is based on the author's work [results] and utilizes the Portuguese portion of the Common Voice dataset (version 11.0) from Mozilla Foundation.

Dataset

The project uses the Portuguese subset of the Common Voice dataset:

• Dataset: "mozilla-foundation/common_voice_11_0", "pt"

Brazilian accents included in the dataset:

• Português do Brasil, Região Sul do Brasil
• Paulistano
• Paulista, Brasileiro
• Carioca
• Mato Grosso
• Mineiro
• Interior Paulista
• Gaúcho
• Nordestino
• And various regional mixes

Model and Processor

The project utilizes the following model and processor:

• Base Model: "facebook/wav2vec2-base-960h"
• Processor: Wav2Vec2Processor.from_pretrained

Model Versions

Was trained three versions of the model with different configurations:

1. (OLD) v 1.1:

   • Epochs: 3
   • Training samples: 1000
   • Validation samples: 200

2. (OLD) v 1.2:

   • Epochs: 10
   • Training samples: 1000
   • Validation samples: 500

3. (NEW) v 1.3:

   • Epochs: 20
   • Training samples: 5000
   • Validation samples: 1000

All models were trained using high RAM GPU on Google Colab Pro.

Model Structure (files)

Each version of the model includes the following files: results config.json | preprocessor_config.json | model.safetensors | special_tokens_map.json | tokenizer_config.json | vocab.json

How to Use

Test with recording or uploading an audio file. To test, I recommend short sentences.

License

This project is licensed under the Eclipse Public License 2.0 (ECL-2.0).

Developer Information

Developed by Ramon Mayor Martins (2024)

• Email: [email protected]
• Homepage: https://rmayormartins.github.io/
• Twitter: @rmayormartins
• GitHub: https://github.com/rmayormartins

Acknowledgements

Special thanks to Instituto Federal de Santa Catarina (Federal Institute of Santa Catarina) IFSC-São José-Brazil.

Contact

For any queries or suggestions, please contact the developer using the information provided above.