wav2vec_pular_maternalhealth

Pulaar Maternal Health ASR

Model Description

This model is a fine-tuned speech recognition model for Pulaar (Fula), specially for the Futa Toro dialect, with a focus on maternal health-related speech.

The model was developed by KitalaAI / YUX Cultural AI Lab to improve automatic speech recognition for low-resource African languages and maternal health applications.

Intended Use

The model is intended for:

• Automatic transcription of Pulaar speech
• Research on low-resource African language ASR
• Maternal health-related speech applications
• Evaluation and benchmarking of ASR systems

Training Data

The model was fine-tuned using a Pulaar speech dataset containing maternal-health-related utterances.

Language: Pulaar (Fula)

Domain: Maternal health

More information about the dataset, speakers, collection process, and data license should be provided here.

Training

The model was fine-tuned from a pretrained speech recognition model.

Training techniques and hyperparameters: Total validated recordings: 754 Training epochs: 23 (early stopping) Learning rate: 1e-5

Base model: Leonel-Maia/Wav2Vec2-Fula Fine-tuning method: LoRA Training examples: 607 Validation examples: 76 Test examples: 71 Total validated recordings: 754 Training epochs: 23 (early stopping) Learning rate: 1e-5

Evaluation

The model was evaluated using:

• Word Error Rate (WER)
• Character Error Rate (CER)

Limitations

• Performance may vary across Pulaar dialects and accents.
• The model may perform differently on speech outside the maternal-health domain.
• The dataset represent only fouta toro dialect
• The model should not be considered a substitute for professional medical advice.

Ethical Considerations

The model is intended for research and responsible development of speech technologies for underserved language communities.

Care should be taken when deploying the model in healthcare contexts, particularly regarding transcription errors, privacy, and potential biases in the training data.

Citation

If you use this model in your research, please cite:

Acknowledgements

Developed by YUX Cultural AI Lab / KitalaAI and supported by the GRAIN Network, IPAR, Sunbird AI and (IDRC).