jcho02/whisper_cleft

Model Card for Fine-Tuned Whisper Model: Hypernasality Identification

This model card details the Whisper model, fine-tuned to identify hypernasality in patients from audio recordings. The model is developed to assist in the diagnosis and monitoring of speech disorders.

• Model Interactive Demo: https://huggingface.co/spaces/jcho02/Transformers_whisper_cleft

Model Details

Model Description

• Developed by: The Data Science team at the Vanderbilt Data Science Institute
• Funded by: This project was supported by healthcare-focused research grants and institutional funding.
• Shared by: The model is shared by Vanderbilt University with the aim to assist clinicians and speech therapists.
• Model type: Deep Learning, Neural Network (Fine-Tuned Whisper Model)
• Language(s) (NLP): The model primarily focuses on English language audio samples but has the potential to be adapted for multilingual use.
• License: The model is released under an open-source license for non-commercial use, details of which can be found in the repository.
• Finetuned from model: The base model is OpenAI's Whisper, which has been fine-tuned using a specialized dataset for hypernasality detection.

Model Sources

• Repository: https://github.com/jaewoocho/Transformers_whisper_cleft/tree/main?tab=readme-ov-file
• Demo : https://huggingface.co/spaces/jcho02/Transformers_whisper_cleft

Uses

Direct Use

• This model can be directly used by healthcare professionals to analyze patient speech samples for signs of hypernasality, aiding in diagnosis and treatment planning.

Downstream Use

• Potential use in automated speech therapy tools or in research settings for studying speech disorders.

Out-of-Scope Use

• Not intended for non-medical use or as a sole diagnostic tool without clinical oversight.

Bias, Risks, and Limitations

• The model might have limitations in accurately identifying hypernasality across diverse accents and languages. User discretion is advised, and results should be considered alongside clinical assessments.

Recommendations

• Healthcare professionals should use this tool as a supplement to traditional diagnostic methods. Continuous updating with diverse datasets is recommended to improve accuracy.

How to Get Started with the Model

• Use the following code snippet to get started with the model:
• https://github.com/jaewoocho/Transformers_whisper_cleft/tree/main?tab=readme-ov-file