digital-health-clinical-asr-finetune

Fine-tunes a clinical speech-recognition model and compares offline evaluation results to decide whether to keep it or expand the training data.

When to use it

Use when priority-category KER exceeds 0.3 and the manifest contains at least 100 rows, including at least five per priority category.

Give it a clinical manifest, its audio, and baseline evaluation results; it produces a fine-tuned model and an offline comparison against the baseline.