Fig. 2: Overview of the AI pipeline used to preprocess and model unstructured audio data.
From: Generative AI and unstructured audio data for precision public health
The pipeline included the following steps: (1) The Whisper-Large model was used to transcribe the recorded first-person accounts of COVID-19 infections, (2) the o1 large language model was used to generate a filtered summary of the transcript, removing terminology that could have compromised the simulation of an early-stage outbreak, (3) the summaries were embedded using the text-embedding-3-large model, (4) a neural network was trained for variant classification.
