Respiratory illnesses can be hard to track and diagnose. Obtaining useful clinical data on these illnesses is difficult because it requires physical interaction, e.g., via nasal or sinus swab. It is known that respiratory illness can impact speech pathways. To this end, this disclosure describes techniques to use readily accessible software to obtain and classify potentially useful data. With user permission, utterances of the user, e.g., activation of a speech-activated device via a hotword, are analyzed to form speaker-ID models. These models are evaluated against additional utterances of the user in a sequential manner. The evaluation scores, along with the timestamps and details of the models, are aggregated to determine if the user has an interval of time where their speaker-ID models are unstable, inconsistent, or lacking self-similarity. This signal can be used as a proxy for detection or as a motivating factor for clinical investigation.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Kracun, Aleksandar and Moreno, Ignacio Lopez, "Health Diagnostics Using User Utterances", Technical Disclosure Commons, (May 07, 2020)