Do I sound sick?

Our last Editorial discussed the need for greater public voices in development of digital health technologies. Rather ironically, this issue journal, Harry Coppock and colleagues comment on how are instead being used to diagnose COVID-19. Vocal patterns such as pitch, tone, rhythm, rate well breathing coughing can be processed by artificial intelligence (AI) develop vocal biomarkers disease. AI-powered developed a multitude disorders including Parkinson's disease, cardiovascular post-traumatic stress disorder. It has also been suggested that voice technology could predict body-mass index. The advent Amazon Alexa, Google Assistant, Apple's Siri helped grow audio data with it, potential advance transform myriad industries care. But is it possible illness from your voice? Coppock, lead author Comment told Lancet Digital Health “These technologies have dramatically improve global through rapid inexpensive diagnosis run personal devices—the perfect tool against pandemic.” However, highlight further evidence needed realise full mass test list seven key issues biomarker research, which prevent clinically useful diagnosing These include possibility AI cannot discriminate between respiratory disorders; subject many undetectable biases; there lack relevant ground truth; external validation still rare current research due available data. As any diagnostic tool, prospective randomised controlled trials real-world evaluation necessary evaluate before scaling up its use. relative simplicity delivering these tools, companies might take their directly public. One company, Sonde Health, uses signal processing machine learning detect symptoms asthma chronic obstructive pulmonary an early warning system In July, 2021, announced new partnership Qualcomm potentially bring smartphones smart speakers, help reach far broader audience. Though soon accessed mobile phone, ethical technological challenges overcome they proven health. Guy Fagherazzi, Group Leader Deep Phenotyping Research Unit at Luxembourg Institute indeed very promising non-invasive tools screening or triage, but reality safety performance algorithms needs confirmed.” Without proper evaluation, engender false sense security public, overwhelm services. Fagherazzi says, “Vocal embedding (smartphone app, assistant, mirrors...) should co-designed end users. And norm, based what I observe market. Acceptability patients general not earned yet.” To generate trust acceptability, privacy protection must core strictly enforced regulation voice-data critical. Voice reveal person's gender, age, language, geographical socio-cultural origins, particularly risk given recent attempts using deepfake clones attempted cybercrime. result, Europol, EU's law enforcement agency, urged member states make “significant investments” deepfakes. some promise field abundance investment exuberance, long way go endorsed clinical community. Better technical standards required fully detection COVID-19 audio: grains saltDigital testing via phone application made ability identify appears confer unique features produced infected individuals,1 breath, cough, speech recordings yielded results.2–4 critique, we present major argue investigation conclusions about detectability made. Full-Text PDF Open Access