There are two categories of parameters that are assessed in the analysis of voice biomarkers. These are acoustic (frequency, amplitude, tone, pitch) and prosodic (pause length, speech rate, vowel duration) characteristics. Linguistic features can also be added to them. All these parameters can indicate the state of the human body.
Machine learning algorithms, based on the obtained parameters, can make a preliminary diagnosis. This will speed up the process of diagnosing the disease, and the cost of such a technique is minimal.
Of course, to extract and identify voice biomarkers, you need to go through several stages. Select the type of recording, collect audio data (for this, patients are asked to tell something about themselves or read out a text), select the sound parameters necessary for training the neural network. It is important that the training is accurate – this reduces the likelihood of an artificial intelligence error. The technology will also help in the diagnosis of COVID-19.
Scientists from different countries and universities are involved in the implementation of the project. For example, in the United States, MIT (Massachusetts Institute of Technology) employees have achieved that the voice of AI determined the symptoms of Alzheimer’s disease, and later – the coronavirus. Sonde Health has created a methodology that can be used to detect depression through a smartphone app.