Researchers reviewed 80 different studies to see how well machine learning can identify multiple types of voice disorders. They looked at how computers are used to analyze voices as signs of other health issues. The review found that while the technology is promising, there is a lot of variation in how these studies are conducted.
Several problems were identified that make it hard to use these tools in real clinics today. These include inconsistent ways of labeling disorders, different amounts of data for each condition, and varying methods for testing the software. Because every study uses different rules, it is currently difficult to compare results or pick a single best method.
Because this was a scoping review of existing research rather than a clinical trial, these findings do not mean the technology won't work in the future. Instead, they show that researchers need more consistency before these tools can be used reliably by doctors. For now, machine learning is still in an early stage of development for diagnosing voice problems.