Výhody včasnej diagnostky Alzheimerovej choroby pomocou sociálneho robota a umelej inteligencie

At the Institute of Informatics of the SAV, they are exploring how speech recognition and synthesis technologies can help in healthcare. Róbert Savo presented projects that assess speech and voice in detecting neurodegenerative diseases and elevate communication to the level of social robots. The aim is not to replace doctors, but to provide them with quick and precise supporting data.

From speech recognition to social robots

Speech recognition converts spoken words into text, while synthesis is the opposite direction—creating a natural-sounding voice from text. Combining these two technologies creates the foundation for communication between human and machine that is meant to be comfortable and comprehensible. This kind of “natural” communication is the core of social robotics, where the machine conducts a dialogue in a human manner.

The team also uses these tools to support the diagnosis of neurodegenerative diseases. Thanks to the analysis of voice and speech, subtle changes can be detected that a person would not normally notice. The result is models that help uncover the risk of Alzheimer’s or Parkinson’s disease, or mild cognitive impairment.

The EVA screening app and a conversation with a robot

A completed project resulted in the EVA application for rapid screening. In it, the user names simple pictures as well as one complex one, while the system analyzes the recording in terms of acoustic and linguistic features. At the end, the user receives information on whether everything is in order or whether they should seek a specialist. EVA therefore does not provide a diagnosis, but purposefully recommends the next step.

The latest advance has come from social robotics: the patient no longer taps on a mobile phone but talks to a robot. A microphone captures the speech, a recognizer converts it into text, a dialogue manager selects a response, and synthesis reads it out with lip articulation. Experience from tests shows that people perceive this mode of communication as pleasant and natural, which is also confirmed by predominantly positive feedback. For many participants, the conversation with the „plastic head“ was surprisingly relaxed and understandable.

What AI monitors in speech and why the doctor decides

Artificial intelligence measures, for example, reaction times on the order of hundredths to thousandths of a second and monitors the fluency of speech. It can precisely count pauses and hesitations and evaluate intonation, that is, the melody of speaking. It also assesses lexical variability, semantic density, and the degree of “crutches” in speech, while the acoustic parameters of the voice also play an important role.

These key indicators can be evaluated in real time, giving the doctor an overview of specific performance in the given task. The models are created from a sufficient amount of annotated data from healthy and ill individuals, but their role is support, not the final verdict. The team emphasizes that AI should not make a binary decision „healthy/sick“. The diagnosis is determined by the doctor based on facts, to which the technology adds precise and rapid measurements.

Download presentation (353kB)