AI recognizes voice problems: Maintaining voice health for teachers and speakers
Teachers' voices are constantly put to the test – If they are impaired, it becomes more difficult to carry out their job. An AI tool is designed to prevent this
(Image: brizmaker/Shutterstock.com)
The Fraunhofer Institute for Digital Media Technology (IDMT) presented its self-developed AI tool VoiceTracker at this year's Didacta. It is designed to support the voices of teachers or people in professions with a lot of speaking to keep an eye on their own vocal health and reduce deterioration and voice-related failures. To this end, it monitors the voice in stressful situations over a longer period of time. It should be able to take into account relevant environmental factors such as background noise and the acoustics of the classroom. According to IDMT, there is no transcription of the spoken word or evaluation of the lesson.
The VoiceTracker has been available since spring 2024. In a precursor project, an initial voice analysis was implemented in the context of hearing disorders, which was tested under practical conditions in a usability study in collaboration with the Hannover Medical School (MHH). The project idea of long-term monitoring of teachers was then developed.
The AI used is based on models developed in the Institute's Hearing, Speech and Audio Technology (HSA) department. According to the IDMT, it is trained in such a way that established voice quality assessments from the voice signal are mapped in the best possible way. The system pays attention to various parameters that describe the spectral characteristics of the voice.
Implementation partners wanted
Initial pilots have so far been carried out under laboratory conditions with employees; among other things, the AI was tested in such a way that the results of the voice material were compared with assessments by speech therapy experts. When analyzing voice health, VoiceTracker initially focuses on objectively depictable components that can describe the performance of a voice. However, the system can be expanded to include subjective assessment.
As Fraunhofer IDMT explained to heise online, the institute is now looking for implementation partners who could make the system available. This involves integrating the software into hardware that has yet to be determined. Various solutions are conceivable, from your own s
martphone to a microphone (system) or the use of existing hardware, such as smartboards. Interested parties were able to try out the tool at Didacta. Teachers were also sought who could provide impetus for further development.
Videos by heise
Voice gone, teacher gone
"Early detection and preventative measures are at the heart of the future solution. Teachers receive feedback on their voice quality from our software. This can be used to derive recommendations on how they can specifically protect and care for their voices," explains Alina Ernst, project manager in the Assistive Speech and Language Analysis group.
According to the institute, teachers complain about voice problems twice as often as people who do not work in this profession. Teachers are also more likely to be absent from work due to voice problems than non-teachers. Problems in this area are often recognized too late.
At Didacta, Fraunhofer IDMT presented other AI tools such as the InsightPersona research tool. It is suitable for more precise media analyses of political reporting.
(kbe)
