Identifying vocal changes due to environmental changes: Real-time Lombard effect

Abstract

School teachers are known to have an elevated risk of voice problems due to speaking demands in the workplace, as well as noise conditions in the classroom. Six elementary school teachers were recorded for a full school day using a neck-worn accelerometer and microphone. Classroom noise levels were also recorded using seven stationary microphones distributed randomly throughout the room. A method was developed that used the accelerometer signal to identify when a teacher is speaking, subsequently calculating a speech-to-noise ratio from six hours of teaching. Speech levels and fundamental frequencies were assessed through the accelerometer signal. The former were compared to temporally and spatially proximate noise levels from the neck-worn microphone when the teacher was not speaking. The speech data was also compared to spatially averaged and temporally proximate classroom noise levels when the teacher was not speaking. While unique in the duration of the observation for the various teachers, the speech-to-noise ratio exhibited trends related to those suggested in previous studies and is congruent with the Lombard effect. A good understanding of how a teacher compensates for realistic occupational noise conditions can lead to better recommendations for the teacher to retain good vocal health. The method and speech trends will be discussed, and future methodological improvements will be suggested.