The assessment of the acoustic quality of learning spaces traditionally concerns rooms’ objective parameters and not the dynamical behavior of students during lectures. In the present work, student activity and speech levels measured in two historical university lecture halls are used to assess the quality of the acoustic environment before and after renovation works. Restoration includes both acoustic treatments and public address (PA) redesign based on line array system. Clustering techniques, Gaussian Mixture Model and K-means, were used to measure the student activity and the speech levels after long-term monitoring of active classrooms.Outcomes show how acoustic treatments, and especially the PA redesign through line arrays, keep the environment quieter. Consequently, the signal-to-noise ratio – assumed in highly occupied large halls as the difference between the speech level and the student activity – increases even considering the variation of occupancy. This is probably due to the uniform coverage of PA within the audience area.Spectral matching confirms the reliability of both unsupervised methods showing two different speech shapes: the anechoic teachers’ speech, and the students’ speech-in-noise. Nevertheless, K-means clustering seems to be more robust to changes in levels than Gaussian Mixture Model.