Combined effect of reverberation and noise on binaural speech recognition and vocal effort in real classroom acoustics

Good acoustics in classrooms is necessary to guarantee appropriate teaching and learning practices. The majority of classrooms in Italy does not comply with national or international standards since they take place in historical buildings with big volumes and vaulted ceilings, generating unfavorable environments. In fact, long reverberation times and excessive noise levels have been proved to be factors that negatively influence the speech communication process. However, most of the past studies refer to laboratory conditions with artificial added reverberation and speech-shaped noise.

This work focuses on investigating the influence of reverberation and noise in real classrooms on speech recognition. Speech recognition was measured adaptively converging to signal-to-noise ratio yielding 80% correct recognition scores (SRT80). Since the noise level was fixed at 60 dB SPL, the speech level can be estimated from the measured SRTs and linked to the vocal effort that a teacher would need to maintain to guarantee a good level of speech communication.

Five experiments were designed based on realistic receiver positions in two representative Italian classrooms, one with an acoustical treatment and one without, where room impulse responses were measured at a head and torso simulator ears. Receiver positions in axis with a speech-source were two in the room with good acoustics, at 1.5 m and at 4 m, and three in the room with bad acoustics, at 1.5 m, 4 m and 6.3 m. In each room noise-sources were placed at different distances and azimuths with respect to the receivers to account for binaural cues in the cocktail party phenomenon, namely at 0°, 120° and 180°. Babble noise was measured in real classrooms during a break between lessons, acquiring the noise produced by children moving and talking. The respective impulse responses were convolved with speech and noise signals of Italian matrix test.

Preliminary results show that lower SRT80s were measured for good room acoustics, indicating the detrimental effect of reverberation on speech recognition. A major effect of the reflections was also evident in the increase in SRT80s when the distance between speech-source and receiver increased. The angular separation of the noise-source was evaluated as speech release from masking, resulting up to 4 dB when noise came from 120° instead of 0° or 180° in both room acoustics. Estimations of the speech level will be performed and compared to the indications of ISO 9921, since a high vocal effort can be a risk factor for vocal health.