Public Address (Systems)
Public address systems are installed at many places in order to play speech to an audience using loudspeakers. Sometimes, local background noise affects the speech intelligibility and increases the listening effort. At railway stations, for example, announcements are hardly understandable while fast trains are passing.
If the background noise cannot be reduced or cancelled, which is usually the case, the announcements can only be enhanced by modifying the speech. We adopt a technique, originally developed for from mobile telephony, which is called Near-end listening enhancement. It improves the intelligibility of the speech announcement by adaptively filtering the speech signal and taking into account the local background noise characteristics.
Compared to mobile telephony, further particular problems arise. The background noise needs to be measured close to the audience to allow an adaptation for the current situation. However, for logistical reasons it is often not possible to install microphones on the bottom of the platform. Instead, they are mounted higher in the air, close to the loudspeakers. In addition to the (desired) background noise, the microphone records also (undesired) extremely powerful echoes of the speech announcement. This renders the task of estimating the background noise during announcements very difficult. In [niermann16a], the problem is addressed by means of echo cancellation combined with a new echo-aware noise estimator. This approach updates the noise estimate only in time-frequency bins with sufficiently low echo power and thus allows for robust listening enhancement.
References
[niermann16]
Markus Niermann, Peter Jax, and Peter Vary
Noise Estimation For Speech Reinforcement in the Presence of Strong Echoes
Proceedings of IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2016