Using a Speech Codec to Suppress Howling in Public Address Systems

Acoustical feedback is present whenever a loudspeaker signal gets redirected to a microphone that feeds its input signal directly or indirectly back into the loudspeaker. If the gain around such a closed feedback loop is close to or higher than unity, unpleasant acoustical artifacts will typically occur and will nearly always lead to a periodic howling sound. Most readers are probably familiar with this noise, which can for example set in when a microphone is accidentally pointed at a speaker. This research project aims to suppress these unwanted effects of acoustical feedback by the insertion of a modified speech coder and decoder into the signal path of the feedback loop. It is demonstrated that the Speex open-source speech codec can be successfully tweaked to increase the maximum stable feedback gain by as much as 3 dB to 7 dB through adjustment of the codec's quality parameter. This enhancement outperforms the simple introduction of shaped noise into the feedback loop and is compared with the performance of a frequency shifter. Tests are conducted using an automated experimental framework for determining the maximum stable gain of a public address system.