Real-time implementation of a model-based feedback suppression and its adaptive parameter adjustment for In-Car Communication
An In-Car Communication (ICC) system is supposed to record the speech via a microphone located next to a speaker and reproduce it via a loudspeaker next to the listener. An important criteria for feedback suppression is the estimation of the feedback signal in the microphone. Here an effective feedback estimation method based on an already existing energy-decay model can be utilized. The feedback is estimated based on the loudspeaker subband signals delayed by the length of the direct acoustic path between loudspeaker and microphone, room coupling factors and the feedback decay behavior. Often, all the model parameters are extracted offline from the room impulse response and do not change over time.In this Paper, however, a method will be presented that adaptively updates the model parameters caused by changing conditions and environments in real-time. The adaptation of the model parameters is performed only in short speech pauses when the microphone records only acoustic feedbacks. For detecting such parts, a reliable detection method is proposed. The resulting improved method for model-based feedback suppression is realized on a real-time hardware and integrated in a demonstrator vehicle. Several objective and subjective performance tests were eventually carried out.