VerSpiFah

InterCom systems are used, among other things, to improve voice communication inside the vehicle. Such a system is able to record voice signals from speaking passengers, process them in real time and play them back to improve voice quality and intelligibility for listening passengers in the vehicle.

Acoustic feedback makes the system unstable because the microphones are close to the speakers. Acoustic conditions and noise are other important challenges to be solved in order to increase speech intelligibility and speech quality in the presence of loud noises on the one hand, and to cause the InterCom system not to amplify noises on the other hand. With the development of suitable algorithms, these difficulties can be solved.

The InterCom system operates in a closed electro-acoustic loop. Microphones pick up the speech signals and play them back through the loudspeakers near the listening passengers. The feedback loop can lead to persistent vibrations that manifest themselves in annoying whistling sounds. The conditions for reproducing a speech signal through loudspeakers are difficult due to the high correlation between the direct speech signal and the loudspeaker output. This requires new model-based approaches to identify feedback paths under difficult conditions for feedback compensation and feedback suppression.

Furthermore, speech communication in a moving car is affected by different types of noise (stationary and transient). For strong or impulsive noises, classical filtering methods are not satisfactory. More powerful machine learning methods for speech signal enhancement are therefore being researched and will be used in the system in the future, with the additional goal of reconstructing highly disturbed speech components.