The SHINE project has the goal of investigating distant-talking interaction in noisy and reverberant environment as well as robustness issues in speech recognition. Distributed microphone networks are used to perform acoustic scene analysis in multi-speaker scenarios.

It is well known that the performance of a Speech Recognizer drops dramatically in noisy environments. This fact is further critical when the talker is at a distance of few meters from the microphones and when the room acoustics introduces reverberation effects.

We investigate on this problem for different applicative contexts as smarthome, office, security and surveillance, video-conferencing, etc. processing aimed to derive an acoustic scene analysis (speech activity detection, speaker localization and tracking, acoustic event detection and classification, etc.) in an automatic way.

By applying multi-microphone based solutions as well as specific signal processing techniques and acoustic modeling, we also try to compensate for this, in order to ensure satisfactory speech recognition performance under realistic conditions as, for instance, in the car environment.