The research interest in technologies for supporting people in their own homes is constantly increasing. In this context this paper proposes a speech-interfaced system for recognizing home automation commands and distress calls. The robustness of the system is increased by employing Power Normalized Cepstral Coefficients as features and by using an adaptive algorithm to reduce known sources of interference. In addition, the mismatch introduced by vocal effort variability is reduced employing a vocal effort classifier and multiple acoustic models. The performance has been evaluated on ITAAL, a recently proposed corpus of home automation commands and distress calls in Italian. The results confirm that the adopted solutions are effective to be employed in a distorted acoustic scenario.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.