Steven Rennie, Trausti Kristjansson, Peder Olsen, Ramesh Gopinath
We consider the problem of robust speech recognition in thecar environment. We present a new dynamic noise adaptation al-gorithm, called DNA, for the robust front-end compensation ofevolving semi-stationary noise as typically encountered in the carsetting. A large dataset of in-car noise was collected for the eval-uation of the new algorithm. This dataset was combined with theAurora II framework to produce a new, publicly available frame-work, called DNA + AURORA II, for the evaluation of adaptivenoise compensation algorithms. We show that DNA consistentlyoutperforms several existing, related state-of-the-art front-end de-noising techniques.