In this paper we present an algorithm to compensate the dynamic cepstral coefficients. This extends the parallel model combination scheme to more general cases where dynamic cepstral coefficients are calculated by linear regression of any length. The algorithm has been applied to a speech database which was recorded in a car and improvement was observed. The performance is improved further by using prefiltering.
Bibliographic reference. Yang, Ruikang / Majaniemi, Markku / Haavisto, Petri (1995): "Dynamic parameter compensation for speech recognition in noise", In EUROSPEECH-1995, 469-472.