Gaussian Selection with Non-Overlapping Clusters for ASR in Embedded Devices

Optimizing components for handheld two-way speech translation for an English-iraqi Arabic system

“…A Gaussian selection algorithm is used to speed up the evaluation of the acoustic model. As a small modification compared to [7], we generated the disjunct clusters using Euclidean distance, and used Mahalanobis distance between a feature vector and the cluster centroids to find the active clusters for each frame during decoding. The covariance of each centroid is calculated by averaging over all Gaussians' covariances that belong to the centroid's cluster.…”

Efficient codebooks for fast and accurate low resource ASR systems

“…Gaussians belonging to this shortlist contribute to that frame likelihood computation. Many extensions of this work have been proposed in the literature (Gales et al (1999); Olsen (2000); Leppänen et al (2006)). They were focused on Gaussian codebook assignments.…”