Posterior - Based Keyword Spotting Approaches Without Filler Models — Iterative Viterbi Decoding and One - Pass Approaches —

This paper addresses the problem of detecting keywords in unconstrained speech without explicit modeling of non-keyword segments. The proposed algorithms are based on recent developments in confidence measures using local posterior probabilities, and searches for the segment maximizing the average observation posterior along the most likely path in the hypothesized keyword model. We can also use more complex matching scores. As known, this approach (sometimes referred to as sliding model method) requires a relaxation of the begin/endpoints of the Viterbi matching, as well as a time normalization of the resulting score, making dynamic programming sub-optimal or more complex (more computation and/or more memory). We present here alternative (quite simple and efficient) solutions to this problem. One of them uses an iterative form of Viterbi decoding algorithm, but which does not require scoring for all possible begin/endpoints. Convergence proof of this algorithm is available. Results obtained with this method on 100 keywords chosen at random from the BREF database [5] are reported. Finally, we also develop a new pruning strategy of a one-pass approach that can optimize the same criteria, and yields the same recognition performance than the iterative Viterbi approach. This report is an extension of [8], where a detailed convergence proof of the Iterative Viterbi Decoder is provided, together with an extension to the one-pass decoder.