New pruning criteria for efficient decoding
نویسنده
چکیده
In large vocabulary continuous speech recognizers the search space needs to be constrained efficiently to make the recognition task feasible. Beam pruning and restricting the number of active paths are the most widely applied techniques for this. In this paper, we present three additional pruning criteria, which can be used to further limit the search space. These new criteria take into account the state of the search space, which enables tighter pruning. In the speech recognition experiments, the new pruning criteria were shown to reduce the search space up to 50% without affecting the search accuracy. We also present a method for optimizing the threshold parameters of the pruning criteria for the selected level of recognition accuracy. With this method even a large number of different pruning thresholds can be determined with little effort.
منابع مشابه
Start- and end-node segmental-HMM pruning
An efficient decoding algorithm for segmental HMMs (SHMMs) is proposed with multi-stage pruning. The generation by SHMMs of a feature trajectory for each state expands the search space and the computational cost of decoding. It is reduced in three ways: pre-cost partitioning, start-node (SN) beam pruning, and conventional endnode (EN) beam pruning. Experiments show that partitioning cuts comput...
متن کاملBeam-Width Prediction for Efficient Context-Free Parsing
Efficient decoding for syntactic parsing has become a necessary research area as statistical grammars grow in accuracy and size and as more NLP applications leverage syntactic analyses. We review prior methods for pruning and then present a new framework that unifies their strengths into a single approach. Using a log linear model, we learn the optimal beam-search pruning parameters for each CY...
متن کاملSearch Based Weighted Multi-Bit Flipping Algorithm for High-Performance Low-Complexity Decoding of LDPC Codes
In this paper, two new hybrid algorithms are proposed for decoding Low Density Parity Check (LDPC) codes. Original version of the proposed algorithms named Search Based Weighted Multi Bit Flipping (SWMBF). The main idea of these algorithms is flipping variable multi bits in each iteration, change in which leads to the syndrome vector with least hamming weight. To achieve this, the proposed algo...
متن کاملReducing computation on parallel decoding using frame-wise confidence scores
Parallel decoding based on multiple models has been studied to cover various conditions and speakers at a time on a speech recognition system. However, running many recognizers in parallel applying all models causes the total computational cost to grow in proportion to the number of models. In this paper, an efficient way of finding and pruning unpromising decoding processes during search is pr...
متن کاملPruning Convolutional Neural Networks for Resource Efficient Inference
We propose a new formulation for pruning convolutional kernels in neural networks to enable efficient inference. We interleave greedy criteria-based pruning with finetuning by backpropagation—a computationally efficient procedure that maintains good generalization in the pruned network. We propose a new criterion based on Taylor expansion that approximates the change in the cost function induce...
متن کامل