Two-pass Continuous Digit String Decoder
نویسندگان
چکیده
In this paper, we present a two-pass continuous digit string decoder using two sets of whole-word HMM models. One set contains context-independent (CI) models used in the first-pass search. The first-pass search results in N-best hypotheses from which a N-best word lattice can be derived. The other set contains context-dependent (CD) HMM models used to search along the N-best word lattice for the best hypothesis, which is called the second-pass search. During the second-pass search, we introduce a tree-structured word lattice to speed up the second-pass search. Compared with one-pass decoder using only CI models, our two-pass decoder achieves 68% reduction of word error rate. Compared with one-pass decoder using only CD models, it achieves a 6.5 times faster search speed. Compared with two-pass decoder using flat-structured word lattice, it achieves about one time faster search speed.
منابع مشابه
Efficient 2-pass n-best decoder
In this paper, we describe the new BBN BYBLOS efcient 2-Pass N-Best decoder used for the 1996 Hub-4 Benchmark Tests. The decoder uses a quick fastmatch to determine the likely word endings. Then in the second pass, it performs a time-synchronous beam search using a detailed continuous-density HMM and a trigram language model to decide the word starting positions. From these word starts, the dec...
متن کاملEvaluation of a Stack Decoder on a Japanese Newspaper Dictation Task
This paper describes the evaluation of the !V$N$>$_!W stack decoder for LVCSR on a 5000 word Japanese newspaper dictation task [3]. Using continuous density acoustic models with 2000 and 3000 states trained on the JNAS/ASJ corpora and a 3-gram LM trained on the RWC text corpus, both models provided by the IPA group, it was possible to reach more than 95% word accuracy on the standard test set. ...
متن کاملContinuous Digit Recognition in Noise: Reservoirs can do an excellent job!
In this paper a formerly proposed continuous digit recognition system based on Reservoir Computing (RC) is improved in two respects: (1) the single reservoir is substituted by a stack of reservoirs, and (2) the straightforward mapping of reservoir outputs to state likelihoods is replaced by a trained non-parametric mapping. Furthermore, it is shown that a reservoir-based method can improve a mo...
متن کاملImproved methods for vocal tract normalization
This paper presents improved methods for vocal tract normalization (VTN) along with experimental tests on three databases. We propose a new method for VTN in training: By using acoustic models with single Gaussian densities per state for selecting the normalization scales it is avoided that the models learn the normalization scales of the training speakers. We show that using single Gaussian de...
متن کاملUniversity of Warsaw Faculty of Mathematics , Informatics and Mechanics Marek Biskup
In compressed data a single bit error propagates because of the corruption of the decoder’s state. This work is a study of error resilience in compressed data and, in particular, of the recovery of as much data as possible after a bit error. It is focused on Huffman codes. In a message encoded with a Huffman code a bit error causes the decoder to lose synchronization with the coder. The error p...
متن کامل