Multiple-Pass Search Strategies

Schwartz, Richard L.; Nguyen, Long; Makhoul, John

doi:10.1007/978-1-4613-1367-0_18

Cited by 31 publications

(16 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The decoding is done by the two-pass (forward and backward) beam search [1,19]. In addition, we use a trigram language model to guide the path search in which the probabilities of all triplets of words in the training dataset are computed.…”

Section: Hmm Recognitionmentioning

confidence: 99%

Applying Discriminatively Optimized Feature Transform for HMM-based Off-Line Handwriting Recognition

Chen

Zhang

Cao

et al. 2012

2012 International Conference on Frontiers in Handwriting Recognition

View full text Add to dashboard Cite

Feature extraction is an important step in off-line handwriting recognition systems to represent raw handwriting in a low-dimensional, tractable feature space. Traditionally, linear feature transforms such as Principle Component Analysis (PCA), Linear Discriminative Analysis (LDA) are commonly used. The assumptions they make, however, usually cannot be satisfied in practice and thus the best performance is not obtained. In this paper, we apply the Region-Dependent nonlinear feature Transform (RDT) to handwriting recognition. RDT is one type of non-linear feature transforms which captures the discriminating power much better than traditional linear ones. We justify the effectiveness of RDT on handwriting features using an HMM-based handwriting recognition system on an Arabic handwriting dataset, which consists of 38K pages of handwriting, over 3M handwritten words. Experimental results show that RDT is able to decrease the word error rates (WERs) relatively by 4% to 7% with statistical significance, comparing to two LDA-based baseline systems. * Jin Chen performed this work during his summer internship at Raytheon BBN Technologies.

show abstract

Section: Hmm Recognitionmentioning

confidence: 99%

Applying Discriminatively Optimized Feature Transform for HMM-based Off-Line Handwriting Recognition

Chen

Zhang

Cao

et al. 2012

2012 International Conference on Frontiers in Handwriting Recognition

View full text Add to dashboard Cite

show abstract

“…This requires multiple copies of word models to account for different contexts and can be extremely resource-intensive for large vocabulary tasks. Approaches taken to address this problem include the use of dynamic search structures [8] and multipass search where more simple, early passes direct the later passes [9].…”

Section: A Viterbi Decodingmentioning

confidence: 99%

“…Such algorithms have been investigated by Bahl et al [1], [12]- [14], Paul [15], Kenny et al [16], and Soong and Huang [17]. In the statistical framework for speech recognition, (8) may be expressed as (9) The first term on the right hand side, , corresponds to in (8) and is the joint log likelihood of the sequence of acoustic vectors to time and the word sequence hypothesized to that time. The second term, , corresponds to in (8) and-in this case-is the sum over all possible word sequences that can follow while accounting for the remaining acoustic evidence .…”

Section: B Stack Decodingmentioning

confidence: 99%

Start-synchronous search for large vocabulary continuous speech recognition

Renals

Hochberg²

1999

IEEE Trans. Speech Audio Process.

View full text Add to dashboard Cite

Abstract-In this paper, we present a novel, efficient search strategy for large vocabulary continuous speech recognition. The search algorithm, based on a stack decoder framework, utilizes phone-level posterior probability estimates (produced by a connectionist/hidden Markov model acoustic model) as a basis for phone deactivation pruning-a highly efficient method of reducing the required computation. The single-pass algorithm is naturally factored into the time-asynchronous processing of the word sequence and the time-synchronous processing of the hidden Markov model state sequence. This enables the search to be decoupled from the language model while still maintaining the computational benefits of time-synchronous processing. The incorporation of the language model in the search is discussed and computationally cheap approximations to the full language model are introduced. Experiments were performed on the North American Business News task using a 60 000 word vocabulary and a trigram language model. Results indicate that the computational cost of the search may be reduced by more than a factor of 40 with a relative search error of less than 2% using the techniques discussed in the paper.Index Terms-Hidden Markov model, large vocabulary continuous speech recognition, phone deactivation pruning, search, stack decoding.

show abstract

“…In general, multi-pass methods (Schwartz et al 1996) work in two or more steps: in the first pass the less likely hypotheses are discarded because of some condition requiring low computational time. Then, in the later passes, only the remaining hypotheses are examined by more complex, reliable evaluations, which will approximate the probabilities of the hypotheses more closely.…”

Section: Multi-pass Search Strategiesmentioning

confidence: 99%

“…To speed up the search process, various ideas are available in the literature: we can do a multi-pass search (Schwartz et al 1996), apply search heuristics (Gosztolya et al 2003), or try out other similar ideas. However, if we were to improve the recognition accuracy by methods not requiring additional computational time, we would have more room to achieve a speed-up while maintaining the same level of accuracy.…”

Section: Introductionmentioning

confidence: 99%

The use of speed-up techniques for a speech recognizer system

Kocsor

Gosztolya

2006

Int J Speech Technol

View full text Add to dashboard Cite

In speech recognition, not just the accuracy of an automatic speech recognition application is important, but also its speed. However, if we want to create a real-time speech recognizer, this requirement limits the time that is spent on searching for the best hypothesis, which can even affect the recognition accuracy. Thus the applied search method plays an important role in the speech recognition task, and so does its efficiency, i.e. how quickly it finds the uttered words. To speed up this search process, various ideas are available in the literature: we can use search heuristics, multi-pass search, or apply a family of aggregation operators. In this paper we test all these methods in turn, and combine them with a set of other novel speed-up ideas. The test results confirm that all of these techniques are valuable: using combinations of them helped make the speech recognition process over 12 times faster than the basic multi-stack decoding algorithm, and almost 11 times faster than the Viterbi beam search method.

show abstract

Multiple-Pass Search Strategies

Cited by 31 publications

References 4 publications

Applying Discriminatively Optimized Feature Transform for HMM-based Off-Line Handwriting Recognition

Applying Discriminatively Optimized Feature Transform for HMM-based Off-Line Handwriting Recognition

Start-synchronous search for large vocabulary continuous speech recognition

The use of speed-up techniques for a speech recognizer system

Contact Info

Product

Resources

About