Finite-State Transducers in Language and Speech Processing

By Mehryar Mohri. 1997

Subfields: Natural Language Processing, Speech & Audio

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Natural Language Processing

In Speech & Audio

Citations in the corpus

Unsupervised Speech Recognition (2021, methodology): “In practice, decoding is often carried out by searching the minimal distortion path on a weighted finite state transducer (WFST; Mohri 1997; Mohri et al. 2002), obtained by composing WFSTs representing HMM emission, HMM transition, pronunciation model, and language model, respectively.”
Federated Learning for Mobile Keyboard Prediction (2018, methodology): “We train an RNN model from scratch in the server and federated environments and achieve recall improvements with respect to the FST decoder baseline.”
Markovian Models for Sequential Data (2004, methodology): “A generic composition operation [8, 9, 10, 11] allows to combine a cascade of transducers and acceptors, e.”
Sphinx-4: a flexible open source framework for speech recognition (2004, background): “For example, the FrontEnd implementations support MFCC, PLP, and LPC feature extraction; the Linguist implementations support a variety of language models, including CFGs, FSTs, and N-Grams; and the Decoder supports a variety of SearchManager implementations, including traditional Viterbi, Bushderby”
Weighted Finite-State Transducer Algorithms. An Overview (2004, methodology): “Weighted transducers are used in many applications such as text, speech and image processing [9, 12, 5].”
A Tutorial on Energy-Based Learning (2006, background): “The operation of most modules can be expressed as the composition of the input graph with another graph, called a transducer, associated with the module [51].”
Weighted finite-state transducers in speech recognition (2002, methodology): “The more general weighted transducer case can be shown to be equivalent to this case by interpreting weight-output label pairs as new ‘weights’ combined by the appropriate semiring[10].”
A bibliographical study of grammatical inference (2005, background): “Another direction has been followed by Mohri [158].”
Introduction to Digital Speech Processing (2007, methodology): “Fortunately, through the use of methods from the field of Finite State Automata Theory, Finite State Network (FSN) methods have evolved that reduce the computational burden by orders of magnitude, thereby enabling exact maximum likelihood solutions in computationally feasible times, even for very la”
OpenFst: A General and Efficient Weighted Finite-State Transducer Library (2007, background): “This treats string and K outputs uniformly and is the natural representation for weighted transducer determinization, minimization, label pushing, and epsilon normalization [10, 11, 13, 12].”
Bridging CNNs, RNNs, and Weighted Finite-State Machines (2018, background): “From a modeling perspective, SoPa is interesting because WF-SAs are well-studied and come with efﬁcient and ﬂexible inference algorithms (Mohri, 1997; Eis-ner, 2002) that SoPa can take advantage of.”
Long short term memory neural network for keyboard gesture decoding (2015, methodology): “In this work, we address these challenges with a hybrid approach, combining a variant of recurrent networks, namely Long Short Term Memories [1] with conventional Finite State Transducer decoding [2].”
Gradient-based learning applied to document recognition (1998, methodology): “Transducers have been applied to speech recognition [100] and language translation [101], and proposals have been made for handwriting recognition [102].”
Hybrid CTC/Attention Architecture for End-to-End Speech Recognition (2017, background): “Although this integration is often efﬁciently handled by ﬁnite state transducers, the construction and implementation of well-optimized transducers are very complicated [5], [6].”
Probabilistic finite-state machines - part I (2005, background): “A different line was successfully followed in [8], where the probabilities are just a special case of abstract weights: The algebra over which the weights are computed then allows us to deal with all cases, whether computable or not.”
Large Vocabulary Speech Recognition on Parallel Architectures (2013, background): “Weighted Finite-State Transducers (WFST) have been introduced in speech recognition by Mohri [26]–[29] and are now used in state of the art speech recognition systems.”
Graphical Models over Multiple Strings (2009, background): “Pereira and Riley (1997) built a speech recognizer in this way, relating acoustic to phonetic to lexical strings.”
Differentiable Weighted Finite-State Transducers (2020, background): “A wealth of prior work exists using weighted ﬁnite-state automata in speech recognition, natural language processing, optical character recognition, and other applications (Breuel, 2008; Knight & May, 2009; Mohri, 1997; Mohri et al., 2008; Pereira et al., 1994).”