An Efficient Method for Vocabulary Addition to WFST Graphs

Bulusheva, Anna; Zatvornitskiy, Alexander; Korenevsky, Maxim

doi:10.1007/978-3-319-45510-5_52

Cited by 3 publications

(5 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The general mechanism of vocabulary expansion with WFST always involves some modifications of the transducers with symbol table modification. Typically the phoneme set and acoustic modelling stays unchanged and so we only need to modify L and G. Modifying G consists of expanding classbased model using Replace operation [17,19,20,21]. Modifying L might look trivial from the first glance but is actually not a simple problem if we want to keep it minimal [18,27].…”

Section: Wfst-based Vocabulary Expansion Techniquesmentioning

confidence: 99%

“…In [18,19] authors suggested a methods that separated newly added words on groups depending on their boundary phonemes while constructing initial CLG/HCLG with several contextdependent auxiliary symbols that are to be replaced with new words. In [17] a simpler approach was proposed to deal with context dependency:…”

Section: Wfst-based Vocabulary Expansion Techniquesmentioning

confidence: 99%

“…The other approach is to modify HMM part of the system in such way that the vocabulary can be later extended without acoustic and language models retraining. This approach is mostly researched for WFST based systems in [17,18,19,20,21]. In this paper we present a research on methods described in [17,21] and present our own improvements of this approach.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Techniques for Vocabulary Expansion in Hybrid Speech Recognition Systems

Malkovsky,

Bataev,

Sviridkin

et al. 2020

Preprint

View full text Add to dashboard Cite

The problem of out of vocabulary words (OOV) is typical for any speech recognition system, hybrid systems are usually constructed to recognize a fixed set of words and rarely can include all the words that will be encountered during exploitation of the system. One of the popular approach to cover OOVs is to use subword units rather then words. Such system can potentially recognize any previously unseen word if the word can be constructed from present subword units, but also nonexisting words can be recognized. The other popular approach is to modify HMM part of the system so that it can be easily and effectively expanded with custom set of words we want to add to the system. In this paper we explore different existing methods of this solution on both graph construction and search method levels. We also present a novel vocabulary expansion techniques which solve some common internal subroutine problems regarding recognition graph processing.

show abstract

Section: Wfst-based Vocabulary Expansion Techniquesmentioning

confidence: 99%

Section: Wfst-based Vocabulary Expansion Techniquesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Techniques for Vocabulary Expansion in Hybrid Speech Recognition Systems

Malkovsky,

Bataev,

Sviridkin

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…In these systems both the language model and lexicon are fixed and encoded as a WFST, this means words that were not part of these systems at training time are impossible to recognize. This has lead to various approaches to modify the WFSTs so that the ASR system can recognize words it had previously no knowledge of [2,3,4,5,6,7]. A complication is that typically the lexicon and language model WFSTs will be composed together to create a static decoding graph that can be used repeatedly during decoding.…”

Section: Introductionmentioning

confidence: 99%

“…Finally, one can try and modify the static decoding graph (HCLG) [5,6,7]. Because of composition and optimization (e.g.…”

Section: Introductionmentioning

confidence: 99%

A Comparison of Methods for OOV-Word Recognition on a New Public Dataset

Braun

Madikeri

Motlíček

2021

ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

A common problem for automatic speech recognition systems is how to recognize words that they did not see during training. Currently there is no established method of evaluating different techniques for tackling this problem. We propose using the CommonVoice dataset to create test sets for multiple languages which have a high out-of-vocabulary (OOV) ratio relative to a training set and release a new tool for calculating relevant performance metrics. We then evaluate, within the context of a hybrid ASR system, how much better subword models are at recognizing OOVs, and how much benefit one can get from incorporating OOV-word information into an existing system by modifying WFSTs. Additionally, we propose a new method for modifying a subword-based language model so as to better recognize OOVwords. We showcase very large improvements in OOV-word recognition and make both the data and code available.

show abstract

Vocabulary Expansion for the Sub-word WFST-Based Automatic Speech Recognition System

Salimbajevs

Kapočiūtė-Dzikienė

2022

Lecture Notes in Networks and Systems

View full text Add to dashboard Cite

An Efficient Method for Vocabulary Addition to WFST Graphs

Cited by 3 publications

References 9 publications

Techniques for Vocabulary Expansion in Hybrid Speech Recognition Systems

Techniques for Vocabulary Expansion in Hybrid Speech Recognition Systems

A Comparison of Methods for OOV-Word Recognition on a New Public Dataset

Vocabulary Expansion for the Sub-word WFST-Based Automatic Speech Recognition System

Contact Info

Product

Resources

About