Minimum Classification Error for Large Scale Speech Recognition Tasks using Weighted Finite State Transducers

AbstractWeighted finite-state transducers have been shown to be a general and efficient representation in many applications such as text and speech processing, computational biology, and machine learning. The composition of weighted finite-state transducers constitutes a fundamental and common operation between these applications. The NP-hardness of the composition computation problem presents a challenge that leads us to devise efficient algorithms on a large scale when considering more than two transducers. This paper describes a parallel computation of weighted finite transducers composition in MapReduce framework. To the best of our knowledge, this paper is the first to tackle this task using MapReduce methods. First, we analyze the communication cost of this problem using Afrati et al. model. Then, we propose three MapReduce methods based respectively on input alphabet mapping, state mapping, and hybrid mapping. Finally, intensive experiments on a wide range of weighted finite-state transducers are conducted to compare the proposed methods and show their efficiency for large-scale data.

Download Full-text

Joint optimization on decoding graphs using minimum classification error criterion

APSIPA Transactions on Signal and Information Processing ◽

10.1017/atsip.2014.5 ◽

2014 ◽

Vol 3 ◽

Author(s):

Abdelaziz A. Abdelhamid ◽

Waleed H. Abdulla

Keyword(s):

Likelihood Estimation ◽

Discriminative Training ◽

Language Models ◽

Classification Error ◽

Error Criterion ◽

Speech Corpora ◽

Finite State ◽

Minimum Classification Error ◽

Speech Features ◽

Weighted Finite State Transducers

Motivated by the inherent correlation between the speech features and their lexical words, we propose in this paper a new framework for learning the parameters of the corresponding acoustic and language models jointly. The proposed framework is based on discriminative training of the models' parameters using minimum classification error criterion. To verify the effectiveness of the proposed framework, a set of four large decoding graphs is constructed using weighted finite-state transducers as a composition of two sets of context-dependent acoustic models and two sets of n-gram-based language models. The experimental results conducted on this set of decoding graphs validated the effectiveness of the proposed framework when compared with four baseline systems based on maximum likelihood estimation and separate discriminative training of acoustic and language models in benchmark testing of two speech corpora, namely TIMIT and RM1.

Download Full-text

A multiplatform speech recognition decoder based on weighted finite-state transducers

2009 IEEE Workshop on Automatic Speech Recognition & Understanding ◽

10.1109/asru.2009.5373404 ◽

2009 ◽

Cited By ~ 1

Author(s):

Emilian Stoimenov ◽

Tanja Schultz

Keyword(s):

Speech Recognition ◽

Finite State Transducers ◽

Finite State ◽

Weighted Finite State Transducers

Download Full-text

Speech Recognition Algorithms Using Weighted Finite-State Transducers

Synthesis Lectures on Speech and Audio Processing ◽

10.2200/s00462ed1v01y201212sap010 ◽

2013 ◽

Vol 9 (1) ◽

pp. 1-162 ◽

Cited By ~ 6

Author(s):

Takaaki Hori ◽

Atsushi Nakamura

Keyword(s):

Speech Recognition ◽

Finite State Transducers ◽

Recognition Algorithms ◽

Finite State ◽

Weighted Finite State Transducers

Download Full-text

Composition of Weighted Finite Transducers in MapReduce

10.21203/rs.3.rs-101167/v1 ◽

2020 ◽

Author(s):

Bilal Elghadyry ◽

Faissal Ouardi ◽

Sébastien Verel

Keyword(s):

Speech Processing ◽

Large Scale ◽

Large Scale Data ◽

Finite State Transducers ◽

Wide Range ◽

Finite State ◽

Common Operation ◽

Efficient Representation ◽

Weighted Finite State Transducers ◽

Np Hardness

Abstract Weighted finite-state transducers have been shown to be a general and efficient representation in many applications such as text and speech processing, computational biology, and machine learning. The composition of weighted finite-state transducers constitutes a fundamental and common operation between these applications. The NP-hardness of the composition computation problem presents a challenge that leads us to devise efficient algorithms on a large scale when considering more than two transducers. This paper describes a parallel computation of weighted finite transducers composition in MapReduce framework. To the best of our knowledge, this paper is the first to tackle this task using MapReduce methods. First, we analyze the communication cost of this problem using Afrati et al. model. Then, we propose three MapReduce methods based respectively on input alphabet mapping, state mapping, and hybrid mapping. Finally, intensive experiments on a wide range of weighted finite-state transducers are conducted to compare the proposed methods and show their efficiency for large-scale data.

Download Full-text