Long Short-Term Memory networks for noise robust speech recognition

Martin Wöllmer; Yang Sun; Florian Eyben; Björn Schuller

Long Short-Term Memory networks for noise robust speech recognition

Martin Wöllmer
, Yang Sun
, Florian Eyben
, Björn Schuller

Chair of Health Informatics

Technical University of Munich

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

2 Scopus citations

Abstract

In this paper we introduce a novel hybrid model architecture for speech recognition and investigate its noise robustness on the Aurora 2 database. Our model is composed of a bidirectional Long Short-Term Memory (BLSTM) recurrent neural net exploiting long-range context information for phoneme prediction and a Dynamic Bayesian Network (DBN) for decoding. The DBN is able to learn pronunciation variants as well as typical phoneme confusions of the BLSTM predictor in order to compensate signal disturbances. Unlike conventional Hidden Markov Model (HMM) systems, the proposed architecture is not based on Gaussian mixture modeling. Even without any feature enhancement, our BLSTM-DBN system outperforms a baseline HMM recognizer by up to 18%.

Original language	English
Title of host publication	Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010
Publisher	International Speech Communication Association
Pages	2966-2969
Number of pages	4
State	Published - 2010

Publication series

Name	Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010

Keywords

Dynamic Bayesian Networks
Long Short-Term Memory
Noise robustness
Recurrent neural networks
Speech recognition

Cite this

Wöllmer, M., Sun, Y., Eyben, F., & Schuller, B. (2010). Long Short-Term Memory networks for noise robust speech recognition. In Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010 (pp. 2966-2969). (Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010). International Speech Communication Association.

Wöllmer, Martin ; Sun, Yang ; Eyben, Florian et al. / Long Short-Term Memory networks for noise robust speech recognition. Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010. International Speech Communication Association, 2010. pp. 2966-2969 (Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010).

@inproceedings{41fb043824934d2ba1f37b5051088e73,

title = "Long Short-Term Memory networks for noise robust speech recognition",

abstract = "In this paper we introduce a novel hybrid model architecture for speech recognition and investigate its noise robustness on the Aurora 2 database. Our model is composed of a bidirectional Long Short-Term Memory (BLSTM) recurrent neural net exploiting long-range context information for phoneme prediction and a Dynamic Bayesian Network (DBN) for decoding. The DBN is able to learn pronunciation variants as well as typical phoneme confusions of the BLSTM predictor in order to compensate signal disturbances. Unlike conventional Hidden Markov Model (HMM) systems, the proposed architecture is not based on Gaussian mixture modeling. Even without any feature enhancement, our BLSTM-DBN system outperforms a baseline HMM recognizer by up to 18\%.",

keywords = "Dynamic Bayesian Networks, Long Short-Term Memory, Noise robustness, Recurrent neural networks, Speech recognition",

author = "Martin W{\"o}llmer and Yang Sun and Florian Eyben and Bj{\"o}rn Schuller",

year = "2010",

language = "English",

series = "Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010",

publisher = "International Speech Communication Association",

pages = "2966--2969",

booktitle = "Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010",

}

Wöllmer, M, Sun, Y, Eyben, F & Schuller, B 2010, Long Short-Term Memory networks for noise robust speech recognition. in Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010. Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010, International Speech Communication Association, pp. 2966-2969.

Long Short-Term Memory networks for noise robust speech recognition. / Wöllmer, Martin; Sun, Yang; Eyben, Florian et al.
Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010. International Speech Communication Association, 2010. p. 2966-2969 (Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Long Short-Term Memory networks for noise robust speech recognition

AU - Wöllmer, Martin

AU - Sun, Yang

AU - Eyben, Florian

AU - Schuller, Björn

PY - 2010

Y1 - 2010

N2 - In this paper we introduce a novel hybrid model architecture for speech recognition and investigate its noise robustness on the Aurora 2 database. Our model is composed of a bidirectional Long Short-Term Memory (BLSTM) recurrent neural net exploiting long-range context information for phoneme prediction and a Dynamic Bayesian Network (DBN) for decoding. The DBN is able to learn pronunciation variants as well as typical phoneme confusions of the BLSTM predictor in order to compensate signal disturbances. Unlike conventional Hidden Markov Model (HMM) systems, the proposed architecture is not based on Gaussian mixture modeling. Even without any feature enhancement, our BLSTM-DBN system outperforms a baseline HMM recognizer by up to 18%.

AB - In this paper we introduce a novel hybrid model architecture for speech recognition and investigate its noise robustness on the Aurora 2 database. Our model is composed of a bidirectional Long Short-Term Memory (BLSTM) recurrent neural net exploiting long-range context information for phoneme prediction and a Dynamic Bayesian Network (DBN) for decoding. The DBN is able to learn pronunciation variants as well as typical phoneme confusions of the BLSTM predictor in order to compensate signal disturbances. Unlike conventional Hidden Markov Model (HMM) systems, the proposed architecture is not based on Gaussian mixture modeling. Even without any feature enhancement, our BLSTM-DBN system outperforms a baseline HMM recognizer by up to 18%.

KW - Dynamic Bayesian Networks

KW - Long Short-Term Memory

KW - Noise robustness

KW - Recurrent neural networks

KW - Speech recognition

UR - https://www.scopus.com/pages/publications/79959819892

M3 - Conference contribution

AN - SCOPUS:79959819892

T3 - Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010

SP - 2966

EP - 2969

BT - Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010

PB - International Speech Communication Association

ER -

Wöllmer M, Sun Y, Eyben F, Schuller B. Long Short-Term Memory networks for noise robust speech recognition. In Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010. International Speech Communication Association. 2010. p. 2966-2969. (Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010).

Long Short-Term Memory networks for noise robust speech recognition

Abstract

Publication series

Keywords

Other files and links

Fingerprint

Cite this