Multi-level adaptive networks in tandem and hybrid ASR systems

P. Bell; P. Swietojanski; S. Renals

doi:10.1109/ICASSP.2013.6639014

Multi-level adaptive networks in tandem and hybrid ASR systems

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

Abstract

In this paper we investigate the use of Multi-level adaptive networks (MLAN) to incorporate out-of-domain data when training large vocabulary speech recognition systems. In a set of experiments on multi-genre broadcast data and on TED lecture recordings we present results using of out-of-domain features in a hybrid DNN system and explore tandem systems using a variety of input acoustic features. Our experiments indicate using the MLAN approach in both hybrid and tandem systems results in consistent reductions in word error rate of 5-10% relative.

Original language	English
Title of host publication	Acoustics, Speech and Signal Processing (ICASSP), 2013 IEEE International Conference on
Publisher	Institute of Electrical and Electronics Engineers
Pages	6975-6979
Number of pages	5
ISBN (Print)	978-1-4799-0356-6
DOIs	https://doi.org/10.1109/ICASSP.2013.6639014
Publication status	Published - 2013

Keywords / Materials (for Non-textual outputs)

error analysis
speech recognition
vocabulary
TED lecture recordings
hybrid ASR systems
multigenre broadcast data
multilevel adaptive networks
out-of-domain data
vocabulary speech recognition systems
word error rate
Acoustics
Adaptation models
Hidden Markov models
Neural networks
Speech
Speech recognition
Training
BBC
MLAN
TED
deep neural networks
hybrid
tandem

Access to Document

10.1109/ICASSP.2013.6639014

Bell Swietojanski Renals 2013 Multi level adaptive networks in tandem and hybrid asr systemsAccepted author manuscript, 143 KB

http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=6639014

Natural Speech Technology
Renals, S. (Principal Investigator) & King, S. (Co-investigator)
EPSRC
1/05/11 → 31/07/16
Project: Research

Cite this

@inproceedings{6dbeeb8da45840febba71a6e80cbc569,

title = "Multi-level adaptive networks in tandem and hybrid ASR systems",

abstract = "In this paper we investigate the use of Multi-level adaptive networks (MLAN) to incorporate out-of-domain data when training large vocabulary speech recognition systems. In a set of experiments on multi-genre broadcast data and on TED lecture recordings we present results using of out-of-domain features in a hybrid DNN system and explore tandem systems using a variety of input acoustic features. Our experiments indicate using the MLAN approach in both hybrid and tandem systems results in consistent reductions in word error rate of 5-10% relative.",

keywords = "error analysis, speech recognition, vocabulary, TED lecture recordings, hybrid ASR systems, multigenre broadcast data, multilevel adaptive networks, out-of-domain data, vocabulary speech recognition systems, word error rate, Acoustics, Adaptation models, Hidden Markov models, Neural networks, Speech, Speech recognition, Training, BBC, MLAN, TED, deep neural networks, hybrid, tandem",

author = "P. Bell and P. Swietojanski and S. Renals",

year = "2013",

doi = "10.1109/ICASSP.2013.6639014",

language = "English",

isbn = "978-1-4799-0356-6",

pages = "6975--6979",

booktitle = "Acoustics, Speech and Signal Processing (ICASSP), 2013 IEEE International Conference on",

publisher = "Institute of Electrical and Electronics Engineers",

address = "United States",

}

TY - GEN

T1 - Multi-level adaptive networks in tandem and hybrid ASR systems

AU - Bell, P.

AU - Swietojanski, P.

AU - Renals, S.

PY - 2013

Y1 - 2013

N2 - In this paper we investigate the use of Multi-level adaptive networks (MLAN) to incorporate out-of-domain data when training large vocabulary speech recognition systems. In a set of experiments on multi-genre broadcast data and on TED lecture recordings we present results using of out-of-domain features in a hybrid DNN system and explore tandem systems using a variety of input acoustic features. Our experiments indicate using the MLAN approach in both hybrid and tandem systems results in consistent reductions in word error rate of 5-10% relative.

AB - In this paper we investigate the use of Multi-level adaptive networks (MLAN) to incorporate out-of-domain data when training large vocabulary speech recognition systems. In a set of experiments on multi-genre broadcast data and on TED lecture recordings we present results using of out-of-domain features in a hybrid DNN system and explore tandem systems using a variety of input acoustic features. Our experiments indicate using the MLAN approach in both hybrid and tandem systems results in consistent reductions in word error rate of 5-10% relative.

KW - error analysis

KW - speech recognition

KW - vocabulary

KW - TED lecture recordings

KW - hybrid ASR systems

KW - multigenre broadcast data

KW - multilevel adaptive networks

KW - out-of-domain data

KW - vocabulary speech recognition systems

KW - word error rate

KW - Acoustics

KW - Adaptation models

KW - Hidden Markov models

KW - Neural networks

KW - Speech

KW - Speech recognition

KW - Training

KW - BBC

KW - MLAN

KW - TED

KW - deep neural networks

KW - hybrid

KW - tandem

U2 - 10.1109/ICASSP.2013.6639014

DO - 10.1109/ICASSP.2013.6639014

M3 - Conference contribution

SN - 978-1-4799-0356-6

SP - 6975

EP - 6979

BT - Acoustics, Speech and Signal Processing (ICASSP), 2013 IEEE International Conference on

PB - Institute of Electrical and Electronics Engineers

ER -

Multi-level adaptive networks in tandem and hybrid ASR systems

Abstract

Keywords / Materials (for Non-textual outputs)

Access to Document

Fingerprint

Projects

Natural Speech Technology

Cite this