A Deep Neural Network for Acoustic-Articulatory Speech Inversion

Benigno Uria; S. Renals; K. Richmond

A Deep Neural Network for Acoustic-Articulatory Speech Inversion

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

Abstract

In this work, we implement a deep belief network for the acoustic-articulatory inversion mapping problem. We find that adding up to 3 hidden-layers improves inversion accuracy. We also show that this improvement is due to the higher ex- pressive capability of a deep model and not a consequence of adding more adjustable parameters. Additionally, we show unsupervised pretraining of the sys- tem improves its performance in all cases, even for a 1 hidden-layer model. Our implementation obtained an average root mean square error of 0.95 mm on the MNGU0 test dataset, beating all previously published results.

Original language	English
Title of host publication	Proc. NIPS 2011 Workshop on Deep Learning and Unsupervised Feature Learning
Publication status	Published - Dec 2011

Access to Document

A Deep Neural Network for Acoustic-ArticulatoryAccepted author manuscript, 529 KB

Fingerprint Dive into the research topics of 'A Deep Neural Network for Acoustic-Articulatory Speech Inversion'. Together they form a unique fingerprint.

2 Finished

Deep Learning for Speech Recognition
Murray, I. & Renals, S.
UK industry, commerce and public corporations
1/10/11 → 31/03/15
Project: Research
ULTRAX: Ultrax: Real-time tongue tracking for speech therapy using ultrasound
Richmond, K., Renals, S., Cleland, J. & Scobbie, J. M.
EPSRC
1/02/11 → 31/07/14
Project: Research

Cite this

@inproceedings{52225b931e2a49d4a5a5139ecf6a2ecb,

title = "A Deep Neural Network for Acoustic-Articulatory Speech Inversion",

abstract = "In this work, we implement a deep belief network for the acoustic-articulatory inversion mapping problem. We find that adding up to 3 hidden-layers improves inversion accuracy. We also show that this improvement is due to the higher ex- pressive capability of a deep model and not a consequence of adding more adjustable parameters. Additionally, we show unsupervised pretraining of the sys- tem improves its performance in all cases, even for a 1 hidden-layer model. Our implementation obtained an average root mean square error of 0.95 mm on the MNGU0 test dataset, beating all previously published results.",

author = "Benigno Uria and S. Renals and K. Richmond",

year = "2011",

month = dec,

language = "English",

booktitle = "Proc. NIPS 2011 Workshop on Deep Learning and Unsupervised Feature Learning",

}

TY - GEN

T1 - A Deep Neural Network for Acoustic-Articulatory Speech Inversion

AU - Uria, Benigno

AU - Renals, S.

AU - Richmond, K.

PY - 2011/12

Y1 - 2011/12

N2 - In this work, we implement a deep belief network for the acoustic-articulatory inversion mapping problem. We find that adding up to 3 hidden-layers improves inversion accuracy. We also show that this improvement is due to the higher ex- pressive capability of a deep model and not a consequence of adding more adjustable parameters. Additionally, we show unsupervised pretraining of the sys- tem improves its performance in all cases, even for a 1 hidden-layer model. Our implementation obtained an average root mean square error of 0.95 mm on the MNGU0 test dataset, beating all previously published results.

AB - In this work, we implement a deep belief network for the acoustic-articulatory inversion mapping problem. We find that adding up to 3 hidden-layers improves inversion accuracy. We also show that this improvement is due to the higher ex- pressive capability of a deep model and not a consequence of adding more adjustable parameters. Additionally, we show unsupervised pretraining of the sys- tem improves its performance in all cases, even for a 1 hidden-layer model. Our implementation obtained an average root mean square error of 0.95 mm on the MNGU0 test dataset, beating all previously published results.

M3 - Conference contribution

BT - Proc. NIPS 2011 Workshop on Deep Learning and Unsupervised Feature Learning

ER -

A Deep Neural Network for Acoustic-Articulatory Speech Inversion

Abstract

Access to Document

Deep Learning for Speech Recognition

ULTRAX: Ultrax: Real-time tongue tracking for speech therapy using ultrasound

Cite this