On representation learning for artificial bandwidth extension

Matthias Zöhrer; Robert Peharz; Franz Pernkopf

On representation learning for artificial bandwidth extension

Matthias Zöhrer
, Robert Peharz
, Franz Pernkopf

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

Abstract

Recently, sum-product networks (SPNs) showed convincing results on the ill-posed task of artificial bandwidth extension (ABE). However, SPNs are just one type of many architectures which can be summarized as representational models. In this paper, using ABE as benchmark task, we perform a comparative study of Gauss Bernoulli restricted Boltzmann machines, conditional restricted Boltzmann machines, higher order contractive autoencoders, SPNs and generative stochastic networks (GSNs). Especially the latter ones are promising architectures in terms of its reconstruction capabilities. Our experiments show impressive results of GSNs, achieving on average an improvement of 3.90dB and 4.08dB in segmental SNR on a speaker dependent (SD) and speaker independent (SI) scenario compared to SPNs, respectively.

Original language	English
Title of host publication	Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH
Publisher	ISCA
Pages	791-795
Number of pages	5
Publication status	Published - 1 Jan 2015
Externally published	Yes
Event	16th Annual Conference of the International Speech Communication Association, INTERSPEECH 2015 - Dresden, Germany Duration: 6 Sept 2015 → 10 Sept 2015

Conference

Conference	16th Annual Conference of the International Speech Communication Association, INTERSPEECH 2015
Country/Territory	Germany
City	Dresden
Period	6/09/15 → 10/09/15

Keywords

Bandwidth extension
General stochastic network
Representation learning
Sum-product network

Cite this

@inproceedings{7121f4b3c5de47c99c729b7343207959,

title = "On representation learning for artificial bandwidth extension",

abstract = "Recently, sum-product networks (SPNs) showed convincing results on the ill-posed task of artificial bandwidth extension (ABE). However, SPNs are just one type of many architectures which can be summarized as representational models. In this paper, using ABE as benchmark task, we perform a comparative study of Gauss Bernoulli restricted Boltzmann machines, conditional restricted Boltzmann machines, higher order contractive autoencoders, SPNs and generative stochastic networks (GSNs). Especially the latter ones are promising architectures in terms of its reconstruction capabilities. Our experiments show impressive results of GSNs, achieving on average an improvement of 3.90dB and 4.08dB in segmental SNR on a speaker dependent (SD) and speaker independent (SI) scenario compared to SPNs, respectively.",

keywords = "Bandwidth extension, General stochastic network, Representation learning, Sum-product network",

author = "Matthias Z{\"o}hrer and Robert Peharz and Franz Pernkopf",

year = "2015",

month = jan,

day = "1",

language = "English",

pages = "791--795",

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

publisher = "ISCA",

note = "16th Annual Conference of the International Speech Communication Association, INTERSPEECH 2015 ; Conference date: 06-09-2015 Through 10-09-2015",

}

TY - GEN

T1 - On representation learning for artificial bandwidth extension

AU - Zöhrer, Matthias

AU - Peharz, Robert

AU - Pernkopf, Franz

PY - 2015/1/1

Y1 - 2015/1/1

N2 - Recently, sum-product networks (SPNs) showed convincing results on the ill-posed task of artificial bandwidth extension (ABE). However, SPNs are just one type of many architectures which can be summarized as representational models. In this paper, using ABE as benchmark task, we perform a comparative study of Gauss Bernoulli restricted Boltzmann machines, conditional restricted Boltzmann machines, higher order contractive autoencoders, SPNs and generative stochastic networks (GSNs). Especially the latter ones are promising architectures in terms of its reconstruction capabilities. Our experiments show impressive results of GSNs, achieving on average an improvement of 3.90dB and 4.08dB in segmental SNR on a speaker dependent (SD) and speaker independent (SI) scenario compared to SPNs, respectively.

AB - Recently, sum-product networks (SPNs) showed convincing results on the ill-posed task of artificial bandwidth extension (ABE). However, SPNs are just one type of many architectures which can be summarized as representational models. In this paper, using ABE as benchmark task, we perform a comparative study of Gauss Bernoulli restricted Boltzmann machines, conditional restricted Boltzmann machines, higher order contractive autoencoders, SPNs and generative stochastic networks (GSNs). Especially the latter ones are promising architectures in terms of its reconstruction capabilities. Our experiments show impressive results of GSNs, achieving on average an improvement of 3.90dB and 4.08dB in segmental SNR on a speaker dependent (SD) and speaker independent (SI) scenario compared to SPNs, respectively.

KW - Bandwidth extension

KW - General stochastic network

KW - Representation learning

KW - Sum-product network

UR - https://www.scopus.com/pages/publications/84959165751

M3 - Conference contribution

AN - SCOPUS:84959165751

SP - 791

EP - 795

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

PB - ISCA

T2 - 16th Annual Conference of the International Speech Communication Association, INTERSPEECH 2015

Y2 - 6 September 2015 through 10 September 2015

ER -

On representation learning for artificial bandwidth extension

Abstract

Conference

Keywords

Other files and links

Fingerprint

Cite this