SepIt: Approaching a Single Channel Speech Separation Bound

Shahar Lutati; Eliya Nachmani; Lior Wolf

doi:10.21437/Interspeech.2022-149

SepIt: Approaching a Single Channel Speech Separation Bound

Shahar Lutati, Eliya Nachmani, Lior Wolf

Meta AI

Research output: Contribution to journal › Conference article › peer-review

11 Scopus citations

Abstract

We present an upper bound for the Single Channel Speech Separation task, which is based on an assumption regarding the nature of short segments of speech. Using the bound, we are able to show that while the recent methods have made great progress for a few speakers, there is room for improvement for five and ten speakers. We then introduce a Deep neural network, SepIt, that iteratively improves the different speakers' estimation. At test time, SpeIt has a varying number of iterations per test sample, based on a mutual information criterion that arises from our analysis. In an extensive set of experiments, SepIt outperforms the state of the art neural networks for 2, 3, 5, and 10 speakers.

Original language	English
Pages (from-to)	5323-5327
Number of pages	5
Journal	Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH
Volume	2022-September
DOIs	https://doi.org/10.21437/Interspeech.2022-149
State	Published - 2022
Event	23rd Annual Conference of the International Speech Communication Association, INTERSPEECH 2022 - Incheon, Korea, Republic of Duration: 18 Sep 2022 → 22 Sep 2022

Keywords

deep learning
single channel
speech separation

Access to Document

10.21437/Interspeech.2022-149

Cite this

@article{30d968c1feca49749622850d9e508267,

title = "SepIt: Approaching a Single Channel Speech Separation Bound",

abstract = "We present an upper bound for the Single Channel Speech Separation task, which is based on an assumption regarding the nature of short segments of speech. Using the bound, we are able to show that while the recent methods have made great progress for a few speakers, there is room for improvement for five and ten speakers. We then introduce a Deep neural network, SepIt, that iteratively improves the different speakers' estimation. At test time, SpeIt has a varying number of iterations per test sample, based on a mutual information criterion that arises from our analysis. In an extensive set of experiments, SepIt outperforms the state of the art neural networks for 2, 3, 5, and 10 speakers.",

keywords = "deep learning, single channel, speech separation",

author = "Shahar Lutati and Eliya Nachmani and Lior Wolf",

note = "Publisher Copyright: Copyright {\textcopyright} 2022 ISCA.; 23rd Annual Conference of the International Speech Communication Association, INTERSPEECH 2022 ; Conference date: 18-09-2022 Through 22-09-2022",

year = "2022",

doi = "10.21437/Interspeech.2022-149",

language = "אנגלית",

volume = "2022-September",

pages = "5323--5327",

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

issn = "2308-457X",

}

TY - JOUR

T1 - SepIt

T2 - 23rd Annual Conference of the International Speech Communication Association, INTERSPEECH 2022

AU - Lutati, Shahar

AU - Nachmani, Eliya

AU - Wolf, Lior

PY - 2022

Y1 - 2022

N2 - We present an upper bound for the Single Channel Speech Separation task, which is based on an assumption regarding the nature of short segments of speech. Using the bound, we are able to show that while the recent methods have made great progress for a few speakers, there is room for improvement for five and ten speakers. We then introduce a Deep neural network, SepIt, that iteratively improves the different speakers' estimation. At test time, SpeIt has a varying number of iterations per test sample, based on a mutual information criterion that arises from our analysis. In an extensive set of experiments, SepIt outperforms the state of the art neural networks for 2, 3, 5, and 10 speakers.

AB - We present an upper bound for the Single Channel Speech Separation task, which is based on an assumption regarding the nature of short segments of speech. Using the bound, we are able to show that while the recent methods have made great progress for a few speakers, there is room for improvement for five and ten speakers. We then introduce a Deep neural network, SepIt, that iteratively improves the different speakers' estimation. At test time, SpeIt has a varying number of iterations per test sample, based on a mutual information criterion that arises from our analysis. In an extensive set of experiments, SepIt outperforms the state of the art neural networks for 2, 3, 5, and 10 speakers.

KW - deep learning

KW - single channel

KW - speech separation

UR - http://www.scopus.com/inward/record.url?scp=85140048063&partnerID=8YFLogxK

U2 - 10.21437/Interspeech.2022-149

DO - 10.21437/Interspeech.2022-149

M3 - ???researchoutput.researchoutputtypes.contributiontojournal.conferencearticle???

AN - SCOPUS:85140048063

SN - 2308-457X

VL - 2022-September

SP - 5323

EP - 5327

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

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

Y2 - 18 September 2022 through 22 September 2022

ER -

SepIt: Approaching a Single Channel Speech Separation Bound

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this