Many-speakers single channel speech separation with optimal permutation training

Shaked Dovrat; Eliya Nachmani; Lior Wolf

doi:https://doi.org/10.21437/Interspeech.2021-493

Many-speakers single channel speech separation with optimal permutation training

Shaked Dovrat, Eliya Nachmani, Lior Wolf

Tel Aviv University

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

Abstract

Single channel speech separation has experienced great progress in the last few years. However, training neural speech separation for a large number of speakers (e.g., more than 10 speakers) is out of reach for the current methods, which rely on the Permutation Invariant Training (PIT). In this work, we present a permutation invariant training that employs the Hungarian algorithm in order to train with an O(C³) time complexity, where C is the number of speakers, in comparison to O(C!) of PIT based methods. Furthermore, we present a modified architecture that can handle the increased number of speakers. Our approach separates up to 20 speakers and improves the previous results for large C by a wide margin.

Original language	American English
Title of host publication	22nd Annual Conference of the International Speech Communication Association, INTERSPEECH 2021
Pages	2408-2412
Number of pages	5
ISBN (Electronic)	9781713836902
DOIs	https://doi.org/10.21437/Interspeech.2021-493
State	Published - 1 Jan 2021
Event	22nd Annual Conference of the International Speech Communication Association, INTERSPEECH 2021 - Brno, Czech Republic Duration: 30 Aug 2021 → 3 Sep 2021

Publication series

Name	Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH
Volume	4

Conference

Conference	22nd Annual Conference of the International Speech Communication Association, INTERSPEECH 2021
Country/Territory	Czech Republic
City	Brno
Period	30/08/21 → 3/09/21

Keywords

Deep learning
Single channel
Speech separation

All Science Journal Classification (ASJC) codes

Language and Linguistics
Human-Computer Interaction
Signal Processing
Software
Modelling and Simulation

Access to Document

https://doi.org/10.21437/Interspeech.2021-493

Cite this

Dovrat, S., Nachmani, E., & Wolf, L. (2021). Many-speakers single channel speech separation with optimal permutation training. In 22nd Annual Conference of the International Speech Communication Association, INTERSPEECH 2021 (pp. 2408-2412). (Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH; Vol. 4). https://doi.org/10.21437/Interspeech.2021-493

@inproceedings{d4bbb574b21e45909e51b6116dcdc227,

title = "Many-speakers single channel speech separation with optimal permutation training",

abstract = "Single channel speech separation has experienced great progress in the last few years. However, training neural speech separation for a large number of speakers (e.g., more than 10 speakers) is out of reach for the current methods, which rely on the Permutation Invariant Training (PIT). In this work, we present a permutation invariant training that employs the Hungarian algorithm in order to train with an O(C3) time complexity, where C is the number of speakers, in comparison to O(C!) of PIT based methods. Furthermore, we present a modified architecture that can handle the increased number of speakers. Our approach separates up to 20 speakers and improves the previous results for large C by a wide margin.",

keywords = "Deep learning, Single channel, Speech separation",

author = "Shaked Dovrat and Eliya Nachmani and Lior Wolf",

note = "Publisher Copyright: Copyright {\textcopyright} 2021 ISCA.; 22nd Annual Conference of the International Speech Communication Association, INTERSPEECH 2021 ; Conference date: 30-08-2021 Through 03-09-2021",

year = "2021",

month = jan,

day = "1",

doi = "https://doi.org/10.21437/Interspeech.2021-493",

language = "American English",

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

pages = "2408--2412",

booktitle = "22nd Annual Conference of the International Speech Communication Association, INTERSPEECH 2021",

}

Dovrat, S, Nachmani, E & Wolf, L 2021, Many-speakers single channel speech separation with optimal permutation training. in 22nd Annual Conference of the International Speech Communication Association, INTERSPEECH 2021. Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH, vol. 4, pp. 2408-2412, 22nd Annual Conference of the International Speech Communication Association, INTERSPEECH 2021, Brno, Czech Republic, 30/08/21. https://doi.org/10.21437/Interspeech.2021-493

Many-speakers single channel speech separation with optimal permutation training. / Dovrat, Shaked; Nachmani, Eliya ; Wolf, Lior.
22nd Annual Conference of the International Speech Communication Association, INTERSPEECH 2021. 2021. p. 2408-2412 (Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH; Vol. 4).

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

TY - GEN

T1 - Many-speakers single channel speech separation with optimal permutation training

AU - Dovrat, Shaked

AU - Nachmani, Eliya

AU - Wolf, Lior

PY - 2021/1/1

Y1 - 2021/1/1

N2 - Single channel speech separation has experienced great progress in the last few years. However, training neural speech separation for a large number of speakers (e.g., more than 10 speakers) is out of reach for the current methods, which rely on the Permutation Invariant Training (PIT). In this work, we present a permutation invariant training that employs the Hungarian algorithm in order to train with an O(C3) time complexity, where C is the number of speakers, in comparison to O(C!) of PIT based methods. Furthermore, we present a modified architecture that can handle the increased number of speakers. Our approach separates up to 20 speakers and improves the previous results for large C by a wide margin.

AB - Single channel speech separation has experienced great progress in the last few years. However, training neural speech separation for a large number of speakers (e.g., more than 10 speakers) is out of reach for the current methods, which rely on the Permutation Invariant Training (PIT). In this work, we present a permutation invariant training that employs the Hungarian algorithm in order to train with an O(C3) time complexity, where C is the number of speakers, in comparison to O(C!) of PIT based methods. Furthermore, we present a modified architecture that can handle the increased number of speakers. Our approach separates up to 20 speakers and improves the previous results for large C by a wide margin.

KW - Deep learning

KW - Single channel

KW - Speech separation

UR - http://www.scopus.com/inward/record.url?scp=85119211107&partnerID=8YFLogxK

U2 - https://doi.org/10.21437/Interspeech.2021-493

DO - https://doi.org/10.21437/Interspeech.2021-493

M3 - Conference contribution

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

SP - 2408

EP - 2412

BT - 22nd Annual Conference of the International Speech Communication Association, INTERSPEECH 2021

T2 - 22nd Annual Conference of the International Speech Communication Association, INTERSPEECH 2021

Y2 - 30 August 2021 through 3 September 2021

ER -

Dovrat S, Nachmani E , Wolf L. Many-speakers single channel speech separation with optimal permutation training. In 22nd Annual Conference of the International Speech Communication Association, INTERSPEECH 2021. 2021. p. 2408-2412. (Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH). doi: https://doi.org/10.21437/Interspeech.2021-493

Many-speakers single channel speech separation with optimal permutation training

Abstract

Publication series

Conference

Keywords

All Science Journal Classification (ASJC) codes

Access to Document

Other files and links

Cite this