End-to-End Optimization of Source Models for Speech and Audio Coding Using a Machine Learning Framework

Tom Bäckström

doi:10.21437/Interspeech.2019-1284

End-to-End Optimization of Source Models for Speech and Audio Coding Using a Machine Learning Framework

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

3 Citations (Scopus)

471 Downloads (Pure)

Abstract

Speech coding is the most commonly used application of speech processing. Accumulated layers of improvements have however made codecs so complex that optimization of individual modules becomes increasingly difficult. This work introduces machine learning methodology to speech and audio coding, such that we can optimize quality in terms of overall entropy. We can then use conventional quantization, coding and perceptual models without modification such that the codec adheres to conventional requirements on algorithmic complexity, latency and robustness to packet loss. Experiments demonstrate that end-to-end optimization of quantization accuracy of the spectral envelope can be used for a lossless reduction in bitrate of 0.4 kbits/s.

Original language	English
Title of host publication	Proceedings of Interspeech
Publisher	International Speech Communication Association (ISCA)
Pages	3401-3405
DOIs	https://doi.org/10.21437/Interspeech.2019-1284
Publication status	Published - Sept 2019
MoE publication type	A4 Conference publication
Event	Interspeech - Graz, Austria Duration: 15 Sept 2019 → 19 Sept 2019 https://www.interspeech2019.org/

Publication series

Name	Interspeech - Annual Conference of the International Speech Communication Association
ISSN (Electronic)	2308-457X

Conference

Conference	Interspeech
Country/Territory	Austria
City	Graz
Period	15/09/2019 → 19/09/2019
Internet address	https://www.interspeech2019.org/

Keywords

speech and audio coding
end-to-end optimization
speech source modeling

Access to Document

10.21437/Interspeech.2019-1284

ELEC_Backstrom_End-to-end_InterspeechFinal published version, 264 KB

https://www.isca-speech.org/archive/Interspeech_2019/pdfs/1284.pdf

Interdisciplinary research on statistical parametric speech synthesis
Alku, P. (Principal investigator), Bäckström, T. (Project Member), Juvela, L. (Project Member), Murtola, T. (Project Member), Nonavinakere Prabhakera, N. (Project Member), Bollepalli, B. (Project Member) & Airaksinen, M. (Project Member)
01/01/2018 → 31/12/2019
Project: Academy of Finland: Other research funding

Cite this

@inproceedings{8cb5f77d88994276a7499988cafc80f4,

title = "End-to-End Optimization of Source Models for Speech and Audio Coding Using a Machine Learning Framework",

abstract = "Speech coding is the most commonly used application of speech processing. Accumulated layers of improvements have however made codecs so complex that optimization of individual modules becomes increasingly difficult. This work introduces machine learning methodology to speech and audio coding, such that we can optimize quality in terms of overall entropy. We can then use conventional quantization, coding and perceptual models without modification such that the codec adheres to conventional requirements on algorithmic complexity, latency and robustness to packet loss. Experiments demonstrate that end-to-end optimization of quantization accuracy of the spectral envelope can be used for a lossless reduction in bitrate of 0.4 kbits/s.",

keywords = "speech and audio coding, end-to-end optimization, speech source modeling",

author = "Tom B{\"a}ckstr{\"o}m",

year = "2019",

month = sep,

doi = "10.21437/Interspeech.2019-1284",

language = "English",

series = "Interspeech - Annual Conference of the International Speech Communication Association",

publisher = "International Speech Communication Association (ISCA)",

pages = "3401--3405",

booktitle = "Proceedings of Interspeech",

address = "France",

note = "Interspeech ; Conference date: 15-09-2019 Through 19-09-2019",

url = "https://www.interspeech2019.org/",

}

Bäckström, T 2019, End-to-End Optimization of Source Models for Speech and Audio Coding Using a Machine Learning Framework. in Proceedings of Interspeech. Interspeech - Annual Conference of the International Speech Communication Association, International Speech Communication Association (ISCA), pp. 3401-3405, Interspeech, Graz, Austria, 15/09/2019. https://doi.org/10.21437/Interspeech.2019-1284

End-to-End Optimization of Source Models for Speech and Audio Coding Using a Machine Learning Framework. / Bäckström, Tom.
Proceedings of Interspeech. International Speech Communication Association (ISCA), 2019. p. 3401-3405 (Interspeech - Annual Conference of the International Speech Communication Association).

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

TY - GEN

T1 - End-to-End Optimization of Source Models for Speech and Audio Coding Using a Machine Learning Framework

AU - Bäckström, Tom

PY - 2019/9

Y1 - 2019/9

N2 - Speech coding is the most commonly used application of speech processing. Accumulated layers of improvements have however made codecs so complex that optimization of individual modules becomes increasingly difficult. This work introduces machine learning methodology to speech and audio coding, such that we can optimize quality in terms of overall entropy. We can then use conventional quantization, coding and perceptual models without modification such that the codec adheres to conventional requirements on algorithmic complexity, latency and robustness to packet loss. Experiments demonstrate that end-to-end optimization of quantization accuracy of the spectral envelope can be used for a lossless reduction in bitrate of 0.4 kbits/s.

AB - Speech coding is the most commonly used application of speech processing. Accumulated layers of improvements have however made codecs so complex that optimization of individual modules becomes increasingly difficult. This work introduces machine learning methodology to speech and audio coding, such that we can optimize quality in terms of overall entropy. We can then use conventional quantization, coding and perceptual models without modification such that the codec adheres to conventional requirements on algorithmic complexity, latency and robustness to packet loss. Experiments demonstrate that end-to-end optimization of quantization accuracy of the spectral envelope can be used for a lossless reduction in bitrate of 0.4 kbits/s.

KW - speech and audio coding

KW - end-to-end optimization

KW - speech source modeling

U2 - 10.21437/Interspeech.2019-1284

DO - 10.21437/Interspeech.2019-1284

M3 - Conference article in proceedings

T3 - Interspeech - Annual Conference of the International Speech Communication Association

SP - 3401

EP - 3405

BT - Proceedings of Interspeech

PB - International Speech Communication Association (ISCA)

T2 - Interspeech

Y2 - 15 September 2019 through 19 September 2019

ER -

End-to-End Optimization of Source Models for Speech and Audio Coding Using a Machine Learning Framework

Abstract

Publication series

Conference

Keywords

Access to Document

Fingerprint

Projects

Interdisciplinary research on statistical parametric speech synthesis

Cite this