Speech - Nonspeech Discrimination using the Information Bottleneck Method and Spectro-Temporal Modulation Index

Maria Markaki; Michael Wohlmayr; Yannis Stylianou

Speech - Nonspeech Discrimination using the Information Bottleneck Method and Spectro-Temporal Modulation Index

Maria Markaki
, Michael Wohlmayr
, Yannis Stylianou

Institute of Signal Processing and Speech Communication (4420)

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

Abstract

In this work, we adopt an information theoretic approach - the Information Bottleneck method - to extract the relevant spectrotemporal modulations for the task of speech / non-speech dis-
crimination - non-speech events include music, noise and animal vocalizations. A compact representation (a “cluster prototype”) is built for each class consisting of the maximally in-
formative features with respect to the classification task. We assess the similarity of a sound to each representative cluster using the spectro-temporal modulation index (STMI) adapted
to handle the contribution of different frequency bands. A simple threshold check is then used for discriminating speech from non-speech events. Conducted experiments have shown that the
proposed method has low complexity and high accuracy of discrimination in low SNR conditions compared to recently proposed methods for the same task.

Original language	English
Title of host publication	Proceedings of the 8th Annual Conference of the International Speech Communication Association
Place of Publication	Red Hook, NY
Publisher	Curran Associates, Inc
Pages	2913-2916
ISBN (Print)	978-1-60560-316-2
Publication status	Published - 2007
Event	8th International Conference on Spoken Language Processing: Interspeech 2007 - Antwerpen, Belgium Duration: 27 Aug 2007 → 31 Aug 2007

Conference

Conference	8th International Conference on Spoken Language Processing
Abbreviated title	ICSLP 2007
Country/Territory	Belgium
City	Antwerpen
Period	27/08/07 → 31/08/07

Access to Document

IS070861Final published version, 211 KB

Cite this

Speech - Nonspeech Discrimination using the Information Bottleneck Method and Spectro-Temporal Modulation Index. / Markaki, Maria; Wohlmayr, Michael; Stylianou, Yannis.
Proceedings of the 8th Annual Conference of the International Speech Communication Association. Red Hook, NY: Curran Associates, Inc, 2007. p. 2913-2916.

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

Markaki, M, Wohlmayr, M & Stylianou, Y 2007, Speech - Nonspeech Discrimination using the Information Bottleneck Method and Spectro-Temporal Modulation Index. in Proceedings of the 8th Annual Conference of the International Speech Communication Association. Curran Associates, Inc, Red Hook, NY, pp. 2913-2916, 8th International Conference on Spoken Language Processing, Antwerpen, Belgium, 27/08/07.

@inproceedings{b85520fbbe8240bdb8a7ea206320b8f0,

title = "Speech - Nonspeech Discrimination using the Information Bottleneck Method and Spectro-Temporal Modulation Index",

abstract = "In this work, we adopt an information theoretic approach - the Information Bottleneck method - to extract the relevant spectrotemporal modulations for the task of speech / non-speech dis- crimination - non-speech events include music, noise and animal vocalizations. A compact representation (a {\textquotedblleft}cluster prototype{\textquotedblright}) is built for each class consisting of the maximally in- formative features with respect to the classification task. We assess the similarity of a sound to each representative cluster using the spectro-temporal modulation index (STMI) adapted to handle the contribution of different frequency bands. A simple threshold check is then used for discriminating speech from non-speech events. Conducted experiments have shown that the proposed method has low complexity and high accuracy of discrimination in low SNR conditions compared to recently proposed methods for the same task.",

author = "Maria Markaki and Michael Wohlmayr and Yannis Stylianou",

year = "2007",

language = "English",

isbn = "978-1-60560-316-2 ",

pages = "2913--2916",

booktitle = "Proceedings of the 8th Annual Conference of the International Speech Communication Association",

publisher = "Curran Associates, Inc",

address = "United States",

note = "8th International Conference on Spoken Language Processing : Interspeech 2007, ICSLP 2007 ; Conference date: 27-08-2007 Through 31-08-2007",

}

TY - GEN

T1 - Speech - Nonspeech Discrimination using the Information Bottleneck Method and Spectro-Temporal Modulation Index

AU - Markaki, Maria

AU - Wohlmayr, Michael

AU - Stylianou, Yannis

PY - 2007

Y1 - 2007

N2 - In this work, we adopt an information theoretic approach - the Information Bottleneck method - to extract the relevant spectrotemporal modulations for the task of speech / non-speech dis- crimination - non-speech events include music, noise and animal vocalizations. A compact representation (a “cluster prototype”) is built for each class consisting of the maximally in- formative features with respect to the classification task. We assess the similarity of a sound to each representative cluster using the spectro-temporal modulation index (STMI) adapted to handle the contribution of different frequency bands. A simple threshold check is then used for discriminating speech from non-speech events. Conducted experiments have shown that the proposed method has low complexity and high accuracy of discrimination in low SNR conditions compared to recently proposed methods for the same task.

AB - In this work, we adopt an information theoretic approach - the Information Bottleneck method - to extract the relevant spectrotemporal modulations for the task of speech / non-speech dis- crimination - non-speech events include music, noise and animal vocalizations. A compact representation (a “cluster prototype”) is built for each class consisting of the maximally in- formative features with respect to the classification task. We assess the similarity of a sound to each representative cluster using the spectro-temporal modulation index (STMI) adapted to handle the contribution of different frequency bands. A simple threshold check is then used for discriminating speech from non-speech events. Conducted experiments have shown that the proposed method has low complexity and high accuracy of discrimination in low SNR conditions compared to recently proposed methods for the same task.

M3 - Conference paper

SN - 978-1-60560-316-2

SP - 2913

EP - 2916

BT - Proceedings of the 8th Annual Conference of the International Speech Communication Association

PB - Curran Associates, Inc

CY - Red Hook, NY

T2 - 8th International Conference on Spoken Language Processing

Y2 - 27 August 2007 through 31 August 2007

ER -

Speech - Nonspeech Discrimination using the Information Bottleneck Method and Spectro-Temporal Modulation Index

Abstract

Conference

Access to Document

Fingerprint

Cite this