Improving Music Genre Classification by Short Time Feature Integration

Anders Meng; Peter Ahrendt; Jan Larsen

Improving Music Genre Classification by Short Time Feature Integration

Anders Meng, Peter Ahrendt, Jan Larsen

Research output: Contribution to conference › Poster › Research

378 Downloads (Pure)

Abstract

Many different short-time features (derived from 10-30ms of audio) have been proposed for music segmentation, retrieval and genre classification. Often the available time frame of the music to make a decision (the decision time horizon) is in the range of seconds instead of milliseconds. The problem of making new features on the larger time scale from the short-time features (feature integration) has only received little attention. This paper investigates different methods for feature integration (early information fusion) and late information fusion (assembling of probabilistic outputs or decisions from the classifier, e.g. majority voting) for music genre classification.

Original language	English
Publication date	2005
Publication status	Published - 2005
Event	IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2005) - Philadelphia, United States Duration: 18 Mar 2005 → 23 Mar 2005

Conference

Conference	IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2005)
Country	United States
City	Philadelphia
Period	18/03/2005 → 23/03/2005

Keywords

Information Fusion
Music genre
Autoregressive Model

Access to Document

Imm3659Final published version, 145 KB

Fingerprint Dive into the research topics of 'Improving Music Genre Classification by Short Time Feature Integration'. Together they form a unique fingerprint.

Cite this

@conference{551fef7845e443bfbae4b2677b1f10fd,

title = "Improving Music Genre Classification by Short Time Feature Integration",

abstract = "Many different short-time features (derived from 10-30ms of audio) have been proposed for music segmentation, retrieval and genre classification. Often the available time frame of the music to make a decision (the decision time horizon) is in the range of seconds instead of milliseconds. The problem of making new features on the larger time scale from the short-time features (feature integration) has only received little attention. This paper investigates different methods for feature integration (early information fusion) and late information fusion (assembling of probabilistic outputs or decisions from the classifier, e.g. majority voting) for music genre classification.",

keywords = "Information Fusion, Music genre, Autoregressive Model",

author = "Anders Meng and Peter Ahrendt and Jan Larsen",

year = "2005",

language = "English",

note = "IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2005), ICASSP '05 ; Conference date: 18-03-2005 Through 23-03-2005",

}

TY - CONF

T1 - Improving Music Genre Classification by Short Time Feature Integration

AU - Meng, Anders

AU - Ahrendt, Peter

AU - Larsen, Jan

PY - 2005

Y1 - 2005

N2 - Many different short-time features (derived from 10-30ms of audio) have been proposed for music segmentation, retrieval and genre classification. Often the available time frame of the music to make a decision (the decision time horizon) is in the range of seconds instead of milliseconds. The problem of making new features on the larger time scale from the short-time features (feature integration) has only received little attention. This paper investigates different methods for feature integration (early information fusion) and late information fusion (assembling of probabilistic outputs or decisions from the classifier, e.g. majority voting) for music genre classification.

AB - Many different short-time features (derived from 10-30ms of audio) have been proposed for music segmentation, retrieval and genre classification. Often the available time frame of the music to make a decision (the decision time horizon) is in the range of seconds instead of milliseconds. The problem of making new features on the larger time scale from the short-time features (feature integration) has only received little attention. This paper investigates different methods for feature integration (early information fusion) and late information fusion (assembling of probabilistic outputs or decisions from the classifier, e.g. majority voting) for music genre classification.

KW - Information Fusion

KW - Music genre

KW - Autoregressive Model

M3 - Poster

T2 - IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2005)

Y2 - 18 March 2005 through 23 March 2005

ER -