BABY AI HUB

Home HomeThammasat University Musical AI…

Scroll Down To Discover

Model TU Musical AI

Thammasat University Musical AI Project : Visual-based Musical Data Representation for Composer Classification

August 31, 2023

Visual-based Musical Data
Representation for Composer Classification

Thammasat University Musical AI Project

Musical AI

This research was created as part of the project "Innovative All-around Learning toward Smart Campus Creation" by Dr. Somrudee Deepaisarn and team under the larger-scheme of AI Ready City Networking in RUN Project conceptualized by Prof. Dr. Virach Sornlertlamvanich.

We aim to use machine learning to classify different music composers based on their compositions. We'll gather a dataset of musical pieces from different composers, extract features from each piece, and train a machine learning model to predict the composer given a musical piece.

• The acoustic features, including pitch, velocity, onset, and duration are extracted as the primary input features for the classification of composers.

• A music note is visually represented on an image at a specified feature channel where a pixel intensity indicates its value for velocity, onset, or duration.

EXPLANATION OF ACOUSTIC FEATURE

• The grayscale single channel images were constructed from a single feature of velocity, onset, and duration.

• These three features were also combined to create multi-channel RGB images.

• In this work, composer classification was performed on the visual-based musical data representation.

• Input data and classification models constituted of adjustable parameters, including the based pre-trained model, length of the temporal index, and acoustic features.

• The F1-score of classification performance on each parameter was investigated as follows:

CLASSIFICATION PERFORMANCE OF MODELS CONSTRUCTED BY VARIOUS FEATURES

BABY AI HUB

Thammasat University Musical AI Project : Visual-based Musical Data Representation for Composer Classification

Visual-based Musical Data
Representation for Composer Classification

Thammasat University Musical AI Project

DISCRIPTION

Musical AI

This research was created as part of the project "Innovative All-around Learning toward Smart Campus Creation" by Dr. Somrudee Deepaisarn and team under the larger-scheme of AI Ready City Networking in RUN Project conceptualized by Prof. Dr. Virach Sornlertlamvanich.

We aim to use machine learning to classify different music composers based on their compositions. We'll gather a dataset of musical pieces from different composers, extract features from each piece, and train a machine learning model to predict the composer given a musical piece.

• The acoustic features, including pitch, velocity, onset, and duration are extracted as the primary input features for the classification of composers.

• A music note is visually represented on an image at a specified feature channel where a pixel intensity indicates its value for velocity, onset, or duration.

EXPLANATION OF ACOUSTIC FEATURE

• The grayscale single channel images were constructed from a single feature of velocity, onset, and duration.

• These three features were also combined to create multi-channel RGB images.

• In this work, composer classification was performed on the visual-based musical data representation.

• Input data and classification models constituted of adjustable parameters, including the based pre-trained model, length of the temporal index, and acoustic features.

• The F1-score of classification performance on each parameter was investigated as follows:

CLASSIFICATION PERFORMANCE OF MODELS CONSTRUCTED BY VARIOUS FEATURES

Thammasat University Musical AI Project : Visual-based Musical Data Representation for Composer Classification

Visual-based Musical Data Representation for Composer Classification

Thammasat University Musical AI Project

DISCRIPTION

Musical AI

This research was created as part of the project "Innovative All-around Learning toward Smart Campus Creation" by Dr. Somrudee Deepaisarn and team under the larger-scheme of AI Ready City Networking in RUN Project conceptualized by Prof. Dr. Virach Sornlertlamvanich.

We aim to use machine learning to classify different music composers based on their compositions. We'll gather a dataset of musical pieces from different composers, extract features from each piece, and train a machine learning model to predict the composer given a musical piece.

• The acoustic features, including pitch, velocity, onset, and duration are extracted as the primary input features for the classification of composers.

• A music note is visually represented on an image at a specified feature channel where a pixel intensity indicates its value for velocity, onset, or duration.

EXPLANATION OF ACOUSTIC FEATURE

• The grayscale single channel images were constructed from a single feature of velocity, onset, and duration.

• These three features were also combined to create multi-channel RGB images.

• In this work, composer classification was performed on the visual-based musical data representation.

• Input data and classification models constituted of adjustable parameters, including the based pre-trained model, length of the temporal index, and acoustic features.

• The F1-score of classification performance on each parameter was investigated as follows:

CLASSIFICATION PERFORMANCE OF MODELS CONSTRUCTED BY VARIOUS FEATURES

Visual-based Musical Data
Representation for Composer Classification