Application of artificial neural networks to the recognition of musical sounds

The aim of the presented work was to train a neural network in order to recognize a class of a chosen musical instrument. As problems related to analysis of sounds are related to human subjective perception abilities, then it seems that such tools of analyses as neural nets should be used for recogn...

Full description

Saved in:
Bibliographic Details
Main Authors: B. KOSTEK, R. KRÓLIKOWSKI
Format: Article
Language:English
Published: Institute of Fundamental Technological Research Polish Academy of Sciences 2014-01-01
Series:Archives of Acoustics
Online Access:https://acoustics.ippt.pan.pl/index.php/aa/article/view/935
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849318546297847808
author B. KOSTEK
R. KRÓLIKOWSKI
author_facet B. KOSTEK
R. KRÓLIKOWSKI
author_sort B. KOSTEK
collection DOAJ
description The aim of the presented work was to train a neural network in order to recognize a class of a chosen musical instrument. As problems related to analysis of sounds are related to human subjective perception abilities, then it seems that such tools of analyses as neural nets should be used for recognition processes. On the other hand, an artificial neural network should not be trained directly with subsequent samples of a sound, thus the feature extraction procedure is needed at first. As, there is no consensus regarding the selection of parameters for feature vector extraction, thus the experiment aimed to check whether calculated parameters are sufficient for creating a set of sound patterns used for neural network training. Some neural nets were investigated in the experiment, they were trained with so-called ELEVEN and FOURTEEN vector types. After the learning procedure was executed, other examples of the previously created database (but not seen by the neural network) were presented to neural nets. Results show that NNs (neural networks) are able to generalize information included in feature vectors. Therefore, when presenting data to NN inputs, there is no problem with variation of parameters within data, and consequently with data clustering, because a NN has the ability to generalize information during the learning phase. In the paper, an analysis of experimental results will be carried on, and conclusions derived from the performed tests will be presented.
format Article
id doaj-art-4d5ca4458c9044dd810b175a80d671b1
institution Kabale University
issn 0137-5075
2300-262X
language English
publishDate 2014-01-01
publisher Institute of Fundamental Technological Research Polish Academy of Sciences
record_format Article
series Archives of Acoustics
spelling doaj-art-4d5ca4458c9044dd810b175a80d671b12025-08-20T03:50:48ZengInstitute of Fundamental Technological Research Polish Academy of SciencesArchives of Acoustics0137-50752300-262X2014-01-01221Application of artificial neural networks to the recognition of musical soundsB. KOSTEK0R. KRÓLIKOWSKI1Sound Engineering Department, Faculty of Electronics, Telecommunications and Informatics, Gdańsk University of TechnologySound Engineering Department, Faculty of Electronics, Telecommunications and Informatics, Gdańsk University of TechnologyThe aim of the presented work was to train a neural network in order to recognize a class of a chosen musical instrument. As problems related to analysis of sounds are related to human subjective perception abilities, then it seems that such tools of analyses as neural nets should be used for recognition processes. On the other hand, an artificial neural network should not be trained directly with subsequent samples of a sound, thus the feature extraction procedure is needed at first. As, there is no consensus regarding the selection of parameters for feature vector extraction, thus the experiment aimed to check whether calculated parameters are sufficient for creating a set of sound patterns used for neural network training. Some neural nets were investigated in the experiment, they were trained with so-called ELEVEN and FOURTEEN vector types. After the learning procedure was executed, other examples of the previously created database (but not seen by the neural network) were presented to neural nets. Results show that NNs (neural networks) are able to generalize information included in feature vectors. Therefore, when presenting data to NN inputs, there is no problem with variation of parameters within data, and consequently with data clustering, because a NN has the ability to generalize information during the learning phase. In the paper, an analysis of experimental results will be carried on, and conclusions derived from the performed tests will be presented.https://acoustics.ippt.pan.pl/index.php/aa/article/view/935
spellingShingle B. KOSTEK
R. KRÓLIKOWSKI
Application of artificial neural networks to the recognition of musical sounds
Archives of Acoustics
title Application of artificial neural networks to the recognition of musical sounds
title_full Application of artificial neural networks to the recognition of musical sounds
title_fullStr Application of artificial neural networks to the recognition of musical sounds
title_full_unstemmed Application of artificial neural networks to the recognition of musical sounds
title_short Application of artificial neural networks to the recognition of musical sounds
title_sort application of artificial neural networks to the recognition of musical sounds
url https://acoustics.ippt.pan.pl/index.php/aa/article/view/935
work_keys_str_mv AT bkostek applicationofartificialneuralnetworkstotherecognitionofmusicalsounds
AT rkrolikowski applicationofartificialneuralnetworkstotherecognitionofmusicalsounds