Neural Networks with Dilated Convolutions for Sound Event Recognition

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216305%3A26220%2F21%3APU140671" target="_blank" >RIV/00216305:26220/21:PU140671 - isvavai.cz</a>

Výsledek na webu

—

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Neural Networks with Dilated Convolutions for Sound Event Recognition

Popis výsledku v původním jazyce

Convolutional neural networks, most commonly deployed in image classification tasks, typically use square-shaped convolutional kernels, which are well suited for feature extraction from two-dimensional data. This study explores the effect of utilizing spectrally aware dilated convolutions specialized for sound event recognition. By extending the base kernels in the time or the frequency dimension, the features extracted from the spectral audio representations should, in theory, better capture the temporal and timbral information of different sound events. The baseline neural network model with squared kernels was compared against three models, which used an increasing dilation factor in the subsequent convolutional layers. The three models were purposefully tuned to focus towards the frequency and time feature extraction. The results have shown that the models with dilated convolutions performed noticeably better in comparison with the baseline model.

Název v anglickém jazyce

Neural Networks with Dilated Convolutions for Sound Event Recognition

Popis výsledku anglicky

Convolutional neural networks, most commonly deployed in image classification tasks, typically use square-shaped convolutional kernels, which are well suited for feature extraction from two-dimensional data. This study explores the effect of utilizing spectrally aware dilated convolutions specialized for sound event recognition. By extending the base kernels in the time or the frequency dimension, the features extracted from the spectral audio representations should, in theory, better capture the temporal and timbral information of different sound events. The baseline neural network model with squared kernels was compared against three models, which used an increasing dilation factor in the subsequent convolutional layers. The three models were purposefully tuned to focus towards the frequency and time feature extraction. The results have shown that the models with dilated convolutions performed noticeably better in comparison with the baseline model.

Druh

O - Ostatní výsledky

CEP obor

—

OECD FORD obor

20201 - Electrical and electronic engineering

Projekt

—

Návaznosti

S - Specificky vyzkum na vysokych skolach

Rok uplatnění

2021

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů