Research Article
3D - Learning Representations From Audio Using Autoencoders
@INPROCEEDINGS{10.4108/eai.7-12-2021.2314968, author={Bharathi A and Prakash J}, title={3D - Learning Representations From Audio Using Autoencoders}, proceedings={Proceedings of the First International Conference on Combinatorial and Optimization, ICCAP 2021, December 7-8 2021, Chennai, India}, publisher={EAI}, proceedings_a={ICCAP}, year={2021}, month={12}, keywords={: audio signal auto correlogram rnn encoder rnn decoder}, doi={10.4108/eai.7-12-2021.2314968} }- Bharathi A
Prakash J
Year: 2021
3D - Learning Representations From Audio Using Autoencoders
ICCAP
EAI
DOI: 10.4108/eai.7-12-2021.2314968
Abstract
Deep learning methods permit us to tackle signal processing challenges from a dissimilar perspective, which is currently overlooked in the composition of music in cinema industry. Audio is inherently added time-sensitive than movie. Audios are encoded using other past methods, resulting in data loss or temporal anomalies. This problem is alleviated by using an auto correlogram with a 3-dimensional view, including time, power, and frequency, to improve accuracy. First, acoustic data should be competently encoded into a compressed format using RNN autoencoder by interrelating with the information. As a result of the compressed format, audio waves should be accurately represented. After that, audio waves are rebuilt into an audio structure with little data loss. The accuracy is improved by 10% by using the RNN encoder and decoder.