International Conference on Computer and Knowledge Engineering

Home / 13th International Conference on Computer and Knowledge Engineering

DPRNN-FORMER: AN EFFICIENT WAY TO DEAL WITH BLIND SOURCE SEPARATION

Authors :

Ramin Ghorbani¹ Sajad Haghzad Klidbary²

1- University of Zanjan 2- University of Zanjan

Keywords :

Blind Source Separation (BSS)،Deep Neural Network (DNN)،Long short term memory(LSTM)،Speech Source Separation،Transformer

Abstract :

Recent advancements in DL have indicated that time-domain methods are more successful than traditional time-frequency-based methods regarding speech separation. However, modeling very long sequences in time-domain separation systems presents some challenges. Recurrent neural networks and 1-D convolutional neural networks are not sufficient for modeling lengthy sequences by themselves. In this paper, a hybrid RNN is proposed, combining a pre-trained DPRNN and transformer. This strategy uses the transformer's ability to perceive context, allowing it to gain insight into the time-evolving data connected to audio signals. To handle extended input sequences, the network partitions them into more manageable sections, performing intra-section and inter-section operations iteratively. The proposed network surpasses current state-of-the-art algorithms, achieving SI-SNR of 11.129 and SDR of 11.285 dB on the public WSj0-3mix dataset.

List of archived papers

A 2D-CNN Architecture for Improving the Classification Accuracy of an Electronic Nose with Different Sensor Positions

Hannaneh Mahdavi - Reza Goldoust - Saeideh Rahbarpour

Multi Model CNN Based Gas Meter Characters Recognition

Sanaz Tarhib - Jafar Tanha - Soodabeh Imanzadeh - Sahar Hassanzadeh Mostafaei

MultiPath ViT OCR: A Lightweight Visual Transformer-based License Plate Optical Character Recognition

Alireza Azadbakht - Saeed Reza Kheradpisheh - Hadi Farahani

An overview of Business Intelligence research in healthcare organizations using a topic modeling approach

Mohammad Mehraeen - Laya Mahmoudi - Mohammad Hossein Sharifi

SGFL: A Federated Learning Approach for Non-IID Data Using Semi-Supervised DCGAN

Alireza Rabiee - Abolfazl Ajdarloo - Mohsen Rahmani

Brain Age Estimation with Twin Vision Transformer using Hippocampus Information Applicable to Alzheimer Dementia Diagnosis

Zahra Qodrati - Seyedeh Masoumeh Taji - Amirhossein Ghaemi - Habibollah Danyali - Kamran Kazemi - Alireza Ghaemi

Machine and Deep Learning Models for Prediction of Small Molecule–Biotech Drug Pair’s Interactions

Fatemeh Nasiri - Mohsen Hooshmand

Bridging the Synthetic-to-Real Gap (BSRG): Creating Simulated Datasets for Domain Adaptation to Enhance Vehicle Detection

Behnaz Sadeghigol - Mohammad Ali Keyvanrad

Improvement of Credit Scoring by LSTM Autoencoder Model

Milad Sattari Maleki - Seyedeh Niusha Motevallian - Faezehsadat Hosseini - Mohammad Sabokrou - Hamidreza Soltanalizadeh Maleki

Automating Theory of Mind Assessment with a LLaMA-3-Powered Chatbot: Enhancing Faux Pas Detection in Autism

Avisa Fallah - Ali Keramati - Mohammad Ali Nazari - Fatemeh Sadat Mirfazeli

more

Samin Hamayesh - Version 41.7.6