---
title: 'Advantage and Disadvantage of Method'
disqus: hackmd
---

Speech Separation
===

[TOC]

### Method
- Independent Component Analysis
- Advantages: Simple and efficient
- Disadvantages:
- Many assumptions are made, some not exist in real life scenario.
- Slow convergence lead to difficult in realtime application.
- Cannot determine order and variances of independent components.
- Weiner Filtering
- Noise Reduction technique in linear time invariant systems, using Wiener Filter aims at estimating a target signal in a noisy signal.
- Advantages:
- Optimal result and effective method for noise reduction.
- Reduce effect of reverberation
- No pre-processing needed
- Disadvantages:
- Work on assumption of stationarity of source signals and the availability of their second order satistics. (target and noise signal mostly be non stationary)
- perform poorly when there no clear way of distinguishing the signal from noise
- Principle Component Analysis
- Advantages:
- Reduces number of dimension
- Disadvantages:
- Removed correlations
- Does not remove higher order dependence

## Common Method and Dataset
1. Dataset - Wall Street Journal (WSJ0) corpus
https://arxiv.org/abs/1508.04306
2. Dataset - TIMIT
3. Dataset - CHiME
4. Evaluation - Source to Distortion Ratio (SDR)
https://ieeexplore.ieee.org/document/1643671

###### tags: `research` `speech-separation`