Speech separation and localisation using particle filtering

Summary

The objective of this project is to solve the “Cocktail Party Problem”.

Supervisor(s)

Associate Professor Craig Jin

Research Location

Electrical and Information Engineering

Program Type

PHD

Synopsis

Humans have a remarkable ability to localise and separate a single speech source in a noisy and cluttered multi-talker environment. This is particularly remarkable when one considers that we only use two ears (sensors) to do so. Steps to solving the "Cocktail Party Problem" include:

1. to recover and estimate the original sound source from the observed mixture signals recorded at microphones and
2. to localise and track a moving sound source accurately from the observed mixture of signals

The Sequential Monte Carlo (SMC) method, also known as particle filtering will be investigated as a technique to achieve these objectives.

Keywords

Audio Engineering, Signal Processing, Source Separation, Sound localisation, Particle Filtering

Opportunity ID

The opportunity ID for this research opportunity is: 274

Other opportunities with Associate Professor Craig Jin

• Pattern analysis techniques for sound synthesis
• Interpolation of binaural impulse responses for virtual auditory displays
• Sound field recording and recreation
• Beamforming with acoustic vector sensors - Multiple acoustic source localisation using acoustic vector sensor arrays
• Mapping 2D Images to 3D Shape
• New technique for studying human brain activity
• Next Generation Audio Coding
• Spherical multi-modal scene analysis
• Statistical models of ear shape and ear acoustics
• Binaural signal processing algorithms for hearing aids
• Electrical Impedance Tomography for stroke, biophysical monitoring and medical device design
• Impedance tomography for cardiac imaging: high speed tomography
• Medical diagnostics for neonates in the developing world
• Novel Electrodes for rapid electrophysiological recording
• FPGA-based low latency machine learning