Yi Li
Improving speech enhancement performance.
Email: [email protected]
Project title
Cocktail party problem based machine learning methods and neural networks
Supervisors
Project description
My research project will provide a novel algorithm for the MSS.
Methodology and objectives
The research will use neural networks.
We will establish a system to train two signal approximation (SA)-long short-term memory (LSTM) blocks sequentially and dereverberate speech mixture. For the next step, we will develop a subband training LSTM based speech enhancement method will.
Result
The proposed method can further improve speech enhancement performance compared with state-of-the-art methods in terms of:
- STOI (short-time objective intelligibility)
- PESQ (Perceptual Evaluation of Speech Quality), and
- SDR (source to distortion ratio)
Publications
- Li Y, Sun Y, Naqvi SM. Sequentially trained DNNs based monaural source separation in real room environments. In: Sensor Signal Processing for Defence Conference (SSPD). 2019, Brighton, UK: IEEE.
- Li Y, Sun Y, Naqvi SM. Monaural source separation based on sequentially trained LSTMs in real room environments. In: 27th European Signal Processing Conference (EUSIPCO). 2019, A Coruna, Spain.
- Li Y, Sun Y, Naqvi SM. PSD and Signal Approximation-LSTM Based Speech Enhancement. In: 13th International Conference on Signal Processing and Communication Systems (ICSPCS). 2019, Gold Coast, Australia: IEEE. In Press.
Qualifications
- MSc in Communications and Signal Processing from Newcastle University