Contribution

1. This repository stores Matlab code of our proposed blind source separation (BSS) algorithm, DUET_ISR, a method combining time-frequency masking and linear spatial filtering (or beamforming). DUET_ISR is fast, analytical and performs better than the state-of-the-art BSS algorithms. The proposed algorithm can separate any number of speech sources using two microphones.

2. This repository also stores Matlab codes of all BSS algorithms we used as a comparison in our ICASSP 2022 paper "HARVESTING PARTIALLY-DISJOINT TIME-FREQUENCY INFORMATION FOR IMPROVING DEGENERATE UNMIXING ESTIMATION TECHNIQUE".

These BSS algorithms including

1. Degenerate unmixing estimation technique (DUET) (Rickard, 2007), a time-frequency masking-based method.
2. Independent vector analysis (IVA) (Kim et al., 2006), independent component analysis (ICA) based method.
3. Independent low-rank matrix analysis (ILRMA) (Kitamura et al., 2016), a method combining ICA and non-negative matrix factorazation (NMF).
4. Multi-channel non-nagetive factorization (MULTI_NMF) (Ozerov et al., 2010), NMF-based method.
5. Full-rank spatial model (FULLRANK) (Duong et al., 2010).

Our paper

He, Yudong, He Wang, Qifeng Chen, and Richard HY So. "Harvesting Partially-Disjoint Time-Frequency Information for Improving Degenerate Unmixing Estimation Technique." In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 506-510. IEEE, 2022.

Remark

1. More separation results and separation time can be found on this page: https://ydcnanhe.github.io/demo-icassp2022/
2. Matlab version: MATLAB R2019a on the Linux system.

Contact

Yudong He, email: yhebh@connect.ust.hk