| Paper ID | AUD-26.2 |
| Paper Title |
A NOVEL NMF-HMM SPEECH ENHANCEMENT ALGORITHM BASED ON POISSON MIXTURE MODEL |
| Authors |
Yang Xiang, Aalborg University & Capturi A/S, Denmark; Liming Shi, Aalborg University, Denmark; Jesper Lisby Højvang, Morten Højfeldt Rasmussen, Capturi A/S, Denmark; Mads Græsbøll Christensen, Aalborg University, Denmark |
| Session | AUD-26: Signal Enhancement and Restoration 3: Signal Enhancement |
| Location | Gather.Town |
| Session Time: | Thursday, 10 June, 16:30 - 17:15 |
| Presentation Time: | Thursday, 10 June, 16:30 - 17:15 |
| Presentation |
Poster
|
| Topic |
Audio and Acoustic Signal Processing: [AUD-SEN] Signal Enhancement and Restoration |
| IEEE Xplore Open Preview |
Click here to view in IEEE Xplore |
| Virtual Presentation |
Click here to watch in the Virtual Conference |
| Abstract |
In this paper, we propose a novel non-negative matrix factorization (NMF) and hidden Markov model (NMF-HMM) based speechenhancement algorithm, which applies the Poisson mixture model(PMM). Compared to our previous proposed NMF-HMM method, the novel PMM-NMF-HMM algorithm uses the Poisson mixture distribution as the state conditional likelihood function for HMM ratherthan the single Poisson distribution. This means that there are the more basis matrices to be used to model the speech and noise signal, so the more signal information can be captured by using the PMM. Our algorithm includes the training and enhancement stage. In thetraining stage, our method can also achieve the computationally efficient multiplicative update (MU) for parameters like our previousNMF-HMM algorithm. In the online speech enhancement stage, a novel PMM-NMF-HMM minimum mean-square error (MMSE) estimator is developed. The experimental results indicate that the proposed PMM-NMF-HMM can acquire higher short-time objec-tive intelligibility (STOI) and perceptual evaluation of speech quality(PESQ) score than NMF-HMM. Additionally, our method can also outperform other state-of-the-art NMF-based supervised speech enhancement algorithms. |
