Refine
Document Type
- Conference Proceeding (6)
- Article (2)
- Doctoral Thesis (1)
Language
- English (9) (remove)
Has Fulltext
- no (9)
Keywords
- MEMS microphones (1)
- MVDR beamforming (1)
- Multichannel Wiener filter (1)
- Speech signal processing (1)
- Sprachakustik (1)
- Sprachverarbeitung (1)
- Wind noise reduction (1)
Institute
In this paper we propose a method to determine the active speaker for each time-frequency point in the noisy signals of a microphone array. This detection is based on a statistical model where the speech signals as well as noise signals are assumed to be multivariate Gaussian random variables in the Fourier domain. Based on this model we derive a maximum-likelihood detector for the active speaker. The decision is based on the a posteriori signal to noise ratio (SNR) of a speaker dependent max-SNR beamformer.