funcwj / Cgmm Mvdr
Implementation of the CGMM-MVDR beamforming
Stars: ✭ 86
Programming Languages
python
139335 projects - #7 most used programming language
Notes
The python version of scripts now perfermance badly(may have some logical errors), please refer to apply_cgmm_beamforming.m, which works ok.
Update: Newest version in cgmm-mask-estimator
Usage
install voicebox for MATLAB first.
copy run_cgmm_beamforming.sh and apply_cgmm_beamforming.m to local/, then run followings instead of baseline local/run_beamform_6ch_track.sh commands
local/run_cgmm_beamforming.sh --nj 15 $chime4_data/data/audio/16kHz/isolated_6ch_track/ $enhancement_data
Results
- 6ch
| Methods | dev-simu | dev-real | eval-simu | eval-real |
|---|---|---|---|---|
| Beamformit(SAT) | 14.36% | 12.99% | 21.24% | 21.55% |
| CGMM(SAT) | 11.38% | 11.30% | 15.34% | 17.27% |
| Beamformit(DNN) | 10.29% | 9.59% | 15.79% | 16.73% |
| CGMM(DNN) | 7.69% | 8.40% | 10.82% | 13.51% |
| Beamformit(sMBR) | 9.11% | 8.46% | 14.54% | 15.07% |
| CGMM(sMBR) | 6.88% | 7.58% | 10.15% | 12.12% |
Reference
T. Higuchi, N. Ito, T. Yoshioka, and T. Nakatani, "Robust mvdr beamforming using time-frequency masks for online/offline asr in noise," in ICASSP, 2016.
Note that the project description data, including the texts, logos, images, and/or trademarks,
for each open source project belongs to its rightful owner.
If you wish to add or remove any projects, please contact us at [email protected].