storage_dir: Path where to store the enhanced data ( /audio//*.wav).Python -m pb_run_rttm with \Ĭhime6_dir= '/net/fastdb/chime6/CHiME6 ' \ If you are using this code please cite the following paper ( pdf, poster): The core code is located in the file pb_chime5/core.py.Īn example script to run the enhancement is in pb_chime5/scripts/run.py and can be executed with python -m pb_ with session_id=dev wpe=True wpe_tabs=2. The front-end consists out of WPE, a spacial mixture model that uses time annotations (GSS), beamforming and masking: The best single system WERs with this enhancement are 41.6 % on the development and 43.2 % on the evaluation set reported in. In combination with an acoustic model presented by the RWTH Aachen this multi-array front-end achieved the third best results during the challenge with 54.56 % on the development and 55.30 % on the evaluation set.Ī later cooperation with Hitachi led to WER of 39.94 % on the development and 41.64 % on the evaluation set, using the multi-array front-end presented in this repository. Using the baseline backend provided by the challenge organizers on the data enhanced with this multi-array front-end using the default parameters which differ slightly from the original paper a WER of 60.89 % was achieved on the development set. This repository includes all components of the CHiME-5 front-end presented by Paderborn University on the CHiME-5 workshop. Pb_chime5: Front-End Processing for the CHiME-5 Dinner Party Scenario
0 Comments
Leave a Reply. |