DENSE

Dynamic Embedding Causal Target Speech Extraction

Official Code for Dynamic Embedding Causal Target Speech Extraction

For the demos, please visit Demo

Introduction

Due to the size limited, large files are not uploaded.

Most of the codes remain same as TD-SpeakerBeam. Thanks for open source.

Some changes:

see egs/libri2mix_sep2vec
For eval.py to use real chunk-wise causal target speech extraction, DDP is used and the parameter (1531) is set fixed as the receptive length of TCN for simply.
For system.py, please carefully check SystemPredictedTeacherForcing and SystemPredictedParis. Some parameters should be checked before running. Sorry for the inconvenience.
Open-sourced without carefully check the code style, maybe update sooner or later!

Name		Name	Last commit message	Last commit date
Latest commit History 42 Commits
bad_cases		bad_cases
dense		dense
dense_img		dense_img
good_cases		good_cases
wav_output		wav_output
README.md		README.md
index.html		index.html