Closed enricguso closed 7 months ago
I ended up doing a speaker-based train/val/test separation of the DNS4 (DNS5 speakerphone track). Attaching the files here in case it is of use to anyone, i hope it's ok to do this. You should join training_set_speech0x
into a single file.
test_set_noise.txt test_set_speech.txt test_set_rirs.txt training_set_speech03.txt training_set_speech02.txt training_set_speech01.txt training_set_speech00.txt training_set_noise.txt training_set_rirs.txt validation_set_speech.txt validation_set_rirs.txt validation_set_noise.txt
I made a guide how to prepare dataset and train model
https://github.com/WoodieDudy/open-source-stuff/tree/main/DeepFilerNet/prepare_dataset
Hello! Thanks for sharing this model, really cool. Is it possible that you add to the repo the .txt file that specifies which DNS4/5 files where used? In my opinion it's a bit misleading trying to replicate the training without it. Thank you!