DNS recipe (#1742)

* First commit

* add recipe in recipes.csv

* fixing pre-commit errors

* update recipes.csv

* update recipes

* remove unused models

* major fix- dns_downloader script, add results

* minor changes

* formatting;  parallel_download more stable

* add Readme

* add recipes/DNS.csv

* add HF link in readme

* remove 8k hparams file

* fix conclficts

* fix HF links

* fix trailing whitepsace

* fix yamllint conflicts

* add scripts for DNSMOS evaluation

* update instructions to download DNSMOS evaluation models

* Suport for baseline model NSNet2 evaluation discarded

* update readme and documentation

* update readme and documentation

* update readme, add docstring, update hparams header

* fix tests/recipes/DNS.csv

* remove unused params

* add script to shard DNS 4 data

* script now saves the synthesized clean-noisy as shards

* update readme

* fix trailing whitespace

* remove option to store synthesized data as wavs

* save train and valid shards

* training uses webdataset

* minor fixes

* fix pre-commit conflicts

* some more fixes

* removed forced dtype change

* fix recipes/DNS.csv

* script now takes all shards in the given dir for training

* readme updated, test-recipe updated, other cosmetic update

* pre-commit conflicts fixed

* Delete tests/samples/annotation/separation_train.csv

this file is no more needed

* remove samples from repo and put them on dropbox (to avoid pushing large files)

* restore separation_train.csv (otherwise pytest fails)

* sort extra req

* add dropbox link

---------

Co-authored-by: root <root@serv-3317.kl.dfki.de>
Co-authored-by: root <root@graz.kl.dfki.de>
Co-authored-by: root <root@serv-9208.kl.dfki.de>
Co-authored-by: sangeet2020 <15uec053@gmail.com>
Co-authored-by: Mirco Ravanelli <mirco.ravanelli@gmail.com>