Add files using upload-large-folder tool

.gitattributes

@@ -4,3 +4,36 @@ UVR_MDXNET_3_9662.onnx filter=lfs diff=lfs merge=lfs -text
 UVR_MDXNET_1_9703.onnx filter=lfs diff=lfs merge=lfs -text
 scnet_checkpoint_musdb18.ckpt filter=lfs diff=lfs merge=lfs -text
 UVR-MDX-NET_Crowd_HQ_1.onnx filter=lfs diff=lfs merge=lfs -text
+7d865c68-3d5dd56b.th filter=lfs diff=lfs merge=lfs -text
+UVR_MDXNET_2_9682.onnx filter=lfs diff=lfs merge=lfs -text
+0d19c1c6-0f06f20e.th filter=lfs diff=lfs merge=lfs -text
+902315c2-b39ce9c9.th filter=lfs diff=lfs merge=lfs -text
+5d2d6c55-db83574e.th filter=lfs diff=lfs merge=lfs -text
+75fc33f5-1941ce65.th filter=lfs diff=lfs merge=lfs -text
+17_HP-Wind_Inst-UVR.pth filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET_Main_390.onnx filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET_Main_406.onnx filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET_Inst_187_beta.onnx filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET_Inst_82_beta.onnx filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET_Main_340.onnx filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET_Inst_90_beta.onnx filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET-Voc_FT.onnx filter=lfs diff=lfs merge=lfs -text
+7ecf8ec1-70f50cc9.th filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET-Inst_1.onnx filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET-Inst_2.onnx filter=lfs diff=lfs merge=lfs -text
+UVR_MDXNET_9482.onnx filter=lfs diff=lfs merge=lfs -text
+9a6b4851-03af0aa6.th filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET-Inst_3.onnx filter=lfs diff=lfs merge=lfs -text
+f7e0c4bc-ba3fe64a.th filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET_Main_427.onnx filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET-Inst_Main.onnx filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET-Inst_full_292.onnx filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET_Main_438.onnx filter=lfs diff=lfs merge=lfs -text
+UVR-De-Reverb-aufr33-jarredou.pth filter=lfs diff=lfs merge=lfs -text
+kuielab_b_drums.onnx filter=lfs diff=lfs merge=lfs -text
+kuielab_a_drums.onnx filter=lfs diff=lfs merge=lfs -text
+d12395a8-e57c48e6.th filter=lfs diff=lfs merge=lfs -text
+kuielab_a_vocals.onnx filter=lfs diff=lfs merge=lfs -text
+kuielab_a_other.onnx filter=lfs diff=lfs merge=lfs -text
+kuielab_b_bass.onnx filter=lfs diff=lfs merge=lfs -text
+UVR-MDX-NET-Inst_HQ_2.onnx filter=lfs diff=lfs merge=lfs -text

0d19c1c6-0f06f20e.th

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f06f20ed6ddc8058fa72ccc4845f3a88916eff7d007b623924193de217bbcf4
+size 178048329

17_HP-Wind_Inst-UVR.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:acc6d472b4b478da9c9ab5af45b167749e05a7f65b30c7d5988b3700a513aeee
+size 223661285

5d2d6c55-db83574e.th

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db83574e05b2308f76e2764819da673f2d16d437b9e619f5fcb72f275fc0e24f
+size 167391595

75fc33f5-1941ce65.th

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1941ce654b11df4132b9f4eae408556b4c83fad6fe26b4bc0dbcb36b975befb3
+size 167407275

7d865c68-3d5dd56b.th

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d5dd56b5bc986f136dff98655ded22b2b033f465ccec7a28640a6b15fd71ed6
+size 167918783

7ecf8ec1-70f50cc9.th

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70f50cc947d08f32e6dd8e2b687d398fa5ef9e51d1bd7600e32205d1f44be6b9
+size 178048329

902315c2-b39ce9c9.th

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b39ce9c97b5603fbaee99ec59fe1d2db570c77a37356c9c85542d881517c3302
+size 167405611

9a6b4851-03af0aa6.th

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03af0aa64af2c4f0795659c265fdead238fab31c064a8a4b5b051c1b22c8ad48
+size 267602121

BS_Inst_EXP_VRL.yaml

@@ -0,0 +1,124 @@
+audio:
+  chunk_size: 485100 #352800 #485100
+  dim_f: 1024
+  dim_t: 801
+  hop_length: 441
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.000
+
+model:
+  dim: 384
+  depth: 12
+  stereo: true
+  num_stems: 1
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  freqs_per_bands: !!python/tuple
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 128
+    - 129
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0
+  ff_dropout: 0
+  flash_attn: true
+  dim_freqs_in: 1025
+  stft_n_fft: 2048
+  stft_hop_length: 441
+  stft_win_length: 2048
+  stft_normalized: false
+  mask_estimator_depth: 2
+  multi_stft_resolution_loss_weight: 1.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+  - 4096
+  - 2048
+  - 1024
+  - 512
+  - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: False
+training:
+  batch_size: 1
+  gradient_accumulation_steps: 1
+  grad_clip: 0
+  instruments:
+  - Vocals
+  - Instrumental
+  lr: 1.0e-04
+  patience: 2
+  reduce_factor: 0.95
+  target_instrument: Instrumental
+  num_epochs: 1
+  num_steps: 1000
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adamw
+  other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental
+  use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true
+
+inference:
+  batch_size: 1
+  dim_t: 1101
+  num_overlap: 2

UVR-De-Reverb-aufr33-jarredou.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fff726b0adf09f2eadc2151645557db81e1a01fe15d91f9bbbfa9b50a007f1fd
+size 58928133

UVR-MDX-NET-Inst_1.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ca53f94b7a0cbb04fcfcc8f3ea5ec1ae22cd8ad044f5e673588859f83976f5e
+size 66759214

UVR-MDX-NET-Inst_2.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3a96a664d28b52db9def0a9cae9a16dbb524d8325bfe8f0ac64ac5d231456bc
+size 66759214

UVR-MDX-NET-Inst_3.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b7834e2972158d8c9864e7376e3a7d084079c80a23f38dc31c4b0a4e901a1cb
+size 66759214

UVR-MDX-NET-Inst_HQ_2.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:197f8ab296df850f961e68c595f6649acb7d9e621b5600b460f3458967299112
+size 66759214

UVR-MDX-NET-Inst_Main.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ab401dfe4a548b87deb64f975294bd56ff946aa32903f53b4b24bb13b2cce1e
+size 52786726

UVR-MDX-NET-Inst_full_292.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:020f6b65fa219fb7c285e4f3fc2863bf22daf03c4c93e547b6d13d5f2757a7ec
+size 66759214

UVR-MDX-NET-Voc_FT.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:534b2070fcc7df514b13ef660dc8cbb328679c2374d04354a5c42bb14ecce111
+size 66762490

UVR-MDX-NET_Inst_187_beta.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c74566f3c3033cacba996328b2ee90bf77ef79ea6c35b7841df183b7906f54a5
+size 66759214

UVR-MDX-NET_Inst_82_beta.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6c268302f09ab53687072618e056a611272a7e2c3fd9b3b59164da152f3588e
+size 66759214

UVR-MDX-NET_Inst_90_beta.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d902868a46575aea6ee2335736ff3b53faf497a6bdaa1b864e0fd84eb1b42a5
+size 66759214

UVR-MDX-NET_Main_340.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78792633b4007755af12ecde20f709b4f0b99563b1d25fe0a501ed2122aff218
+size 66759214

UVR-MDX-NET_Main_390.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:286c4f0847ca837e2c3f4c4058f756d5f150cbf080506aa6f33a2847aba92e8c
+size 66759214

UVR-MDX-NET_Main_406.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f158816a44eef1f0ba0f48b813cbfcf460ed1c70a754af3609ade44aaf7d1b23
+size 66759214

UVR-MDX-NET_Main_427.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95275802a27801b97e3c0552b6eaa69f9bb3bd7df53cdf0536cce0a753f702cc
+size 66759214

UVR-MDX-NET_Main_438.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5e1ad93587a163a0987a0168b99a2ad875c0d9bfc3afb596b7c36b09c7f5c26
+size 66759214

UVR_MDXNET_2_9682.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1deb7295acd3206bc9582a5d92f1b0a74bf3f41c7c1fb78a0ac0123cde4372db
+size 29704436

UVR_MDXNET_9482.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4f365207c56deb115bceedff3ad8fe98a751c745f9e370cecec6226b8b47184
+size 29704436

assets/download_models.py

@@ -0,0 +1,26 @@
+import requests
+import os
+import argparse
+from huggingface_hub import snapshot_download
+from model_tools import download_files_from_txt, download_hf_repo
+
+REPO_OWNER = "lainlives"
+REPO_NAME = "audio-separator-models"
+repo_type = "model"
+
+
+if __name__ == "__main__":
+    token = os.getenv("HF_TOKEN")
+    repo_id = f"{REPO_OWNER}/{REPO_NAME}"
+    parser = argparse.ArgumentParser(description="Model Download Tool")
+    local_dir = os.getcwd()
+    parser.add_argument(
+        "--fromtxt",
+        action="store_true",
+        help="Download or process based on a text file instead of cloning Hugging Face repo",
+    )
+    args = parser.parse_args()
+    if args.fromtxt:
+        download_files_from_txt("assets/links.txt", local_dir)
+    else:
+        download_hf_repo(repo_id, local_dir, token)

assets/json_link_processor.py

@@ -0,0 +1,7 @@
+import json
+import urllib.request
+import os
+import model_tools as mt
+
+if __name__ == "__main__":
+    mt.get_links_from_json("models.json")

assets/luvr5-ui-models.txt

@@ -0,0 +1,248 @@
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_aspiration_mel_band_roformer.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_aspiration_mel_band_roformer.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_bs_roformer_instrumental_resurrection_unwa.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_bs_roformer_vocals_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_bs_roformer_vocals_resurrection_unwa.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_bs_roformer_vocals_revive_unwa.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_bs_roformer_vocals_revive_unwa.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_bs_roformer_vocals_revive_unwa.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_chorus_male_female_bs_roformer.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_chorus_male_female_bs_roformer.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_dereverb-echo_mel_band_roformer.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_dereverb-echo_mel_band_roformer_sdr_13.4843_v2.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_dereverb_echo_mel_band_roformer_v2.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_dereverb_echo_mel_band_roformer_v2.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_dereverb_echo_mel_band_roformer_v2.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_dereverb_mdx23c.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_drumsep_mdx23c.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_bleed_suppressor_v1.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_becruily.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_instrumental_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_karaoke_becruily.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_karaoke_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_karaoke_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_kim_ft_unwa.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_kim_ft_unwa.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_kim_ft_unwa.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_kim_ft_unwa.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_vocal_fullness_aname.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_vocals_becruily.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_vocals_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_vocals_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_vocals_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_vocals_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_vocals_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_vocals_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_vocals_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_mel_band_roformer_vocals_gabox.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_melband_roformer_big_beta5e.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_melbandroformer_big_beta4.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_melbandroformer_big_beta6.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_melbandroformer_big_beta6x.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_melbandroformer_inst.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_melbandroformer_inst.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_melbandroformer_inst.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_melbandroformer_inst.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_melbandroformer_inst_v2.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_melbandroformer_instvoc_duality.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_melbandroformer_instvoc_duality.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_vocals_mel_band_roformer_big_v1_ft.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_vocals_mel_band_roformer_ft.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_vocals_mel_band_roformer_ft.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_vocals_mel_band_roformer_ft.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/config_vocals_mel_band_roformer_ft.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768_config.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/denoise_mel_band_roformer_aufr33_sdr_27.9959_config.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/dereverb_mel_band_roformer_anvuew.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/dereverb_mel_band_roformer_anvuew.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/dereverb_mel_band_roformer_anvuew.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/deverb_bs_roformer_8_384dim_10depth_config.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144_config.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956_config.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/model_2_stem_061321.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/model_2_stem_full_band_8k.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/model_2_stem_full_band_8k.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/model_bs_roformer_ep_317_sdr_12.9755.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/model_bs_roformer_ep_368_sdr_12.9628.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/model_bs_roformer_ep_937_sdr_10.5309.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/model_mel_band_roformer_ep_3005_sdr_11.4360.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/vocals_mel_band_roformer.yaml
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/10_SP-UVR-2B-32000-1.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/11_SP-UVR-2B-32000-2.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/12_SP-UVR-3B-44100.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/13_SP-UVR-4B-44100-1.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/14_SP-UVR-4B-44100-2.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/15_SP-UVR-MID-44100-1.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/16_SP-UVR-MID-44100-2.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/17_HP-Wind_Inst-UVR.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/1_HP-UVR.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/2_HP-UVR.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/3_HP-Vocal-UVR.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/4_HP-Vocal-UVR.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/5_HP-Karaoke-UVR.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/6_HP-Karaoke-UVR.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/7_HP2-UVR.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/8_HP2-UVR.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/9_HP2-UVR.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MGM_HIGHEND_v4.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MGM_LOWEND_A_v4.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MGM_LOWEND_B_v4.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MGM_MAIN_v4.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-BVE-4B_SN-44100-1.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-BVE-4B_SN-44100-2.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-De-Echo-Aggressive.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-De-Echo-Normal.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-De-Reverb-aufr33-jarredou.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-DeEcho-DeReverb.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-DeNoise-Lite.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-DeNoise.pth
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MDX23C-8KFFT-InstVoc_HQ.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MDX23C-8KFFT-InstVoc_HQ_2.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MDX23C-De-Reverb-aufr33-jarredou.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MDX23C-DrumSep-aufr33-jarredou.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MDX23C_D1581.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MelBandRoformerBigSYHFTV1.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MelBandRoformerSYHFT.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MelBandRoformerSYHFTV2.5.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MelBandRoformerSYHFTV2.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/MelBandRoformerSYHFTV3Epsilon.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/aspiration_mel_band_roformer_less_aggr_sdr_18.1201.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/aspiration_mel_band_roformer_sdr_18.9845.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/bs_roformer_instrumental_resurrection_unwa.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/bs_roformer_male_female_by_aufr33_sdr_7.2889.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/bs_roformer_vocals_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/bs_roformer_vocals_resurrection_unwa.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/bs_roformer_vocals_revive_unwa.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/bs_roformer_vocals_revive_v2_unwa.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/bs_roformer_vocals_revive_v3e_unwa.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/dereverb-echo_mel_band_roformer_sdr_10.0169.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/dereverb-echo_mel_band_roformer_sdr_13.4843_v2.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/dereverb_big_mbr_ep_362.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/dereverb_echo_mbr_fused.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/dereverb_mel_band_roformer_anvuew_sdr_19.1729.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/dereverb_mel_band_roformer_less_aggressive_anvuew_sdr_18.8050.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/dereverb_mel_band_roformer_mono_anvuew.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/dereverb_super_big_mbr_ep_346.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/deverb_bs_roformer_8_384dim_10depth.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_bleed_suppressor_v1.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_denoise_debleed_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_2_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_3_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_becruily.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_bleedless_v1_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_bleedless_v2_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_bleedless_v3_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_fullness_noise_v4_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_fullness_v1_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_fullness_v2_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_fullness_v3_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_fv7z_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_fv8_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_fvx_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_instv5_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_instv5n_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_instv6_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_instv6n_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_instv7_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_instv7n_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_instv8_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_instrumental_instv8n_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_karaoke_becruily.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_karaoke_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_karaoke_gabox_v2.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_kim_ft2_bleedless_unwa.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_kim_ft2_unwa.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_kim_ft3_unwa.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_kim_ft_unwa.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_vocal_fullness_aname.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_vocals_becruily.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_vocals_fv1_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_vocals_fv2_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_vocals_fv3_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_vocals_fv4_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_vocals_fv5_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_vocals_fv6_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_vocals_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/mel_band_roformer_vocals_v2_gabox.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/melband_roformer_big_beta4.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/melband_roformer_big_beta5e.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/melband_roformer_big_beta6.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/melband_roformer_big_beta6x.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/melband_roformer_inst_v1.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/melband_roformer_inst_v1_plus.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/melband_roformer_inst_v1e.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/melband_roformer_inst_v1e_plus.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/melband_roformer_inst_v2.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/melband_roformer_instvoc_duality_v1.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/melband_roformer_instvox_duality_v2.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/model_bs_roformer_ep_317_sdr_12.9755.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/model_bs_roformer_ep_368_sdr_12.9628.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/model_bs_roformer_ep_937_sdr_10.5309.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/model_chorus_bs_roformer_ep_267_sdr_24.1275.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/model_mel_band_roformer_ep_3005_sdr_11.4360.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/vocals_mel_band_roformer.ckpt
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/Kim_Inst.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/Kim_Vocal_1.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/Kim_Vocal_2.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/Reverb_HQ_By_FoxJoy.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET-Inst_1.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET-Inst_2.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET-Inst_3.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET-Inst_HQ_1.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET-Inst_HQ_2.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET-Inst_HQ_3.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET-Inst_HQ_4.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET-Inst_HQ_5.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET-Inst_Main.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET-Inst_full_292.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET-Voc_FT.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET_Crowd_HQ_1.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET_Inst_187_beta.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET_Inst_82_beta.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET_Inst_90_beta.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET_Main_340.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET_Main_390.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET_Main_406.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET_Main_427.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR-MDX-NET_Main_438.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR_MDXNET_1_9703.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR_MDXNET_2_9682.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR_MDXNET_3_9662.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR_MDXNET_9482.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR_MDXNET_KARA.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR_MDXNET_KARA_2.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/UVR_MDXNET_Main.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/kuielab_a_bass.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/kuielab_a_drums.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/kuielab_a_other.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/kuielab_a_vocals.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/kuielab_b_bass.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/kuielab_b_drums.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/kuielab_b_other.onnx
+https://huggingface.co/lainlives/audio-separator-models/resolve/main/kuielab_b_vocals.onnx

assets/model_url_sorter.py

@@ -0,0 +1,5 @@
+import os
+from model_tools import sort_links_by_extension
+
+if __name__ == "__main__":
+    sort_links_by_extension("links.txt", "sorted_links.txt")

assets/upload_models.py

@@ -0,0 +1,33 @@
+#! /usr/bin/env python3
+import os
+import argparse
+from huggingface_hub import HfApi
+import model_tools as mt
+
+REPO_OWNER = "lainlives"
+REPO_NAME = "audio-separator-models"
+repo_type = "model"
+HF_TOKEN = os.getenv("HF_TOKEN")
+
+
+if __name__ == "__main__":
+    repo_id = f"{REPO_OWNER}/{REPO_NAME}"
+
+    parser = argparse.ArgumentParser(description="Model Upload Tool")
+    parser.add_argument(
+        "--large",
+        action="store_true",
+        help="Upload using upload-large-folder",
+    )
+    parser.add_argument(
+        "--token",
+        type=str,
+        default=HF_TOKEN,
+        dest="HF_TOKEN",
+        help=f"Your access token.",
+    )
+    args = parser.parse_args()
+    if args.large:
+        mt.push_large_folder_to_hf(repo_id, repo_type)
+    else:
+        mt.push_to_hf(repo_id, repo_type)

aufr33-jarredou_DrumSep_model_mdx23c_ep_141_sdr_10.8059.yaml

@@ -0,0 +1,87 @@
+audio:
+  chunk_size: 130560
+  dim_f: 1024
+  dim_t: 256
+  hop_length: 512
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.001
+
+model:
+  act: gelu
+  bottleneck_factor: 4
+  growth: 128
+  norm: InstanceNorm
+  num_blocks_per_scale: 2
+  num_channels: 128
+  num_scales: 5
+  num_subbands: 4
+  scale:
+  - 2
+  - 2
+
+training:
+  batch_size: 12
+  gradient_accumulation_steps: 1
+  grad_clip: 0
+  instruments:
+  - Kick
+  - Snare
+  - Toms
+  - Hh
+  - Ride
+  - Crash
+  lr: 9.0e-05
+  patience: 30
+  reduce_factor: 0.95
+  target_instrument: null
+  num_epochs: 1000
+  num_steps: 1268
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adam
+  other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental
+  use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true
+
+augmentations:
+  enable: true # enable or disable all augmentations (to fast disable if needed)
+  loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max)
+  loudness_min: 0.5
+  loudness_max: 1.5
+  mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3)
+  mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02)
+    - 0.2
+    - 0.02
+  mixup_loudness_min: 0.5
+  mixup_loudness_max: 1.5
+
+  # apply mp3 compression to mixture only (emulate downloading mp3 from internet)
+  mp3_compression_on_mixture: 0.0
+  mp3_compression_on_mixture_bitrate_min: 32
+  mp3_compression_on_mixture_bitrate_max: 320
+  mp3_compression_on_mixture_backend: "lameenc"
+
+  all:
+    channel_shuffle: 0.5 # Set 0 or lower to disable
+    random_inverse: 0.01 # inverse track (better lower probability)
+    random_polarity: 0.5 # polarity change (multiply waveform to -1)
+    mp3_compression: 0.0
+    mp3_compression_min_bitrate: 32
+    mp3_compression_max_bitrate: 320
+    mp3_compression_backend: "lameenc"
+    pitch_shift: 0.1
+    pitch_shift_min_semitones: -3
+    pitch_shift_max_semitones: 3
+    seven_band_parametric_eq: 0.5
+    seven_band_parametric_eq_min_gain_db: -6
+    seven_band_parametric_eq_max_gain_db: 6
+    tanh_distortion: 0.2
+    tanh_distortion_min: 0.1
+    tanh_distortion_max: 0.5
+    
+inference:
+  batch_size: 1
+  dim_t: 256
+  num_overlap: 4

calculate-model-hashes.py

@@ -0,0 +1,105 @@
+#!/usr/bin/env python3
+
+import os
+import sys
+import json
+import hashlib
+import requests
+
+MODEL_CACHE_PATH = "/tmp/audio-separator-models"
+VR_MODEL_DATA_LOCAL_PATH = f"{MODEL_CACHE_PATH}/vr_model_data.json"
+MDX_MODEL_DATA_LOCAL_PATH = f"{MODEL_CACHE_PATH}/mdx_model_data.json"
+
+MODEL_DATA_URL_PREFIX = "https://raw.githubusercontent.com/TRvlvr/application_data/main"
+VR_MODEL_DATA_URL = f"{MODEL_DATA_URL_PREFIX}/vr_model_data/model_data_new.json"
+MDX_MODEL_DATA_URL = f"{MODEL_DATA_URL_PREFIX}/mdx_model_data/model_data_new.json"
+
+OUTPUT_PATH = f"{MODEL_CACHE_PATH}/model_hashes.json"
+
+
+def get_model_hash(model_path):
+    """
+    Get the hash of a model file
+    """
+    # print(f"Getting hash for model at {model_path}")
+    try:
+        with open(model_path, "rb") as f:
+            f.seek(-10000 * 1024, 2)  # Move the file pointer 10MB before the end of the file
+            hash_result = hashlib.md5(f.read()).hexdigest()
+            # print(f"Hash for {model_path}: {hash_result}")
+            return hash_result
+    except IOError:
+        with open(model_path, "rb") as f:
+            hash_result = hashlib.md5(f.read()).hexdigest()
+            # print(f"IOError encountered, hash for {model_path}: {hash_result}")
+            return hash_result
+
+
+def download_file_if_missing(url, local_path):
+    """
+    Download a file from a URL if it doesn't exist locally
+    """
+    print(f"Checking if {local_path} needs to be downloaded from {url}")
+    if not os.path.exists(local_path):
+        print(f"Downloading {url} to {local_path}")
+        with requests.get(url, stream=True, timeout=10) as r:
+            r.raise_for_status()
+            with open(local_path, "wb") as f:
+                for chunk in r.iter_content(chunk_size=8192):
+                    f.write(chunk)
+        print(f"Downloaded {url} to {local_path}")
+    else:
+        print(f"{local_path} already exists. Skipping download.")
+
+
+def load_json_data(file_path):
+    """
+    Load JSON data from a file
+    """
+    print(f"Loading JSON data from {file_path}")
+    try:
+        with open(file_path, "r", encoding="utf-8") as file:
+            data = json.load(file)
+            print(f"Loaded JSON data successfully from {file_path}")
+            return data
+    except FileNotFoundError:
+        print(f"{file_path} not found.")
+        sys.exit(1)
+
+
+def iterate_and_hash(directory):
+    """
+    Iterate through a directory and hash all model files
+    """
+    print(f"Iterating through directory {directory} to hash model files")
+    model_files = [(file, os.path.join(root, file)) for root, _, files in os.walk(directory) for file in files if file.endswith((".pth", ".onnx"))]
+
+    download_file_if_missing(VR_MODEL_DATA_URL, VR_MODEL_DATA_LOCAL_PATH)
+    download_file_if_missing(MDX_MODEL_DATA_URL, MDX_MODEL_DATA_LOCAL_PATH)
+
+    vr_model_data = load_json_data(VR_MODEL_DATA_LOCAL_PATH)
+    mdx_model_data = load_json_data(MDX_MODEL_DATA_LOCAL_PATH)
+
+    combined_model_params = {
+        **vr_model_data,
+        **mdx_model_data,
+    }
+
+    model_info_list = []
+    for file, file_path in sorted(model_files):
+        file_hash = get_model_hash(file_path)
+        model_info = {
+            "file": file,
+            "hash": file_hash,
+            "params": combined_model_params.get(file_hash, "Parameters not found"),
+        }
+        model_info_list.append(model_info)
+
+    print(f"Writing model info list to {OUTPUT_PATH}")
+    with open(OUTPUT_PATH, "w", encoding="utf-8") as json_file:
+        json.dump(model_info_list, json_file, indent=4)
+        print(f"Successfully wrote model info list to {OUTPUT_PATH}")
+
+
+if __name__ == "__main__":
+    iterate_and_hash(MODEL_CACHE_PATH)

config_bs_roformer_karaoke_anvuew.yaml

@@ -0,0 +1,131 @@
+audio:
+  chunk_size: 640000
+  dim_f: 1024
+  dim_t: 801 # don't work (use in model)
+  hop_length: 441 # don't work (use in model)
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.000
+
+model:
+  dim: 256
+  depth: 12
+  stereo: true
+  num_stems: 1
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  linear_transformer_depth: 0
+  freqs_per_bands: !!python/tuple
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 128
+    - 129
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0.0
+  ff_dropout: 0.0
+  flash_attn: true
+  dim_freqs_in: 1025
+  stft_n_fft: 2048
+  stft_hop_length: 512
+  stft_win_length: 2048
+  stft_normalized: false
+  mask_estimator_depth: 2
+  multi_stft_resolution_loss_weight: 1.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+  - 4096
+  - 2048
+  - 1024
+  - 512
+  - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: False
+  mlp_expansion_factor: 4
+  use_torch_checkpoint: True
+  skip_connection: False
+
+
+training:
+  batch_size: 1
+  gradient_accumulation_steps: 1
+  grad_clip: 0
+  instruments: ['Vocals', 'Instrumental']
+  lr: 5.0e-5
+  patience: 7
+  reduce_factor: 0.75
+  target_instrument: Vocals
+  num_epochs: 1000
+  num_steps: 1000
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adam
+  other_fix: False # it's needed for checking on multisong dataset if other is actually instrumental
+  use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true
+
+
+
+
+inference:
+  batch_size: 2
+  dim_t: 1251
+  num_overlap: 4

config_chorus_male_female_bs_roformer.yaml

@@ -0,0 +1,125 @@
+audio:
+  chunk_size: 352800
+  dim_f: 1024
+  dim_t: 801 # don't work (use in model)
+  hop_length: 441 # don't work (use in model)
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.000
+
+model:
+  dim: 384
+  depth: 8
+  stereo: true
+  num_stems: 2
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  freqs_per_bands: !!python/tuple
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 128
+    - 129
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0.0
+  ff_dropout: 0.0
+  flash_attn: true
+  dim_freqs_in: 1025
+  stft_n_fft: 2048
+  stft_hop_length: 441
+  stft_win_length: 2048
+  stft_normalized: false
+  mask_estimator_depth: 2
+  multi_stft_resolution_loss_weight: 1.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+  - 4096
+  - 2048
+  - 1024
+  - 512
+  - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: False
+
+training:
+  batch_size: 1
+  gradient_accumulation_steps: 1
+  grad_clip: 0
+  instruments:
+  - male
+  - female
+  lr: 1.0e-05
+  patience: 2
+  reduce_factor: 0.95
+  target_instrument: null
+  num_epochs: 1000
+  num_steps: 1000
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adam
+  other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental
+  use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true
+
+inference:
+  batch_size: 1
+  dim_t: 801
+  num_overlap: 2

config_dereverb-echo_mel_band_roformer_sdr_13.4843_v2.yaml

@@ -0,0 +1,64 @@
+audio:
+  chunk_size: 352800
+  dim_f: 1024
+  dim_t: 801
+  hop_length: 441
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.000
+
+model:
+  dim: 256
+  depth: 8
+  stereo: true
+  num_stems: 1
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  num_bands: 60
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0.1
+  ff_dropout: 0.1
+  flash_attn: True
+  dim_freqs_in: 1025
+  sample_rate: 44100
+  stft_n_fft: 2048
+  stft_hop_length: 441
+  stft_win_length: 2048
+  stft_normalized: False
+  mask_estimator_depth: 2
+  multi_stft_resolution_loss_weight: 1.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+  - 4096
+  - 2048
+  - 1024
+  - 512
+  - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: False
+
+training:
+  batch_size: 1
+  gradient_accumulation_steps: 8
+  grad_clip: 0
+  instruments:
+  - dry
+  - No dry
+  lr: 1.0e-05
+  patience: 2
+  reduce_factor: 0.95
+  target_instrument: dry
+  num_epochs: 1000
+  num_steps: 1000
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adam
+  other_fix: false
+  use_amp: true
+
+inference:
+  batch_size: 1
+  dim_t: 801
+  num_overlap: 4

config_dereverb_echo_mel_band_roformer_v2.yaml

@@ -0,0 +1,64 @@
+audio:
+  chunk_size: 352800
+  dim_f: 1024
+  dim_t: 801
+  hop_length: 441
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.000
+
+model:
+  dim: 256
+  depth: 8
+  stereo: true
+  num_stems: 1
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  num_bands: 60
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0.1
+  ff_dropout: 0.1
+  flash_attn: True
+  dim_freqs_in: 1025
+  sample_rate: 44100
+  stft_n_fft: 2048
+  stft_hop_length: 441
+  stft_win_length: 2048
+  stft_normalized: False
+  mask_estimator_depth: 2
+  multi_stft_resolution_loss_weight: 1.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+  - 4096
+  - 2048
+  - 1024
+  - 512
+  - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: False
+
+training:
+  batch_size: 1
+  gradient_accumulation_steps: 8
+  grad_clip: 0
+  instruments:
+  - dry
+  - other
+  lr: 1.0e-05
+  patience: 2
+  reduce_factor: 0.95
+  target_instrument: dry
+  num_epochs: 1000
+  num_steps: 1000
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adam
+  other_fix: false
+  use_amp: true
+
+inference:
+  batch_size: 1
+  dim_t: 801
+  num_overlap: 4

config_dereverb_echo_mel_band_roformer_v2.yaml.1

@@ -0,0 +1,64 @@
+audio:
+  chunk_size: 352800
+  dim_f: 1024
+  dim_t: 801
+  hop_length: 441
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.000
+
+model:
+  dim: 256
+  depth: 8
+  stereo: true
+  num_stems: 1
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  num_bands: 60
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0.1
+  ff_dropout: 0.1
+  flash_attn: True
+  dim_freqs_in: 1025
+  sample_rate: 44100
+  stft_n_fft: 2048
+  stft_hop_length: 441
+  stft_win_length: 2048
+  stft_normalized: False
+  mask_estimator_depth: 2
+  multi_stft_resolution_loss_weight: 1.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+  - 4096
+  - 2048
+  - 1024
+  - 512
+  - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: False
+
+training:
+  batch_size: 1
+  gradient_accumulation_steps: 8
+  grad_clip: 0
+  instruments:
+  - dry
+  - other
+  lr: 1.0e-05
+  patience: 2
+  reduce_factor: 0.95
+  target_instrument: dry
+  num_epochs: 1000
+  num_steps: 1000
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adam
+  other_fix: false
+  use_amp: true
+
+inference:
+  batch_size: 1
+  dim_t: 801
+  num_overlap: 4

config_dereverb_echo_mel_band_roformer_v2.yaml.2

@@ -0,0 +1,64 @@
+audio:
+  chunk_size: 352800
+  dim_f: 1024
+  dim_t: 801
+  hop_length: 441
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.000
+
+model:
+  dim: 256
+  depth: 8
+  stereo: true
+  num_stems: 1
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  num_bands: 60
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0.1
+  ff_dropout: 0.1
+  flash_attn: True
+  dim_freqs_in: 1025
+  sample_rate: 44100
+  stft_n_fft: 2048
+  stft_hop_length: 441
+  stft_win_length: 2048
+  stft_normalized: False
+  mask_estimator_depth: 2
+  multi_stft_resolution_loss_weight: 1.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+  - 4096
+  - 2048
+  - 1024
+  - 512
+  - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: False
+
+training:
+  batch_size: 1
+  gradient_accumulation_steps: 8
+  grad_clip: 0
+  instruments:
+  - dry
+  - other
+  lr: 1.0e-05
+  patience: 2
+  reduce_factor: 0.95
+  target_instrument: dry
+  num_epochs: 1000
+  num_steps: 1000
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adam
+  other_fix: false
+  use_amp: true
+
+inference:
+  batch_size: 1
+  dim_t: 801
+  num_overlap: 4

config_dereverb_mdx23c.yaml

@@ -0,0 +1,135 @@
+audio:
+  chunk_size: 261120
+  dim_f: 4096
+  dim_t: 256
+  hop_length: 1024
+  n_fft: 8192
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.001
+
+model:
+  act: gelu
+  bottleneck_factor: 4
+  growth: 128
+  norm: InstanceNorm
+  num_blocks_per_scale: 2
+  num_channels: 128
+  num_scales: 5
+  num_subbands: 4
+  scale:
+  - 2
+  - 2
+
+training:
+  batch_size: 2
+  gradient_accumulation_steps: 1
+  grad_clip: 0
+  instruments:
+  - dry
+  - No dry
+  lr: 1.0e-06
+  patience: 4
+  reduce_factor: 0.93
+  target_instrument: null
+  num_epochs: 40
+  num_steps: 1000
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adamw
+  read_metadata_procs: 8 # Number of processes to use during metadata reading for dataset. Can speed up metadata generation
+  other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental
+  use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true
+
+augmentations:
+  enable: false # enable or disable all augmentations (to fast disable if needed)
+  loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max)
+  loudness_min: 0.5
+  loudness_max: 1.5
+  mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3)
+  mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02)
+    - 0.2
+    - 0.02
+  mixup_loudness_min: 0.5
+  mixup_loudness_max: 1.5
+
+  all:
+    channel_shuffle: 0.5 # Set 0 or lower to disable
+    random_inverse: 0.05 # inverse track (better lower probability)
+    random_polarity: 0.5 # polarity change (multiply waveform to -1)
+
+    # pedalboard chorus block
+    pedalboard_chorus: 0.001
+    pedalboard_chorus_rate_hz_min: 1.0
+    pedalboard_chorus_rate_hz_max: 7.0
+    pedalboard_chorus_depth_min: 0.25
+    pedalboard_chorus_depth_max: 0.95
+    pedalboard_chorus_centre_delay_ms_min: 3
+    pedalboard_chorus_centre_delay_ms_max: 10
+    pedalboard_chorus_feedback_min: 0.0
+    pedalboard_chorus_feedback_max: 0.01
+    pedalboard_chorus_mix_min: 0.1
+    pedalboard_chorus_mix_max: 0.9
+
+    # pedalboard phazer block
+    pedalboard_phazer: 0.001
+    pedalboard_phazer_rate_hz_min: 1.0
+    pedalboard_phazer_rate_hz_max: 10.0
+    pedalboard_phazer_depth_min: 0.25
+    pedalboard_phazer_depth_max: 0.95
+    pedalboard_phazer_centre_frequency_hz_min: 200
+    pedalboard_phazer_centre_frequency_hz_max: 12000
+    pedalboard_phazer_feedback_min: 0.0
+    pedalboard_phazer_feedback_max: 0.5
+    pedalboard_phazer_mix_min: 0.1
+    pedalboard_phazer_mix_max: 0.9
+
+    # pedalboard pitch shift block
+    pedalboard_pitch_shift: 0.01
+    pedalboard_pitch_shift_semitones_min: -7
+    pedalboard_pitch_shift_semitones_max: 7
+
+    # pedalboard resample block
+    pedalboard_resample: 0.001
+    pedalboard_resample_target_sample_rate_min: 4000
+    pedalboard_resample_target_sample_rate_max: 44100
+
+    mp3_compression_min_bitrate: 32
+    mp3_compression_max_bitrate: 320
+    mp3_compression_backend: "lameenc"
+
+  dry:
+    # pedalboard distortion block
+    pedalboard_distortion: 0.001
+    pedalboard_distortion_drive_db_min: 1.0
+    pedalboard_distortion_drive_db_max: 25.0
+    
+    tanh_distortion: 0.05
+    tanh_distortion_min: 0.1
+    tanh_distortion_max: 0.7
+    # pedalboard bitcrash block
+    pedalboard_bitcrash: 0.005
+    pedalboard_bitcrash_bit_depth_min: 4
+    pedalboard_bitcrash_bit_depth_max: 16
+
+    seven_band_parametric_eq: 0.24
+    seven_band_parametric_eq_min_gain_db: -9
+    seven_band_parametric_eq_max_gain_db: 9
+    
+    gaussian_noise: 0.005
+    gaussian_noise_min_amplitude: 0.001
+    gaussian_noise_max_amplitude: 0.01
+    
+    time_stretch: 0.01
+    time_stretch_min_rate: 0.8
+    time_stretch_max_rate: 1.25
+  other:
+    seven_band_parametric_eq: 0.24
+    seven_band_parametric_eq_min_gain_db: -9
+    seven_band_parametric_eq_max_gain_db: 9
+
+inference:
+  batch_size: 2
+  dim_t: 256
+  num_overlap: 4

config_dnr_bandit_bsrnn_multi_mus64.yaml

@@ -0,0 +1,78 @@
+name: "MultiMaskMultiSourceBandSplitRNN"
+audio:
+  chunk_size: 264600
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.001
+
+model:
+  in_channel: 1
+  stems: ['speech', 'music', 'effects']
+  band_specs: "musical"
+  n_bands: 64
+  fs: 44100
+  require_no_overlap: false
+  require_no_gap: true
+  normalize_channel_independently: false
+  treat_channel_as_feature: true
+  n_sqm_modules: 8
+  emb_dim: 128
+  rnn_dim: 256
+  bidirectional: true
+  rnn_type: "GRU"
+  mlp_dim: 512
+  hidden_activation: "Tanh"
+  hidden_activation_kwargs: null
+  complex_mask: true
+  n_fft: 2048
+  win_length: 2048
+  hop_length: 512
+  window_fn: "hann_window"
+  wkwargs: null
+  power: null
+  center: true
+  normalized: true
+  pad_mode: "constant"
+  onesided: true
+
+training:
+  batch_size: 4
+  gradient_accumulation_steps: 4
+  grad_clip: 0
+  instruments:
+  - Speech
+  - Music
+  - Effects
+  lr: 9.0e-05
+  patience: 2
+  reduce_factor: 0.95
+  target_instrument: null
+  num_epochs: 1000
+  num_steps: 1000
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adam
+  other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental
+  use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true
+
+augmentations:
+  enable: true # enable or disable all augmentations (to fast disable if needed)
+  loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max)
+  loudness_min: 0.5
+  loudness_max: 1.5
+  mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3)
+  mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02)
+    - 0.2
+    - 0.02
+  mixup_loudness_min: 0.5
+  mixup_loudness_max: 1.5
+  all:
+    channel_shuffle: 0.5 # Set 0 or lower to disable
+    random_inverse: 0.1 # inverse track (better lower probability)
+    random_polarity: 0.5 # polarity change (multiply waveform to -1)
+
+inference:
+  batch_size: 1
+  dim_t: 256
+  num_overlap: 4

config_dnr_bandit_v2_mus64.yaml

@@ -0,0 +1,78 @@
+cls: Bandit
+
+audio:
+  chunk_size: 384000
+  num_channels: 2
+  sample_rate: 48000
+  min_mean_abs: 0.000
+
+kwargs:
+  in_channels: 1
+  stems: ['speech', 'music', 'sfx']
+  band_type: musical
+  n_bands: 64
+  normalize_channel_independently: false
+  treat_channel_as_feature: true
+  n_sqm_modules: 8
+  emb_dim: 128
+  rnn_dim: 256
+  bidirectional: true
+  rnn_type: "GRU"
+  mlp_dim: 512
+  hidden_activation: "Tanh"
+  hidden_activation_kwargs:
+  complex_mask: true
+  use_freq_weights: true
+  n_fft: 2048
+  win_length: 2048
+  hop_length: 512
+  window_fn: "hann_window"
+  wkwargs:
+  power:
+  center: true
+  normalized: true
+  pad_mode: "reflect"
+  onesided: true
+
+training:
+  batch_size: 4
+  gradient_accumulation_steps: 4
+  grad_clip: 0
+  instruments:
+  - Speech
+  - Music
+  - Sfx
+  lr: 9.0e-05
+  patience: 2
+  reduce_factor: 0.95
+  target_instrument:
+  num_epochs: 1000
+  num_steps: 1000
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adam
+  other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental
+  use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true
+
+augmentations:
+  enable: true # enable or disable all augmentations (to fast disable if needed)
+  loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max)
+  loudness_min: 0.5
+  loudness_max: 1.5
+  mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3)
+  mixup_probs: !!python/tuple
+                              # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02)
+  - 0.2
+  - 0.02
+  mixup_loudness_min: 0.5
+  mixup_loudness_max: 1.5
+  all:
+    channel_shuffle: 0.5 # Set 0 or lower to disable
+    random_inverse: 0.1 # inverse track (better lower probability)
+    random_polarity: 0.5 # polarity change (multiply waveform to -1)
+
+inference:
+  batch_size: 8
+  dim_t: 256
+  num_overlap: 4

config_drumsep_mdx23c.yaml

@@ -0,0 +1,87 @@
+audio:
+  chunk_size: 130560
+  dim_f: 1024
+  dim_t: 256
+  hop_length: 512
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.001
+
+model:
+  act: gelu
+  bottleneck_factor: 4
+  growth: 128
+  norm: InstanceNorm
+  num_blocks_per_scale: 2
+  num_channels: 128
+  num_scales: 5
+  num_subbands: 4
+  scale:
+  - 2
+  - 2
+
+training:
+  batch_size: 12
+  gradient_accumulation_steps: 1
+  grad_clip: 0
+  instruments:
+  - kick
+  - snare
+  - toms
+  - hh
+  - ride
+  - crash
+  lr: 9.0e-05
+  patience: 30
+  reduce_factor: 0.95
+  target_instrument: null
+  num_epochs: 1000
+  num_steps: 1268
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adam
+  other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental
+  use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true
+
+augmentations:
+  enable: true # enable or disable all augmentations (to fast disable if needed)
+  loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max)
+  loudness_min: 0.5
+  loudness_max: 1.5
+  mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3)
+  mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02)
+    - 0.2
+    - 0.02
+  mixup_loudness_min: 0.5
+  mixup_loudness_max: 1.5
+
+  # apply mp3 compression to mixture only (emulate downloading mp3 from internet)
+  mp3_compression_on_mixture: 0.0
+  mp3_compression_on_mixture_bitrate_min: 32
+  mp3_compression_on_mixture_bitrate_max: 320
+  mp3_compression_on_mixture_backend: "lameenc"
+
+  all:
+    channel_shuffle: 0.5 # Set 0 or lower to disable
+    random_inverse: 0.01 # inverse track (better lower probability)
+    random_polarity: 0.5 # polarity change (multiply waveform to -1)
+    mp3_compression: 0.0
+    mp3_compression_min_bitrate: 32
+    mp3_compression_max_bitrate: 320
+    mp3_compression_backend: "lameenc"
+    pitch_shift: 0.1
+    pitch_shift_min_semitones: -3
+    pitch_shift_max_semitones: 3
+    seven_band_parametric_eq: 0.5
+    seven_band_parametric_eq_min_gain_db: -6
+    seven_band_parametric_eq_max_gain_db: 6
+    tanh_distortion: 0.2
+    tanh_distortion_min: 0.1
+    tanh_distortion_max: 0.5
+    
+inference:
+  batch_size: 1
+  dim_t: 256
+  num_overlap: 4

config_mel_band_roformer_bleed_suppressor_v1.yaml

@@ -0,0 +1,51 @@
+audio:
+  chunk_size: 485100
+  dim_f: 1024
+  dim_t: 801
+  hop_length: 441
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.000
+
+model:
+  dim: 384
+  depth: 6
+  stereo: true
+  num_stems: 1
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  num_bands: 60
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0
+  ff_dropout: 0
+  flash_attn: True
+  dim_freqs_in: 1025
+  sample_rate: 44100  # needed for mel filter bank from librosa
+  stft_n_fft: 2048
+  stft_hop_length: 441
+  stft_win_length: 2048
+  stft_normalized: False
+  mask_estimator_depth: 2
+  multi_stft_resolution_loss_weight: 1.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+  - 4096
+  - 2048
+  - 1024
+  - 512
+  - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: False
+
+training:
+  instruments:
+  - Instrumental
+  - Bleed
+  target_instrument: Instrumental
+  use_amp: True
+
+inference:
+  batch_size: 1
+  dim_t: 801
+  num_overlap: 2

config_mel_band_roformer_karaoke.yaml

@@ -0,0 +1,71 @@
+audio:
+  chunk_size: 352800
+  dim_f: 1024
+  dim_t: 256
+  hop_length: 441
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 000
+
+model:
+  dim: 384
+  depth: 6
+  stereo: true
+  num_stems: 1
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  num_bands: 60
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0
+  ff_dropout: 0
+  flash_attn: true
+  dim_freqs_in: 1025
+  sample_rate: 44100  # needed for mel filter bank from librosa
+  stft_n_fft: 2048
+  stft_hop_length: 441
+  stft_win_length: 2048
+  stft_normalized: false
+  mask_estimator_depth: 2
+  multi_stft_resolution_loss_weight: 1.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+  - 4096
+  - 2048
+  - 1024
+  - 512
+  - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: false
+
+training:
+  batch_size: 4
+  gradient_accumulation_steps: 1
+  grad_clip: 0
+  instruments:
+  - Vocals
+  - Instrumental
+  lr: 1.0e-05
+  patience: 2
+  reduce_factor: 0.95
+  target_instrument: Vocals
+  num_epochs: 1000
+  num_steps: 2000
+  augmentation: false # enable augmentations by audiomentations and pedalboard
+  augmentation_type:
+  use_mp3_compress: false # Deprecated
+  augmentation_mix: false # Mix several stems of the same type with some probability
+  augmentation_loudness: false # randomly change loudness of each stem
+  augmentation_loudness_type: 1 # Type 1 or 2
+  augmentation_loudness_min: 0
+  augmentation_loudness_max: 0
+  q: 0.95
+  coarse_loss_clip: false
+  ema_momentum: 0.999
+  optimizer: adam
+  other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental
+  use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true
+inference:
+  batch_size: 1
+  dim_t: 1101
+  num_overlap: 4

config_mel_band_roformer_karaoke_gabox.yaml

@@ -0,0 +1,72 @@
+audio:
+  chunk_size: 485100
+  dim_f: 1024
+  dim_t: 256
+  hop_length: 441
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.000
+
+model:
+  dim: 384
+  depth: 6
+  stereo: true
+  num_stems: 1
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  num_bands: 60
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0
+  ff_dropout: 0
+  flash_attn: true
+  dim_freqs_in: 1025
+  sample_rate: 44100  # needed for mel filter bank from librosa
+  stft_n_fft: 2048
+  stft_hop_length: 441
+  stft_win_length: 2048
+  stft_normalized: false
+  mask_estimator_depth: 2
+  multi_stft_resolution_loss_weight: 1.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+    - 4096
+    - 2048
+    - 1024
+    - 512
+    - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: true
+
+training:
+  batch_size: 1
+  gradient_accumulation_steps: 1
+  grad_clip: 0
+  instruments:
+    - Vocals
+    - Instrumental
+  lr: 0.0005
+  patience: 2
+  reduce_factor: 0.95
+  target_instrument: Vocals
+  num_epochs: 1000
+  num_steps: 1000
+  augmentation: false # enable augmentations by audiomentations and pedalboard
+  augmentation_type:
+  use_mp3_compress: false # Deprecated
+  augmentation_mix: false # Mix several stems of the same type with some probability
+  augmentation_loudness: false # randomly change loudness of each stem
+  augmentation_loudness_type: 1 # Type 1 or 2
+  augmentation_loudness_min: 0
+  augmentation_loudness_max: 0
+  q: 0.95
+  coarse_loss_clip: false
+  ema_momentum: 0.999
+  optimizer: adamw
+  other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental
+  use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true
+
+inference:
+  batch_size: 1
+  dim_t: 1101
+  num_overlap: 8

config_melband_roformer_big_beta5e.yaml

@@ -0,0 +1,51 @@
+audio:
+  chunk_size: 485100
+  dim_f: 1024
+  dim_t: 801
+  hop_length: 441
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.000
+
+model:
+  dim: 384
+  depth: 6
+  stereo: true
+  num_stems: 1
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  num_bands: 60
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0
+  ff_dropout: 0
+  flash_attn: True
+  dim_freqs_in: 1025
+  sample_rate: 44100  # needed for mel filter bank from librosa
+  stft_n_fft: 2048
+  stft_hop_length: 441
+  stft_win_length: 2048
+  stft_normalized: False
+  mask_estimator_depth: 3
+  multi_stft_resolution_loss_weight: 1.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+  - 4096
+  - 2048
+  - 1024
+  - 512
+  - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: False
+
+training:
+  instruments:
+  - vocals
+  - other
+  target_instrument: vocals
+  use_amp: True
+
+inference:
+  batch_size: 1
+  dim_t: 801
+  num_overlap: 2

config_melbandroformer_big_beta4.yaml

@@ -0,0 +1,51 @@
+audio:
+  chunk_size: 485100
+  dim_f: 1024
+  dim_t: 1101
+  hop_length: 441
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.000
+
+model:
+  dim: 384
+  depth: 12
+  stereo: true
+  num_stems: 1
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  num_bands: 60
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0
+  ff_dropout: 0
+  flash_attn: True
+  dim_freqs_in: 1025
+  sample_rate: 44100  # needed for mel filter bank from librosa
+  stft_n_fft: 2048
+  stft_hop_length: 441
+  stft_win_length: 2048
+  stft_normalized: False
+  mask_estimator_depth: 3
+  multi_stft_resolution_loss_weight: 1.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+  - 4096
+  - 2048
+  - 1024
+  - 512
+  - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: False
+
+training:
+  instruments:
+  - vocals
+  - other
+  target_instrument: vocals
+  use_amp: True
+
+inference:
+  batch_size: 1
+  dim_t: 1101
+  num_overlap: 2