Add clean LabelNoise logic

data/dataset/label_noise_bleed.py

@@ -3,6 +3,7 @@
 from data.dataset import BaseDataset
 import os
 from pathlib import Path
+import csv
 
 from aimless.utils import MDX_SOURCES as SOURCES
 
@@ -11,8 +12,8 @@
 
 # Run scripts/dataset_split_and_mix.py first!
 class LabelNoiseBleed(BaseDataset):
-    def __init__(self, root: str, split: str, **kwargs):
-        tracks, track_lengths = load_tracks(root, split)
+    def __init__(self, root: str, split: str, clean_csv_path: str = None, **kwargs):
+        tracks, track_lengths = load_tracks(root, split, clean_csv_path)
         super().__init__(
             **kwargs,
             tracks=tracks,
@@ -22,7 +23,7 @@ def __init__(self, root: str, split: str, **kwargs):
         )
 
 
-def load_tracks(root: str, split: str):
+def load_tracks(root: str, split: str, clean_csv_path: str):
     root = Path(os.path.expanduser(root))
     if split == "train":
         split_root = root / "train"
@@ -34,6 +35,12 @@ def load_tracks(root: str, split: str):
         raise ValueError("Invalid split: {}".format(split))
 
     tracks = sorted([x for x in split_root.iterdir() if x.is_dir()])
+    if clean_csv_path is not None:
+        with open(clean_csv_path, "r") as f:
+            reader = csv.reader(f)
+            clean_tracks = [x[0] for x in reader if x[1] == "Y"]
+        tracks = [x for x in tracks if x.name in clean_tracks]
+
     for x in tracks:
         assert torchaudio.info(str(x / "mixture.wav")).sample_rate == LabelNoiseBleed.sr
 

data/lightning/label_noise.csv

@@ -0,0 +1,204 @@
+ID,Clean,Notes
+0a589d65-50a3-4999-8f16-b5b6199bceee,,
+0d528a19-cb0f-4421-b250-444f9343e51c,,
+0e0d57cd-8662-4091-86d4-ed3e35d04ef6,,
+0f5fb60c-51d4-4618-871d-650c9e927b79,,
+1a32e987-cae8-458f-9c26-d9a6abf5348d,,
+1ade22ad-99bc-4954-b2b9-fb9e8fa06d41,Y,
+1afe1b3b-3e2e-48d3-b859-f50e222cbaf4,,
+1ee489e8-55ad-4d4a-8277-24051b85c02d,,
+1f98fe4d-26c7-460f-9f68-33964bc4d8d3,Y,
+1fc37390-1769-452d-9bea-19025be4c467,Y,
+2a3fd99d-c86f-4275-9321-ad7146364503,Y,
+2b4e8304-c92d-4347-b09e-cfb9b3e29bf2,,
+2c020edb-5947-4fa7-afea-ebc592cea683,,
+2dc237cd-1637-46f0-8f58-ca68dc6f6031,Y,
+2e5d996d-43f3-4359-b7c5-afebe9997556,,
+02ee37da-eea3-42b4-83bf-ab7f243afa13,,
+3a047d1a-f56d-4bf4-9910-f4f77206e53d,,
+3a6023c3-4cd8-46c0-a376-d353743648a9,,
+3c3b5fdb-f15e-4ba4-884a-b083ce2426c6,,
+3c557409-3a34-43c2-9159-5421bbad5ecb,,
+3e41f238-7c48-4a42-ba70-5ee39824a844,,
+3e656eec-84d4-4a45-b410-d3817d849f92,,
+3e7985e5-408f-4cf8-92b9-b9f62f738dd3,Y,
+3e389000-8fdc-4b63-b8b8-ab044273790d,,
+3f5233cb-57fa-4772-b389-f295a6f416ae,,
+4a896cde-57c6-4646-b610-1b0b654d0349,,
+4b9f86f4-23e4-458b-839e-8a63b584bea3,,
+4cbd6c36-87a2-4d50-86e3-52d39b98fad3,,
+5a6df5c7-a58a-479e-bdfb-c5946c221933,Y,"20s of harmonica, some drum bleed in vox"
+05b0ab77-2495-438d-8831-e3d81f96c16d,Y,"Bass sometimes doesn't sound like ""bass"""
+05e7af85-9721-4b42-952a-ccd34feb6033,,
+5f04798d-c7be-4b8a-90bd-1fcd9946e875,,
+6b168ae6-9d8a-4dc2-9d27-898e6871bf8b,,
+06bfc6e7-e5ac-4827-bf8e-ffaf1675872f,Y,Crash cymbal in other
+6c70d5e0-5972-444a-86f8-a558dbb92d92,,
+6cd44645-ed19-4ecc-a57c-58d400005b29,,
+6ce087b4-e571-4472-9be2-04b5340311c6,,
+6ceda40a-88bc-4e98-87c3-dd5c91725d41,,
+6e50565e-8179-4913-af54-a2a7f0dcab2f,,
+7a9f3169-7bdf-483c-9634-8e5b097d50df,,
+7ba734f0-547e-4142-a9b4-01a0c25b9d1f,,
+7bfa233c-24ed-4c7a-9096-11e3aa00c55d,,
+7dd515b0-e218-425d-b8bf-a75056237d6a,,
+7e74ce11-1603-440b-8794-b6e665b917e1,,
+7efcb55d-c0db-472f-b765-1739ad7536aa,,
+07fb2df2-91d6-458d-9230-9638b4edac08,,
+8a6c9c1f-4865-404f-a805-1949de36a33c,,
+8b83ba75-5e35-48c4-b42f-5419da2e6301,Y,
+8ce11544-9a6f-4f1e-ac2f-fc10343f15c8,,
+8e2d0c5c-6764-4d74-a740-391d7931ffd7,,
+8f36f17f-c033-4c1b-a793-80dce43d507b,,
+9ac2612b-e25f-4d27-8d43-b957e7e5a74b,Y,
+9c8a5c66-f6d8-4425-8671-6b7aa6a2663b,,
+9ce23a79-20eb-431a-80d2-eda3260ef503,,
+9eb8bc50-cffb-4c19-be0e-d27423e3e102,,
+9f581867-9f63-4ec4-8a43-e62c3c4230a6,,
+9fc0cff7-bb02-496f-9fa1-db67c52b1b4b,,
+13f233aa-a2e5-4683-8533-2f1e344b55b4,,
+014f3712-293b-42af-9f29-0ed1785be792,,
+16fb6c39-3834-4cfc-abf4-abb4a8d4646c,,
+22d265ef-ee2b-4aba-8d60-c3430295cd6d,Y,
+22ea41a3-1766-4a76-8071-380b27f1869a,,
+24f8a652-0168-4702-8725-42f9924e6729,,
+30cfc60a-5a57-4000-a05e-65006c8f6f74,,
+35a19148-49bf-451d-9a0e-5ab8e914c367,,
+36ee7fc6-604c-4a75-b4f0-0a9ceef3b9cb,Y,
+43ca388d-3e62-4df2-b72d-abf407a7aa5a,,
+045dcfd1-e960-4332-80cc-fdacc4a7c6a7,,
+046ab651-a333-46e1-9d27-ab14ee036c42,,
+46bc5393-7753-44ae-913b-bd5fa8f33e98,,
+49ed7cef-8ffb-4833-bc96-6df7b8ff5b43,,
+58efddfb-04b3-4951-858e-e7dbcfccfc21,,
+58fa04aa-426f-4e16-a112-29eb6e2f2d3e,,
+63b68795-0076-476b-a917-dec9e89bf91e,Y,
+72c6f013-11ea-4bf7-93b2-c6ef2c117718,,
+78ef22ce-472f-4f82-8656-16df73b9465f,Y,
+87a5da23-f17b-44da-accf-c04832f81a14,,
+88b545e5-4d06-4d55-a306-1bd3a2915ee5,,
+89c515c9-5e93-4cb4-9806-20432d2d074d,,
+89f2c781-5c67-4508-a2d6-236744b8c197,,
+93dda0e8-dd76-49e4-b08f-54a82387cdf6,,
+94fafb2a-9f4e-4a01-bee9-998008f95f41,,
+97b07e0e-274e-4212-a66b-44210a48724d,Y,Strange sfx in Vox @ 0:30
+125fc63d-9b69-4170-a46a-42c91bc28446,,
+152d4f5d-4093-4fa4-a4a4-8a9b3502d89d,,
+174a115f-3688-45dc-8c39-9d05f21758e1,,
+0177be35-64de-469b-908b-2d9edb49c053,,
+212bb137-fd01-465e-80f3-a890fb0ebcdd,,
+260c431b-72d4-4ac6-bae5-ee49ce5c0fe4,,
+312bec8d-1c61-43e0-924a-1fb87ddc3e41,Y,
+322f4d9d-b0c9-4ab3-9e30-544a25331ffd,,
+334ed0e5-1761-4ee9-9d39-a555eb9b64a0,,
+0358fd1e-244a-4422-9a42-29b5d68f6e4b,Y,No Bass track
+390b4fea-92be-4fe3-9576-86529f80b4ae,,
+664cc931-4e00-441e-9a78-e7a292515cea,,
+704f1de9-1d02-4c2b-af05-107a7700a51d,,
+747d5c98-665b-4470-a696-7a6cf6968ef1,,
+765d5131-afd9-4ad7-8786-8bef5705c1c2,Y,
+825f697d-5fcd-4429-ba33-23163c726ca7,,
+1921a83e-0373-4bf7-8dc9-6cc9401c9309,,
+2973adc9-6bf8-4422-9f7a-6fd0038eb565,Y,
+4380ad97-2620-4419-a011-ddfa29a87f54,,
+4999a0bf-a753-4e0e-85b1-690259dabf96,,
+6031b120-f6e2-4999-96ba-a1e31be68ea8,,
+06114cc2-e34d-4f8f-82d3-cf6981572f2f,,
+6681f493-c996-424a-9bdb-c671912ea9db,,
+7180bffa-dd48-49b9-bc02-f6e3f7f165b0,,
+8042b88a-6179-406b-9ec4-b45a4cdd4a71,,
+8804c154-6294-481a-ad63-bc61162cae2f,,
+28748b6e-6125-42f7-998d-2ad734e39b6c,,
+49478a32-483f-48d2-a594-d272b44bf587,Y,
+53808b95-cfe9-461d-a113-ffadf32817a1,,
+95378cf3-e939-42e0-b486-ebf2ca951664,,
+169628c5-266d-4e11-993b-440fa5fa2167,Y,
+378742ba-5ba8-44c2-9cfd-8a609decca57,,
+553048ce-7afd-4e0e-b4cb-4896620287a1,,
+731893c6-67b9-42f0-aea6-d1f70c2c9870,,
+737356b2-ce9c-448b-877b-e42b3ed94563,,
+763641c7-488f-4959-a554-fdbce9582644,,
+04204031-4f98-44ba-9c47-98c2f2e6b8fc,,
+04798708-6915-4dbc-842e-d394d545d4eb,,
+4857878a-e44b-4143-90e9-b65d0b704306,,
+5640831d-7853-4d06-8166-988e2844b652,,
+7524054e-dc67-47e0-8c26-ea1d4d70d2fb,Y,
+8427760a-b82e-4136-8f12-dfd53cad9bc9,,
+25789239-1075-43b9-bfc9-51dff4a29590,,
+a0b9a4e4-51f5-4c98-a090-0317fb891056,,
+a0eae9d2-d97f-401a-a495-1e1d1cb84a9c,Y,
+a1dcaeb2-f4e6-4818-b490-09f44d624afc,,
+a6bccb70-62b5-42aa-bfc9-3b0b886a2b2d,Y,
+a56d9450-3a26-485c-8ac3-24b6b54e2c1d,,
+a199697c-3cdc-47c7-a9c7-c1b07dd6c9dd,,
+aa600069-0a98-45e6-94fa-4000bfe46c25,,
+aaf0fc7b-d7f5-412c-b3b9-313a7c483666,,
+ad6c3742-e517-42eb-9dec-e74ed05387cc,,
+aefc1609-976b-423e-8516-f7d588d64ff7,,
+afca84b2-0277-4b1b-8696-5f14543f338c,Y,
+b8a79d39-346e-4258-a810-572b3b2c9ab1,,
+b8d6f3eb-f2d6-4342-af90-6d09f5257b6b,,
+b92cb1ca-baa9-4c74-b6dc-36389671ed76,,
+b207da3d-4baf-485a-98e1-657602479b3a,,
+b876b54b-6007-4d36-a6f4-efed8829d5fc,,
+bacbb01f-b877-4d62-8050-992f1d85543a,,
+baea951d-526a-49aa-8329-c8de676341fb,,
+bb45cf1a-4c58-4fe3-88ea-97ef27527507,,
+bbf40b5a-8ef9-4aec-a6c3-8b8706eb2ba0,,
+bc1f2967-f834-43bd-aadc-95afc897cfe7,,
+bc964128-da16-4e4c-af95-4d1211e78c70,,
+bd25c90e-d307-4cd9-adfb-46f5e323a81b,,
+bdcc429e-ed95-40d3-a1af-bad268d66b25,,
+bdd109ec-d5dd-4d91-92ad-66b679518026,,
+c6d73235-1dd5-4085-a3b3-50a3466c6168,,
+c8f42ad5-5a2f-4398-b9d9-207fd4fcc551,Y,
+c15ade79-43c0-4271-9f00-a121cefc92e5,Y,
+c70471f9-9c4a-41c9-b8f8-20ac38847a8e,Y,
+c228818e-eabe-434b-9d60-2fb84a6c5b2a,,
+c2330200-ad8e-4848-8c2b-b70612f4b80e,Y,
+c8752696-4ae5-4e47-b2ad-622496966fa9,,
+ca080447-fe99-4f6d-98c9-c69b68dacba3,,
+cc3e4991-6cce-40fe-a917-81a4fbb92ea6,,
+cc7f7675-d3c8-4a49-a2d7-a8959b694004,,
+cda46831-26d1-4dd8-ac50-6004d27d45dd,,
+cff5bcde-6a15-4c5b-b529-e1c528c46335,Y,Some slight vocal bleed in other
+d4df499c-e394-4753-b459-e167e6a58bad,,
+d4fe2408-c123-4739-93bb-22f558ae99d7,,
+d7d28204-a8ac-4c2b-bb3c-c941f4a00b85,Y,
+d8f0e410-5761-4d4a-9000-effe11089bbd,,
+d028d7c2-45c1-4846-b7df-4964238fd460,,
+d45bb3a6-eb80-44b3-b2ef-56cc9d5b4914,,
+d072debf-ea5b-4e8a-a447-cc1868cfc5fa,,
+d890ff35-300d-49f2-8054-49ab47262987,Y,
+d2401d3d-967c-46be-b9a0-3da571105158,,
+d624037a-1a76-4dd9-9e60-4ba380748a0b,,
+d4262245-3143-4c05-8423-6cbdc6253042,,
+dda2c057-6d73-43fa-a130-d7d6562c09ca,Y,
+dfb0e076-cb6b-4dcc-9934-c60070ff04d7,,
+e2ccbc17-44bf-431a-af2b-4cf2fbd19a72,,
+e2e4ce50-cd0d-4144-809d-4cf8c8e4912e,,
+e3ab3975-033b-40e2-b538-09396b3d4244,Y,
+e4de8632-6f69-4c63-8081-f4c2b77b40df,Y,
+e37cdb09-e648-4e9b-bc06-d178a964161c,Y,
+e62afdcd-0c96-4bee-80c7-1c17b897a6d7,,
+e78fa5de-cfdd-44fa-87c1-10a337b7011f,,
+e9336d31-c0df-4c91-be2b-7c4420c9cd34,Y,
+e1108928-9776-434c-bc57-c32dfdb7839c,,
+ea29ab4d-7f72-4331-b2a4-d3945c754211,,
+ea898682-08e7-4818-b516-8c0e10a4c20a,,
+ebea0f1d-8e23-469e-8eed-5269a9c684f0,,
+ed90a89a-bf22-444d-af3d-d9ac3896ebd2,,
+ee082817-dbda-4fbf-b5aa-8dce2320ae35,,
+ef1510e0-ba23-4b59-ba53-14181d73f213,,
+f0c565c5-fc73-4da1-b979-0fac0167f671,,
+f4b735de-14b1-4091-a9ba-c8b30c0740a7,Y,
+f9e58f4d-e361-4598-9c9a-d0a83529cc68,Y,
+f40ffd10-4e8b-41e6-bd8a-971929ca9138,,
+f76e2c13-9a9a-4cac-b6dd-45b5111aac6d,Y,
+fa46f72c-696d-45bc-bcc5-2b3305800565,,
+faad432d-6ad0-492d-96f1-321eeb9685b5,,
+fac94d9a-59da-4f83-9027-3eafe082ad16,Y,
+fcd1937f-2b21-4a78-889a-7b7e63e0ebdd,,
+fd6e4b4a-f33a-4f3c-aa6e-7c65fd5dc0bc,,
+fe3ae408-d35f-4c17-aa33-402238725a9d,Y,
+ff486935-7ce2-4e23-8908-0ff5fcc50856,Y,

data/lightning/label_noise.py

@@ -2,7 +2,7 @@
 from torch.utils.data import DataLoader
 import pytorch_lightning as pl
 from typing import List
-
+import os
 
 from data.dataset import LabelNoiseBleed
 from data.augment import CPUBase
@@ -35,6 +35,10 @@ def __init__(
             self.transforms = Compose(transforms)
 
     def setup(self, stage=None):
+        label_noise_path = None
+        current_dir = os.path.dirname(os.path.realpath(__file__))
+        if "label_noise.csv" in os.listdir(current_dir):
+            label_noise_path = os.path.join(current_dir, "label_noise.csv")
         if stage == "fit":
             self.train_dataset = LabelNoiseBleed(
                 root=self.hparams.root,
@@ -44,13 +48,15 @@ def setup(self, stage=None):
                 random=self.hparams.random,
                 random_track_mix=self.hparams.random_track_mix,
                 transform=self.transforms,
+                clean_csv_path=label_noise_path,
             )
 
         if stage == "validate" or stage == "fit":
             self.val_dataset = LabelNoiseBleed(
                 root=self.hparams.root,
                 split="valid",
                 seq_duration=self.hparams.seq_duration,
+                clean_csv_path=label_noise_path,
             )
 
     def train_dataloader(self):