New branch dcase2016

AudioLoader/Audio_event_detection.py

@@ -0,0 +1,158 @@
+
+# import requests
+# from tqdm import tqdm
+# import tarfile
+
+from torch.utils.data import Dataset
+import torchaudio
+from pathlib import Path
+from typing import Tuple, Union
+from torch import Tensor
+import torch
+import os
+import time
+import json
+import tqdm
+import shutil
+import glob
+import multiprocessing as mp
+import warnings
+from distutils.dir_util import copy_tree
+from torchaudio.compliance import kaldi # for downsampling
+from torchaudio.datasets.utils import (
+    download_url,
+    extract_archive,
+)
+import hashlib
+import json
+import subprocess
+
+
+class DCASE2016(Dataset):
+    """Dataset class for DCASE2016 dataset.
+    Args:
+        root (str or Path): Path to the directory where the dataset is found or downloaded.
+
+        split (str):
+            Choose different dataset splits such as ``"train"``, or ``"test"`` or ``"dev"``. (default: ``"train"``).
+
+        download (bool, optional):
+            Whether to download the dataset if it is not found at root path. (default: ``False``).
+
+    """
+
+
+    def __init__(self,
+                 root: Union[str, Path],
+                 split: str = "train",                 
+                 download: bool = False
+                ):
+        self.split = split
+        if self.split == "train" or self.split =="dev":
+            url = "https://archive.org/compress/dcase2016_task2_train_dev/formats=ZIP&file=/dcase2016_task2_train_dev.zip"
+        else:
+            url = "https://archive.org/download/dcase2016_task2_test_public/dcase2016_task2_test_public.zip"
+        folder_name = 'DCASE2016'
+
+        download_path = os.path.join(root, folder_name)
+        assert self.split.upper()=="TRAIN" or self.split.upper()=="TEST" or self.split.upper()=="DEV", f"split={split} is not present in this dataset"
+
+        if self.split =="train":
+            archive_name = f'dcase2016_task2_train_dev.zip'
+            self._path = os.path.join(download_path, f"dcase2016_task2_train_dev",f"dcase2016_task2_train") 
+        elif self.split =="dev":
+            archive_name = f'dcase2016_task2_train_dev.zip'
+            self._path = os.path.join(download_path, f"dcase2016_task2_train_dev",f"dcase2016_task2_dev","sound") 
+        elif self.split == "test":
+            archive_name = f'dcase2016_task2_test_public.zip'
+            self._path = os.path.join(download_path, f"dcase2016_task2_test_public", "sound") 
+
+        checksum_dict = {"test": "ac98768b39a08fc0c6c2ddd15a981dd7", "train":"0eab0635cff8e2d76e9e38b7a7de342d", "dev":"0eab0635cff8e2d76e9e38b7a7de342d"} #same value for train&dev
+
+        if download: #bad data format==> unzipped file has same name as zipped file, create tmp path and remove as a workaround
+            #file exists and extracted
+            if os.path.isfile(os.path.join(download_path,archive_name)) and os.path.exists(self._path):
+                print(f"Dataset archive exists, all files are extracted. Using all file from {self._path} ")
+            #file exists but not extracted
+            if os.path.isfile(os.path.join(download_path,archive_name)) and not os.path.exists(self._path):
+                print(f"Dataset archive exists, extracting archive:{os.path.join(download_path,archive_name)}")
+                if self.split!="test":
+                    tmp_path = os.path.join(download_path,"tmp")
+                    extract_archive(os.path.join(download_path, archive_name),to_path = tmp_path)
+                    extract_archive(os.path.join(tmp_path, archive_name),to_path = download_path)
+                    os.system(f"rm -rf {tmp_path}")
+                else:
+                    extract_archive(os.path.join(download_path, archive_name))
+                print(f"Using all file from {self._path} ")
+            #file not exist
+            elif not os.path.isfile(os.path.join(download_path,archive_name)):
+                print(f"archive {os.path.join(download_path,archive_name)} not exists, try downloading")
+                if not os.path.exists(download_path):
+                    os.makedirs(download_path)
+                try:
+                    download_url(url, download_path) #, hash_value = checksum_dict[self.split], hash_type = "md5"
+                    if self.split!="test":
+                        tmp_path = os.path.join(download_path,"tmp")
+                        extract_archive(os.path.join(download_path, archive_name),to_path = tmp_path)
+                        extract_archive(os.path.join(tmp_path, archive_name),to_path = download_path)
+                        os.system(f"rm -rf {tmp_path}")
+                    else:
+                        extract_archive(os.path.join(download_path, archive_name))
+                    print(f"All files are extracted. Using all file from {self._path} ")
+
+                except:
+                    raise Exception('Auto download fails. '+
+                                    'You may want to download it manually from:\n'+
+                                    url+ '\n' +
+                                    f'Then, put it inside {download_path}')   
+        else:   
+            #archive is downloaded and extracted
+            if os.path.isfile(os.path.join(download_path, archive_name)) and not os.path.exists(self._path):
+                print(f"Dataset archive exists, all files are extracted. Using all file from {self._path} ")
+            #archive is downloaded but not extracted
+            elif os.path.isfile(os.path.join(download_path, archive_name)) and not os.path.exists(self._path):
+                print(f'archive:{os.path.join(download_path, archive_name)} exists, extracting...')
+                if self.split!="test":
+                    tmp_path = os.path.join(download_path,"tmp")
+                    extract_archive(os.path.join(download_path, archive_name),to_path = tmp_path)
+                    extract_archive(os.path.join(tmp_path, archive_name),to_path = download_path)
+                    os.system(f"rm -rf {tmp_path}")
+                else:
+                    extract_archive(os.path.join(download_path, archive_name))
+                print(f"Using all file from {self._path} ")
+            else:
+                raise FileNotFoundError(f"Dataset not found at {self._path}, please specify the correct location or set `download=True`")
+
+        self._walker = glob.glob(f"{self._path}/*.wav") 
+        self.label_dict = {'clearthroat':0, 'cough':1, 'doorslam':2, 'drawer':3, 'keyboard':4, 'keys':5, 'knock':6, 'laughter':7, 'pageturn':8, 'phone':9, 'speech':10} #keys == keysDrop; convert keysDrop in trainset to keys
+
+
+    def __getitem__(self, n):
+        """Load the n-th sample from the dataset.
+        Args:
+            n (int): The index of the sample to be loaded
+        Returns:
+            tuple: ``(path, waveform, sample_rate, start_end_event)``
+        """
+        file_path = self._walker[n]
+        waveform, sample_rate = torchaudio.load(file_path)
+
+        if self.split!="train":  
+            label_path = os.path.dirname(os.path.dirname(file_path)) +"/annotation/"+ os.path.basename(file_path).split(".")[0]+".txt"
+            lst_of_events = [e[:-1].split("\t") for e in open(label_path,"r").readlines()] #read lines, remove "\n", split
+            lst_of_events_encoded = [(float(x[0]), float(x[1]), self.label_dict[x[2]]) for x in lst_of_events] #lists of (start, end, class)
+        else: 
+            label_name = os.path.basename(file_path).split(".")[0][:-3]
+            if label_name == "keysDrop":
+                label_name = "keys" #check readme.txt, and http://dcase.community/challenge2016/task-sound-event-detection-in-synthetic-audio; keysDrop and keys are equivalent
+            lst_of_events_encoded= [("dummy_start","dummy_end",label_name)]
+        batch = {'path':file_path,
+                 'waveform': waveform,
+                 'sample_rate': sample_rate,
+                 'start_end_event':lst_of_events_encoded
+                }
+
+        return batch
+
+    def __len__(self) -> int:
+        return len(self._walker)