fix: loading script

.gitignore

@@ -4,3 +4,5 @@ debug_dataset.py
 test_loading_script.py
 asr_dataset-backup.py
 **/__pycache__
+-asr_dataset.py
+*.ipynb

asr_dataset.py

@@ -1,149 +0,0 @@
-import os
-import tarfile
-import pandas as pd
-import datasets
-from datasets import Audio, Value, Features
-import logging
-from typing import Dict, Generator, Tuple
-
-logger = logging.getLogger(__name__)
-
-
-_DESCRIPTION = """
-This dataset consists of various Youtube videos in Persian language.
-
-Note: This dataset contains raw, unvalidated transcriptions. Users are advised to:
-1. Perform their own quality assessment
-2. Create their own train/validation/test splits based on their specific needs
-3. Validate a subset of the data if needed for their use case
-"""
-
-_CITATION = """
-Use this repo info/link for citation.
-"""
-
-_LICENSE = "MIT"
-
-_DATA_URL = [
-    "clips/clips_001.tar",
-    "clips/clips_002.tar",
-    "clips/clips_003.tar",
-    "clips/clips_004.tar",
-    "clips/clips_005.tar",
-    "clips/clips_006.tar",
-    "clips/clips_007.tar",
-    "clips/clips_008.tar",
-    "clips/clips_009.tar",
-    "clips/clips_010.tar",
-    "clips/clips_011.tar",
-    "clips/clips_012.tar",
-    "clips/clips_013.tar",
-    "clips/clips_014.tar",
-    "clips/clips_015.tar",
-    "clips/clips_016.tar",
-    "clips/clips_017.tar",
-    "clips/clips_018.tar",
-    "clips/clips_019.tar",
-    "clips/clips_020.tar",
-    "clips/clips_021.tar",
-]
-#_DATA_URL = ["ytDataset/" + x for x in _DATA_URL][:2]
-
-_PROMPTS_URLS = {
-    #"train": "ytDataset/clips/metadata.csv"
-    "train": "clips/metadata.csv"
-}
-
-class ASRDataset(datasets.GeneratorBasedBuilder):
-    """ASR dataset with audio files stored in tar archives."""
-
-    VERSION = datasets.Version("1.0.0")
-
-    def _info(self):
-        return datasets.DatasetInfo(
-            description=_DESCRIPTION,
-            features=datasets.Features({
-                "file_name": datasets.Value("string"),
-                "audio": datasets.Audio(sampling_rate=16000),
-                "sentence": datasets.Value("string"),
-                #"tar_file": datasets.Value("string"),
-            }),
-            supervised_keys=None,
-            citation=_CITATION,
-        )
-
-    def _split_generators(self, dl_manager):
-        """Returns SplitGenerators with added error handling."""
-
-        prompts_paths = dl_manager.download_and_extract(_PROMPTS_URLS)
-        archive = dl_manager.download(_DATA_URL)
-        train_dir = "clips"
-
-        try:
-            return [
-                datasets.SplitGenerator(
-                    name=datasets.Split.TRAIN,
-                    gen_kwargs={
-                        #"split": "train",
-                        #"data_dir": self.config.data_dir,
-                        "prompts_path": prompts_paths["train"],
-                        "path_to_clips": train_dir,
-                        "audio_files": dl_manager.iter_archive(archive)
-                    },
-                ),
-            ]
-        except Exception as e:
-            logger.error(f"Error in _split_generators: {e}")
-            logger.error(traceback.format_exc())
-            raise
-
-    def _generate_examples(self, prompts_path, path_to_clips, audio_files):
-        """Yields examples as (key, example) tuples."""
-
-        sentence_map  = {}
-        with open(prompts_path, encoding="utf-8") as f:
-            for row in f:
-                data = row.strip().split("\t", 1)
-                file_name = data[0].strip()
-                sentence_map[audio_path] = sentence
-
-        id_ = 0
-
-        tar_files = [
-            "clips/clips_001.tar",
-            "clips/clips_002.tar",
-            "clips/clips_003.tar",
-            "clips/clips_004.tar",
-            "clips/clips_005.tar",
-            "clips/clips_006.tar",
-            "clips/clips_007.tar",
-            "clips/clips_008.tar",
-            "clips/clips_009.tar",
-            "clips/clips_010.tar",
-            "clips/clips_011.tar",
-            "clips/clips_012.tar",
-            "clips/clips_013.tar",
-            "clips/clips_014.tar",
-            "clips/clips_015.tar",
-            "clips/clips_016.tar",
-            "clips/clips_017.tar",
-            "clips/clips_018.tar",
-            "clips/clips_019.tar",
-            "clips/clips_020.tar",
-            "clips/clips_021.tar",
-        ]
-        for tar_file in tar_files:
-            with tarfile.open(tar_file, 'r') as tar:
-                for member in tar.getmembers():
-                    file_name = member.name #"/".join([path_to_clips, member.name])
-
-                    sentence = sentence_map.get(file_name, "")
-
-                    audio_file =tar.extractfile(member)
-                    audio = {"path": file_name, "bytes": audio_file.read()}
-                    yield id_, {
-                        "file_name": file_name,
-                        "audio": audio,
-                        "sentence": sentence
-                    }
-                    id_ += 1

audio_dataset.py

@@ -0,0 +1,113 @@
+import os
+import tarfile
+import datasets
+import pandas as pd
+from typing import Dict, List
+import io
+
+_DESCRIPTION = """
+This dataset consists of various Youtube videos in Persian language.
+
+Note: This dataset contains raw, unvalidated transcriptions. Users are advised to:
+1. Perform their own quality assessment
+2. Create their own train/validation/test splits based on their specific needs
+3. Validate a subset of the data if needed for their use case
+"""
+
+_CITATION = """
+Use this repo info/link for citation.
+"""
+
+_LICENSE = "CC0"
+
+class AudioTarDataset(datasets.GeneratorBasedBuilder):
+    """Audio dataset with files stored in tar archives."""
+    
+    VERSION = datasets.Version("1.0.0")
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            features=datasets.Features({
+                "audio": datasets.Audio(sampling_rate=16_000),  # Adjust sampling rate as needed
+                "text": datasets.Value("string"),
+                "file_name": datasets.Value("string"),
+            }),
+            supervised_keys=None,
+        )
+
+    def _split_generators(self, dl_manager):
+        """Returns SplitGenerators."""
+        # Paths to your tar files and metadata
+        tar_dir = "./clips"  # Update this
+        metadata_path = "./clips/metadata.csv"  # Update this
+        
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,  # Or adjust splits as needed
+                gen_kwargs={
+                    "tar_dir": tar_dir,
+                    "metadata_path": metadata_path,
+                },
+            ),
+        ]
+
+    def _read_tar_index(self, tar_paths: List[str]) -> Dict[str, str]:
+        """Create an index mapping audio filenames to their tar files."""
+        file_to_tar = {}
+        for tar_path in tar_paths:
+            with tarfile.open(tar_path, "r") as tar:
+                for member in tar.getmembers():
+                    if member.isfile():
+                        file_to_tar[os.path.basename(member.name)] = tar_path
+        return file_to_tar
+
+    def _generate_examples(self, tar_dir, metadata_path):
+        """Yields examples."""
+        # Load TSV metadata
+        metadata_df = pd.read_csv(metadata_path, sep=',', encoding='utf-8')
+        
+        # Get all tar files
+        tar_paths = [os.path.join(tar_dir, f) for f in os.listdir(tar_dir) if f.endswith('.tar')]
+        
+        # Create index of which audio file is in which tar
+        file_to_tar = self._read_tar_index(tar_paths)
+        
+        # Keep tar files open during iteration
+        tar_handles = {path: tarfile.open(path, "r") for path in tar_paths}
+        
+        try:
+            for idx, row in metadata_df.iterrows():
+                audio_filename = row['file_name']
+                tar_path = file_to_tar.get(audio_filename)
+                
+                if not tar_path:
+                    print(f"Warning: Could not find tar file for {audio_filename}")
+                    continue
+                
+                # Extract audio file from tar
+                tar = tar_handles[tar_path]
+                try:
+                    audio_member = tar.getmember(audio_filename)
+                except KeyError:
+                    print(f"Warning: Could not find {audio_filename} in tar file")
+                    continue
+                    
+                audio_file = tar.extractfile(audio_member)
+                
+                if audio_file is None:
+                    print(f"Warning: Could not extract {audio_filename}")
+                    continue
+                
+                # Read audio data into memory
+                audio_data = audio_file.read()
+                
+                yield idx, {
+                    "audio": {"path": audio_filename, "bytes": audio_data},
+                    "text": row['sentence'],
+                    "file_name": audio_filename,
+                }
+        
+        finally:
+            # Clean up tar handles
+            for tar in tar_handles.values():
+                tar.close()