Upload common_voice_sq_local.py with huggingface_hub

common_voice_sq_local.py

@@ -0,0 +1,67 @@
+import os
+import pandas as pd
+import tarfile
+from datasets import Dataset, DatasetDict, Audio, Features, Value, Sequence
+
+def load_common_voice_dataset(data_dir):
+    # Extract tar if not already extracted
+    tar_path = os.path.join(data_dir, "common_voice_sq_20_local.tar")
+    extract_dir = os.path.join(data_dir, "extracted")
+    if not os.path.exists(extract_dir):
+        os.makedirs(extract_dir)
+        with tarfile.open(tar_path, 'r') as tar_ref:
+            tar_ref.extractall(extract_dir)
+    
+    splits = {}
+    for split in ["train", "dev", "validated"]:
+        tsv_path = os.path.join(extract_dir, f"{split}.tsv")
+        if not os.path.exists(tsv_path):
+            continue
+        
+        df = pd.read_csv(tsv_path, sep="	")
+        # Use 'path' column from TSV, prepending 'clips/' to construct full audio paths
+        audio_paths = [os.path.join(extract_dir, "clips", p) for p in df["path"]]
+        
+        data = {
+            "audio": audio_paths,
+            "sentence": df["sentence"].tolist(),
+            "client_id": df["client_id"].tolist(),
+            "sentence_id": df["sentence_id"].tolist(),
+            "sentence_domain": df["sentence_domain"].tolist(),
+            "up_votes": df["up_votes"].tolist(),
+            "down_votes": df["down_votes"].tolist(),
+            "age": df["age"].tolist(),
+            "gender": df["gender"].tolist(),
+            "accents": df["accents"].tolist(),
+            "variant": df["variant"].tolist(),
+            "locale": df["locale"].tolist(),
+            "segment": df["segment"].tolist()
+        }
+        
+        features = Features({
+            "audio": Audio(sampling_rate=16000),
+            "sentence": Value("string"),
+            "client_id": Value("string"),
+            "sentence_id": Value("string"),
+            "sentence_domain": Value("string"),
+            "up_votes": Value("int32"),
+            "down_votes": Value("int32"),
+            "age": Value("string"),
+            "gender": Value("string"),
+            "accents": Value("string"),
+            "variant": Value("string"),
+            "locale": Value("string"),
+            "segment": Value("string")
+        })
+        
+        splits[split] = Dataset.from_dict(data, features=features)
+    
+    return DatasetDict(splits)
+
+def get_dataset(data_dir):
+    return load_common_voice_dataset(data_dir)
+
+if __name__ == "__main__":
+    import sys
+    dataset = get_dataset(sys.argv[1])
+    print(dataset)