fix(g2pw): avoid redundant download when model files already exist

paddlespeech/t2s/frontend/g2pw/onnx_api.py

@@ -41,6 +41,33 @@
 model_version = '1.1'
 
 
+def get_g2pw_model_path(model_dir: os.PathLike, model_version: str) -> str:
+    """Resolve the G2PW ONNX model directory path.
+
+    Checks if the model file 'g2pW.onnx' exists in the expected location.
+    If not, downloads and decompresses the model archive
+
+    Args:
+        model_dir (os.PathLike): Base directory to store models (e.g., ~/.paddlespeech).
+        model_version (str): Model version string (e.g., '1.1').
+
+    Returns:
+        str: Path to the model directory containing 'g2pW.onnx'.
+    """
+
+    archive_info = g2pw_onnx_models['G2PWModel'][model_version]
+    archive_fname = os.path.basename(
+        archive_info['url'])  # e.g., "G2PWModel_1.1.zip"
+    expected_extract_name = os.path.splitext(archive_fname)[
+        0]  # e.g., "G2PWModel_1.1"
+    expected_model_dir = os.path.join(model_dir, expected_extract_name)
+    uncompress_path = expected_model_dir
+    onnx_file_path = os.path.join(expected_model_dir, 'g2pW.onnx')
+    if not os.path.isfile(onnx_file_path):
+        uncompress_path = download_and_decompress(archive_info, model_dir)
+    return uncompress_path
+
+
 def predict(session, onnx_input: Dict[str, Any],
             labels: List[str]) -> Tuple[List[str], List[float]]:
     all_preds = []
@@ -70,8 +97,7 @@ def __init__(self,
                  style: str='bopomofo',
                  model_source: str=None,
                  enable_non_tradional_chinese: bool=False):
-        uncompress_path = download_and_decompress(
-            g2pw_onnx_models['G2PWModel'][model_version], model_dir)
+        uncompress_path = get_g2pw_model_path(model_dir, model_version)
 
         sess_options = onnxruntime.SessionOptions()
         sess_options.graph_optimization_level = onnxruntime.GraphOptimizationLevel.ORT_ENABLE_ALL