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

[AmzGrainRain]

PR types

Function optimization

PR changes

Others

Describe

Currently, G2PWOnnxConverter uses download_and_decompress() unconditionally to obtain the model path. This function checks for the presence of the downloaded .zip archive to decide whether to skip downloading.

However, some users often delete the .zip file after extraction to save disk space (since the archive is no longer needed once the model is extracted). In such cases, even though the actual ONNX model file (g2pW.onnx) exists and is fully usable, the system will re-download the entire .zip archive on next run — although it won’t re-extract if the model directory is intact.

Changes

• Introduce a helper function get_g2pw_model_path() that directly checks for the existence of the essential model file: g2pW.onnx.
• Only call download_and_decompress() if this file is missing.
• Preserve all existing behavior for first-time usage.

[paddle-bot]

Thanks for your contribution!

[CLAassistant]

All committers have signed the CLA.