Inference time increases when processing multiple files

[Jinming00]

When processing a folder of audio files, the inference time per file gradually increases. The first few files are processed very quickly, but the processing time becomes progressively slower for subsequent files, eventually stabilizing at a much higher duration.

For example, my audio files have an average processing time of 2-3 seconds. However, at the beginning, the model takes only about 1 second to infer a single audio file. As it continues, the inference time per file slows down and eventually stabilizes at around 5 seconds per file.

Could you please provide guidance on what might be causing this issue?