Torchaudio.functional.vad()

Hi everyone, I’m trying to implement torchaudio.functional.vad() but I’m having problems with it. First it takes to much time to work. Is there any method how to filter all the data and then save them? I mean to filter the data once and every time I want to train my modell just to use the filtered one. Here stays: The effect can trim only from the front of the audio, so in order to trim from the back, the reverse effect must also be used.
I’m not understanding how to trim from the back. Can someone please help me?