hallicination for full silent audio input

[yijiegao2006]

If audio input is full silent, it will output/ return things like "MBC 뉴스 이재경입니다.".

Seems VAD has been applied in SYSTRAN/faster-whisper since for sample with long silence but verbal at the end it could return correct text.

[thiswillbeyourgithub]

This is inherent to all whisper based models.

Also related to #108 as cleaningup audio will return a very short result if it's silence so it wouls be ignored as under the duration threshold