Voice separation package spleeter reduces accuracy

Please do not waste your time to improve audio inputs to this project...it's models are trained to handle everything,...thinking that feeding a voice only input (by splitting voice & background)...I may improve accuracy of the generated subtitles files....by using another highly accurate AI project like [Spleeter](https://github.com/deezer/spleeter)...but found that it only decreases both detection & the quality.

Just select the best file available and do not try to change the source (by encoding or any other means) ...it will only hamper your output.