How to produce the best results?

Hello, so I would like to know how I could create the possibly best results for correct subtitle outputs. So the Whisper app has lots of parameters we can use when we print out the help but the most of them I don't know what they are good for to adjust the results and quality. After using Whisper for a while and testing it I always stuck on some problems I would like to know how to prevent them by using any parameters. Problem 1: Hallucinations / Word Sentence repeating ----------------------------------------------------------------- In some cases till often it happens that words / sentences getting repeated couple times even nothing is spoken or something else is spoken in the audio. What can I do about it to reduce or prevent it? Problem 2: No text even something is spoken ----------------------------------------------------------------- Sometimes Whisper does output nothing even the speaker is telling something or the voice is too quiet. I'm using the MEDIUM model in normal case. I also tried other models like TURBO with ffmpeg in build Whisper what does recognize it better when audio is quiet but ffmpeg + Whisper has limited parameters we can't use 'Sentence' option I normally use. Is there a way to use parameters in Whisper to adjust the sensitivity to recognize also quiet spoken voices like ASMR style? Problem 3: Different output results ----------------------------------------------------------------- So when I try to change some parameter/s then it happens that output results are somehow different in text. Words or sentences can be else. My question would be how to make Whisper working to re-produce the spoken words 1:1 as in audio so accurate as possible? Problem 4: Censorship of words in some cases ----------------------------------------------------------------- I also see that problem that Whisper does censor words in some cases with stars (*) instead to print the word out. That's pretty annoying and I wanna disable it to get an 1:1 output even when some swear words are used in the audio like 'Shit' or 'Fuck' etc. I was testing parameters like ' --condition_on_previous_text=False' and True too what seems to reduce some hallucinations but does increase the word * censoring. But also in both cases using False or True with this parameter it can happen that some words getting marked with *** instead to print it out. How to disable that completely to censor anything in text? I even don't understand why anyone added a censorship for words when you can hear it in the audio. Makes no sense. So that's all questions I have and I hope that you guy's could give me some advice what Whisper command-line parameters I should use to prevent or reduce those issues getting an best possible result out. Thank you.