I disagree that knowing the type of data being encrypted makes the technique flawed. Let's say you wanted to build a Speech-to-Text API that takes some encrypte

I disagree that knowing the type of data being encrypted makes the technique flawed. Let's say you wanted to build a Speech-to-Text API that takes some encrypted audio, transcribes it, and returns the transcriptions encrypted. Sure, you know that any files sent to you are going to be encrypted audio files, but you don't know what is being said in them which is the point of this technique.