Generally speaking, it does make you wonder if the method used for transcribing some records (and I'm not saying this is how it's done in these cases) is by recording people reading text, and then using speech-to-text software to 'transcribe' it. Some examples sound as if it's the syllable sounds that are transcribed possibly? It may be a really effective method a lot of the time but come with the errors that we see at times? Just a thought, that's all.