Oh my mistake. I totally missed your point. Yeah an i5 covers a huge range. In my mind I was just assuming they meant the slowest i5, but yeah that doesn’t make sense
I interpreted the whole comparison as an expectation management for rpi4 users - A 2:1 length:processing time is well above what I would have guessed TTS would take, so highlighting this limitation is important.
A pi 4 has a very specific performance to power profile. An "i5" has no specificity.
An underclocked i5 could either be a whisper or an inferno compared to this ARM chip.
It also mentioned useful agents like GPT 3.5 as a sort of distant afterthought, which is both paid and highly non-local.