I've had similar confusion to what you've said (same sources too). From what I've tested (iPhone 16 Pro), yes it's all on device. That may only be on more recent devices though. If you disable anything that can connect you to a network, it still functions as expected, at the same speed, and at the same quality (from what I can tell).
The Dictation (uppercase D) features happens on-device[1].
There are other voice-to-text features, such as Translation, which may use the cloud, but those aren't in the context of the information you provided.
1. https://www.apple.com/legal/privacy/data/en/ask-siri-dictati...
I've had similar confusion to what you've said (same sources too). From what I've tested (iPhone 16 Pro), yes it's all on device. That may only be on more recent devices though. If you disable anything that can connect you to a network, it still functions as expected, at the same speed, and at the same quality (from what I can tell).