If you say an offensive word (as deemed by Google's speech recognition service) using speech input, such as bitch, fuck, shit, ass, etc., it is always censored as "####". There needs to be a configuration option to disable this censor, such as the option present in Android.
This is not a webkit issue, it is the embedder that captures audio, does recognition and returns the recognized results to webkit. WebKit does not censor any results and such an option to not censor words should be part of the embedder. If you are looking at a particular embedder please file a bug in that project.