I can appreciate the effort and think it is an amazing idea. However, why build that into Signal which is E2EE and use an API for the speech-to-text?

It would be much more logical to do it with Messenger (for example by injecting JS with Tampermonkey) or to use a local model to perform S2T.