TalkTastic doesn't generate text in real-time - that's not how our model works with its current configuration. While it's an interesting idea, we've actually debated and tested this extensively. What we found is that real-time transcription produces too many errors, which becomes really distracting unless the accuracy is extremely high. When you're seeing lots of transcription errors while speaking, it disrupts your flow. We originally had this feature but ultimately decided to disable it based on our testing.