For years, speech-to-text has been a software feature that I’m loathe to try. All previous implementations require me to learn a “language” just to work with the tool. Now, with the advent of learning language models, I’ve figured it’s just a matter of time before the predictive capabilities of Learning Language Models (LLMs) are applied to this domain and yield a much more natural interaction. In fact the promise of such a future is now as the text you’re reading began with me dictating and revising to Aqua Voice.

With my idea in mind, I began dictating the above paragraph to Aqua, and then I provided prompts like, “Revise my first sentence for clarity,” or “Replace the word ‘technology’ in my last sentence.” No longer was I constrained to get it perfect as I said it, rather Aqua was able to apply my natural language instructions just like a human scribe. Amazing!

Try it out for yourself at