Speech-to-text

Speech-to-text is a category of software that takes an audio input — usually a live microphone stream or a recorded file — and outputs a written transcript of the words that were spoken. Most modern speech-to-text systems combine an acoustic model (which maps sound waves to phonemes) with a language model (which predicts the most likely words given the sounds and context).

You talk. Software listens. It writes down what you said. Some speech-to-text runs in the cloud on large servers. Some runs on your device. Some, like the browser Web Speech API that Zahvox uses today, is built into your browser and handled by the browser vendor.

Speech-to-text is the foundation of every voice-first writing workflow. Without it, you cannot dictate emails, capture voice notes, or build a voice-to-writing product. Its quality — how accurately it captures your words — sets the ceiling for everything downstream, including AI cleanup, formatting, and reuse.

Speech-to-text is not the same as voice-to-writing. Raw STT output is often a messy wall of words. Turning that into a usable email, note, or reply is a separate editing and formatting step — that's what Zahvox is built around.