Is speech to text generative AI?

No, not by default. Speech to text is AI, because the recognizer is a trained model. But a recognizer transcribes: it picks words that match the audio you gave it. Generative AI writes text that was not chosen from the audio. That second step is optional, and it sits on top of the recognizer.

A plain recognizer transcribes, so it is not generative

Microsoft describes Copilot dictation as transcribing your speech into text in real time ¹. A transcript is a record of what was said, in words the recognizer selected from the audio.

Transcripts are not always verbatim. A 2025 study of dementia speech reports that Whisper tends to omit filler words such as uh and um ². That study is a preprint with a side observation, so it shows a pattern, not a rate across all speakers. Dropping a filler is still a small edit. It is not the invention of new sentences, which is what most people mean by generative.

A language model rewrite is what makes it generative

A large language model can produce words that were not in the recognizer's output. In the HyPoradise benchmark, a language model can correct words that are missing from the recognizer's N-best list, the set of best candidate transcripts, because it generates text rather than choosing among candidates ³. Researchers call this generative error correction. Yang and colleagues study whether language models can act as speech recognition post-processors that perform rescoring and error correction ⁴.

Products now ship this step under a name. Microsoft says fluid dictation in Windows Voice Access automatically corrects grammar, punctuation and filler words as you speak ⁵, and that the experience is powered by on-device small language models ⁶. Apple says that with Apple Intelligence in iOS 27, dictation uses an on-device model to improve spelling, punctuation and capitalization, in English on the iPhone models it names ⁷.

The diagram shows the two steps. Only the second one generates.

Why the difference matters for the text you produce

A rewrite changes your wording, not only its spelling. A language model that fixes grammar has to decide what you meant, and it can get a name, a number or a negation wrong. Microsoft tells Copilot users to review the transcribed or generated content for accuracy, because AI may occasionally misinterpret speech or produce errors ⁸. That warning covers both steps.

This is why policy wording matters. If a school, employer or client restricts generative AI, the rewrite is the part most likely to fall under the rule. A plain transcript may not, although some policies define AI more broadly. The question on whether voice typing counts as AI covers that case.

How to tell which one you have

1. Look for a named rewrite setting. Fluid dictation, AI cleanup or a similar switch means a rewrite is available. Check each app, because one product's setting says nothing about another.
2. Read the vendor's verbs. "Transcribes" or "converts speech to text" describes the recognizer. "Corrects grammar" or "improves wording" describes a rewrite.
3. Run a test sentence. Dictate a sentence with "um" in it and a deliberate grammar error. If the output changes words you did not say, a rewrite is on.
4. Compare with the setting off. If the app can dictate with its cleanup switched off, compare the two outputs side by side.

Where to read more

The sibling questions cover the products one at a time: whether Microsoft dictation uses AI and whether Google voice typing uses AI. For the full picture, read the overview of whether speech to text is AI and the guide to AI cleanup for dictation. To keep the transcript plain, see speech to text without AI. The overview of voice and AI lists every guide on the subject.