What Is AI Dictation and How Does It Work?
AI dictation turns what you say into text you can use in a document, message, or other text field. Some tools stop after speech recognition. Others add a second step that removes filler words, restructures a sentence, or formats a draft. The useful question is what you want to see at your cursor when you finish speaking.
What happens between speaking and typing?
An automatic speech recognition model, or ASR model, converts microphone audio into words. The model may add punctuation and capitalization as part of that step. An optional language model can then edit the transcript according to instructions. A desktop tool can deliver the result to the app where your cursor is active.
| Step | What it does | What to check |
|---|---|---|
| Capture | Records your speech from the selected microphone | Input source, clipping, background sound |
| Recognize | Converts audio into a transcript | Names, numbers, unfamiliar terms |
| Rewrite | Optionally applies instructions to the transcript | Meaning, dates, commitments, tone |
| Insert | Delivers the text to the focused field | Correct app, field, and cursor position |
Product names overlap. A feature called voice typing may already handle punctuation, and a feature called AI dictation may offer plain transcription. Check whether there is a separate rewrite step and whether you can control it.
Try the same sentence in two modes
Say this into a scratch document.
Um, tell Priya the build is ready, and could she check ticket four oh seven before Thursday? I can send the logs if she needs them.
A plain transcript might read like this.
Um, tell Priya the build is ready, and could she check ticket 407 before Thursday? I can send the logs if she needs them.
A short-message rewrite could look like this.
Hi Priya, the build is ready. Could you check ticket 407 before Thursday? I can send the logs if you need them.
These are illustrative outputs. Compare what your tool actually produces. The rewrite is useful only if it preserves the ticket number, the timing, and the fact that sending logs is an offer. Changing “before Thursday” to “on Thursday” would make the result less useful, even if it read smoothly.
When should you use plain transcription?
Use plain transcription for ticket numbers, bug reports, quotations, or prompts with exact constraints. Review the words and punctuation before pasting or submitting. If a name or identifier is repeatedly wrong, type it yourself or follow the speech recognition troubleshooting guide.
Use a rewrite when you already know the substance and want a specific format. A voice memo can become a short email or a list of tasks. Give the rewrite a narrow instruction and review its output.
Turn this transcript into a short email. Keep names, numbers, dates, conditions, and commitments as spoken. Preserve uncertainty. If a detail is unclear, write [check] in its place.
Does AI dictation need the internet?
It depends on where each step runs. A local ASR model can recognize speech after its files are installed. An online rewrite service may still receive the transcript. A local rewrite model can run on your computer if you have already set it up. If you dictate into a cloud app, that app handles the inserted text under its own settings. Offline voice typing guide
Try both modes with Voconly
Voconly is a free, open-source voice typing app for Windows and macOS. It runs speech recognition locally and supports optional text processing with a local or online language model. Begin with plain transcription in a local editor, then set up a rewrite for one task you repeat often. The scenario guide explains the text processing settings. Voconly source and documentation
For your first real draft, pick something you can review at a glance, such as a three-sentence follow-up. Keep the microphone input, model, and sample sentence the same when comparing modes. The edit that saves you the most time is the one you can verify quickly.
Want to Try Offline Voice Typing?
Voconly is a free, open-source desktop voice input tool for Windows and macOS. It runs speech recognition locally, supports multiple ASR models, and can optionally process text with local or online LLMs.