Voice dictation glossary
This glossary defines the terms you will meet when comparing dictation apps, in plain English. Knowing a handful of them, ASR, WER, on-device versus cloud, and latency, makes it much easier to read reviews and pick the right tool. Each definition is short and practical, not academic.
Core terms
ASR (Automatic Speech Recognition). The technology that turns spoken audio into text. Every dictation app is built on some ASR system, whether its own model or an open one.
Speech-to-text (STT). Another name for the same thing: converting speech into written text.
Dictation. Live speech-to-text as you talk, typed into an app. Different from transcription, which usually means converting a recorded file after the fact.
Transcription. Turning recorded audio or video into text, after the fact. Tools like MacWhisper do this; dictation apps like Wispr Flow do live typing.
Accuracy and quality
WER (Word Error Rate). A common accuracy measure: the percentage of words a system gets wrong. Lower is better. Benchmarks vary hugely by audio quality, accent, and vocabulary, so treat any single WER number with caution.
Diarization. Identifying who spoke which words (speaker separation). It matters for meeting transcription, less so for personal dictation.
VAD (Voice Activity Detection). Detecting when you are actually speaking versus silence or noise, so the app knows when to transcribe.
Where processing happens
On-device (local). The audio is converted to text on your own machine, so your voice never leaves it. Best for privacy; see cloud vs on-device dictation.
Cloud. The audio is sent to a server for processing. This is how most polished apps work by default, including Wispr Flow. It enables strong accuracy without local hardware.
BYOK (bring your own key). Some tools let you plug in your own API key for a cloud model, so you control and pay for the processing directly.
Experience
Latency. The delay between speaking and seeing text. Lower latency feels more responsive.
Custom dictionary. A personal list of words (names, jargon) you teach the app to spell correctly. See the custom dictionary guide.
Command Mode. Editing or formatting text with spoken commands rather than dictating new words. See Command Mode.
Frequently asked questions
Automatic Speech Recognition: the technology that converts spoken audio into text. It is the engine inside every dictation app.
Lower is better, but there is no single "good" number: WER depends heavily on audio quality, accent, and vocabulary, so compare benchmarks only within the same test conditions.
On-device processes your voice locally so it never leaves your machine (better for privacy); cloud sends audio to a server (often better accuracy and features). See our post on the trade-off.
Related: voice typing explained · cloud vs on-device · accuracy