Push-to-talk dictation on a Mac: hold, toggle or stream
Every dictation tool has to answer two questions: when does recording start and stop, and when does text appear? Push to talk means you hold a key while you speak. Toggle means you press once to start and again to stop. Streaming shows words while you're still talking. Each one changes how dictating a prompt feels, so here's how the common Mac tools behave and why we built FnFlow around holding a key.
Three ways to start and stop
- Hold to talk. Recording runs while the key is down. Release ends it. There's no ambiguity about where a dictation begins and ends.
- Toggle. Press once to start and once to stop. Your hands are free in between, but you have to remember to stop, and a silence timer often does it for you.
- Streaming. This is about when text appears, not how you start. Words show up while you speak and get revised as the model hears more. It can go with either hold or toggle.
How the tools you already have behave
Apple Dictation is a toggle. Apple's guide says to press your shortcut or choose Edit > Start Dictation, then press Escape or the shortcut to stop. Text appears as you speak, and dictation stops by itself after 30 seconds without speech.
Claude Code's /voice offers both. Hold mode is the default: hold Space, and recording starts after a short warmup while Claude Code watches for key repeat. Text appears dimmed until it's final, and on release it waits for you to press Enter. Tap mode starts and stops with a tap and sends the prompt automatically when the transcript has at least three words. Recording also stops after 15 seconds of silence or two minutes in total.
VS Code dictation works as a toggle, and its docs add that holding the shortcut gives you push to talk. A session stops at 20 minutes. Chat dictation fills in the input and doesn't send it.
Warp uses hold to talk by default on macOS: hold Fn, speak, release to transcribe.
Superwhisper defaults to a toggle on ⌥Space and has a separate push-to-talk shortcut, where a quick tap acts as a toggle. Its realtime mode, which shows text while you record, needs Pro and a model that supports it.
Why holding fits prompts
A prompt for an AI agent is a single unit. You think, say it and want it in the input box, untouched, until you decide to send. Holding a key matches that well.
- Nothing cuts you off. No silence timer ends the take while you scroll back through a diff to find a function name.
- No stray recordings. A toggle you forgot to stop keeps listening. A held key can't be left on.
- Clear end, clean text. When you release, the tool knows the take is complete and can transcribe it in one go, with punctuation that fits the whole thought.
- Nothing is sent for you. Auto-submit is handy for chat, but a prompt with one misheard file name shouldn't go out before you read it.
Where a toggle wins
For anything longer than a breath or two, holding a key gets tiring and ties up a hand you might want for scrolling. Long specs, review notes and planning sessions are easier hands-free. Most tools offer both for that reason, and so does FnFlow.
The cost of streaming
Seeing words appear is reassuring, and with a cloud model or a streaming engine it can feel instant. The cost is that early words are guesses. They get revised as the sentence continues, and a tool that commits them too soon can leave fragments or stray periods in fields that don't expect live edits, terminals especially. If you dictate into agents and terminals, getting the text once, finished, is usually worth a second of waiting.
What FnFlow does
FnFlow is hold-to-talk on fn, or right ⌘ if your 🌐 key already does something. The local Whisper server starts the moment you press the key, so the model loads while you're still talking. When you let go, the take is transcribed on your Mac and pasted once at the cursor. For a 7.6-second sentence on an M4 Pro with large-v3-turbo kept warm, that's about 1.2 seconds. FnFlow never presses Enter.
A few details came from watching real use. Taps shorter than 0.6 seconds are ignored, because in our first 1,295 recordings nearly every failure was a brushed fn key. Pressing another key while holding turns the press into a shortcut, so ⌘C with right ⌘ never records. For long dictation, double-tap to go hands-free and tap again to finish.
The trade-off is that FnFlow doesn't stream into the field. You see a recording pill with a level meter while you speak, not the words. If live text matters more to you than a single clean paste, a streaming tool will suit you better. Key setup is covered in the fn key and dictation, and terminal behaviour in dictation in Claude Code and the terminal.
Questions
What is push-to-talk dictation?
You hold a key while you speak and release it when you're done, like a walkie-talkie. Recording runs only while the key is down, so the start and end of each dictation are always clear.
Does Apple Dictation support push to talk?
Apple's guide describes a toggle: press the shortcut to start, then Escape or the shortcut again to stop. It also stops by itself after 30 seconds without speech.
Is streaming dictation faster?
It shows words sooner, but the final text often changes as the model hears more. With a model kept loaded, waiting for the whole take can still be quick. FnFlow takes about 1.2 seconds after release for a 7.6-second sentence on an M4 Pro.
How do I dictate something long without holding a key?
In FnFlow, double-tap fn to go hands-free and tap again to finish. Other tools have their own toggle or tap modes.
Local Wispr Flow alternativeFnFlow vs VoiceInkVoice input for CursorHow 7 Mac dictation apps get you to your first dictationWhisper.cpp on Apple Silicon: which large-v3-turbo quantization to shipHow I built a local dictation appVoice dictation in Claude Code and the terminalWhy holding fn doesn't work for dictation on Mac (and how to fix it)How to run Whisper locally on an Apple Silicon MacSpokenly alternative for developersFnFlow vs EnviousWisprHow to dictate a good prompt for AI coding toolsVoice dictation in VS Code on a MacTeaching Whisper your jargon: initial prompt and replacementsWhy Whisper types "Thank you" on silence, and how to filter itPrivate dictation on a Mac at work: what actually leaves your machineVoice input for Codex CLI and the ChatGPT appVoice coding on a Mac: what actually worksWriting git commit messages by voiceDictating code review comments on GitHubDictation in IntelliJ IDEA, PyCharm and WebStorm on a MacDictation in Zed on a MacDictation in Terminal, iTerm2 and WarpWhich Whisper model to use on an Apple Silicon MacSuperwhisper alternative for developersClaude Code voice mode: how to enable /voice and fix itWhisper apps: five ways to run Whisper locallyTalon voice coding vs a dictation appPush to talk on a MacDoes Claude have a voice mode?VoiceInk vs Wispr FlowVoiceInk vs SuperwhisperParakeet vs Whisper on a MacApple SpeechAnalyzer vs WhisperWhisper.cpp vs faster-whisper on a MacMlx-whisper vs whisper.cppGitHub Copilot voice inputGemini CLI voice inputVoice input in Google AntigravityVibe coding with voice input