Psst / Voice coding
Voice coding on a Mac: the setup, including the room
Most voice coding guides compare dictation apps. The thing that ruins a take more often is the room: music from the speakers, a video in a tab, a notification.
Short answer: Pick a push-to-talk tool, a microphone close to your mouth, and make sure nothing else plays while you talk. Psst does the last part automatically.
1. Push-to-talk beats always-on
Hold a key, talk, release. Claude Code's /voice works this way by default with the space bar, and most dictation apps offer a hotkey. You decide when the microphone listens, and short takes are easy to redo.
2. A microphone close to your mouth
A laptop's built-in microphone sits a few centimetres from its speakers. A headset or a desk microphone pointed at you hears more of you and less of the room. With headphones, little of your music reaches the microphone at all.
3. A tool that knows code
Claude Code's dictation is tuned for development terms and adds your project and branch names as hints. Dictation apps such as Wispr Flow or Superwhisper work in every app, editors included. Both are fine; use what fits your flow.
4. Quiet while you talk
Background audio does two things: the microphone picks it up, and it pulls your attention while you are thinking out loud. Pausing by hand every time breaks the flow that made talking faster than typing. Psst mutes every other app the moment any of these tools opens the microphone, and brings the sound back when you stop.
A setup that works
- Claude Code in your terminal or VS Code,
/voiceon, hold space. - Headset or a microphone close to you.
- Psst in the menu bar, “Keep calls audible” on, so a huddle never gets muted.
Questions
Do I need a special microphone for voice coding?
No. Any microphone works; one close to your mouth picks up less of the room than a laptop's built-in microphone.
Does background music hurt transcription?
It can. Music and speech from speakers reach the microphone too. Anthropic's Claude Code docs suggest reducing background noise when no speech is detected.