What is Push To Talk?
Also called PTT, hold to speak, tap to talk.
Push to talk is a voice input mode where the user holds or taps a control to mark exactly when they are speaking. The microphone captures audio only within that window, so the system never has to infer where an utterance begins or ends. It is the deterministic alternative to wake words and automatic endpointing.
The pattern comes from two-way radio, where a physical key opened the transmitter. In software it appears as a held button, a keyboard shortcut, or a tap-to-start and tap-to-stop toggle. Because the boundaries are supplied by the user, the system skips voice activity detection and end-of-turn prediction entirely, removing two of the most common sources of awkward interruption in voice interfaces.
Push to talk is favored in noisy environments, shared offices, and any setting where continuous listening would be socially or legally awkward. It also gives a clear privacy story: the microphone is demonstrably active only while the control is engaged, and interfaces usually reinforce this with a visible indicator and a start tone.
The cost is friction and a class of user errors. People release the control early and truncate their last words, or hold it and forget to speak. Interfaces mitigate this with a short trailing capture window after release, a live waveform or transcript so the speaker can see they are being heard, and a minimum-duration check that discards accidental taps.
Many voice agents offer several modes side by side. A hands-free mode uses continuous listening with endpointing, a manual mode uses push to talk, and the interface remembers the choice per device. Accessibility guidance suggests offering both, since neither continuous listening nor sustained button holding suits every user.
Key points
- User-supplied boundaries replace automatic endpointing
- Microphone is active only while the control is engaged
- Well suited to noisy or shared environments
- Trailing capture window prevents clipped final words
- Often offered alongside a hands-free listening mode
In practice
A field technician uses a voice agent inside a mobile app while standing beside a running compressor. Continuous listening picked up machine noise and cut utterances short, so the app switched to a hold-to-speak button. The technician holds it, dictates a fault code and reading, and releases. The app keeps recording for a fraction of a second past release, transcribes the clip, and reads back the parsed values for confirmation.