The short answer

DTMF is how a caller sends data to a phone system by pressing keys. For voice agents it matters in two directions: navigating another party’s IVR when transferring or calling out, and collecting digits where speech is unreliable or unsafe — card numbers, PINs, long reference codes that no recogniser should be asked to guess at.

In detail

Each key maps to one tone from a low-frequency group and one from a high-frequency group, and the pair is decoded at the far end. Using two tones rather than one makes accidental triggering by speech or noise very unlikely, which is why the scheme has survived essentially unchanged since it replaced pulse dialling.

In IP telephony, DTMF is usually not sent as audio. RFC 2833 and its successor carry key presses as named RTP events out of band, because low-bitrate codecs distort the tones enough to break in-band detection. A mismatch in DTMF transport between two systems is a classic interoperability failure: the call audio is fine and the digits simply never arrive.

DTMF remains the right tool for a specific job even in a voice-first system. Spoken digit strings are error-prone over narrowband audio, and reading a card number aloud is a bad idea when the call is being recorded. Collecting sensitive digits by keypad — with recording suppressed for that segment — is both more accurate and easier to defend than transcribing them.

Back to the glossary

All 53 glossary terms