Interactive voice response (IVR)

Interactive voice response (IVR) is a phone system that answers calls automatically, plays recorded or synthesized prompts, and collects the caller’s input from keypad presses or speech to serve the request or route the call.

How it works. An IVR combines a few standard pieces:

• Keypad input. Pressing a key sends a dual-tone multi-frequency (DTMF) signal. On internet telephony, RFC 4733 defines how DTMF digits and other telephony events travel in RTP packets. It replaced RFC 2833.
• Speech input and output. Speech recognition turns the caller’s words into text, and text-to-speech produces the prompts that are not prerecorded. MRCPv2 (RFC 6787) lets an IVR platform control speech recognizers, synthesizers, speaker verifiers and recorders on separate servers, with SIP setting up the sessions.
• Dialog logic. The W3C’s VoiceXML 2.0, a Recommendation since March 2004, is a standard language for writing IVR dialogs. It covers synthesized speech, digitized audio, recognition of spoken and DTMF input, recording, telephony and mixed-initiative conversation. In its model, a document server supplies VoiceXML documents, an interpreter runs the dialog, and an implementation platform handles the audio and the call. SSML controls how synthesized prompts sound.

Typical uses. Menus that route to the right queue, balance and order-status lookups, payments by keypad, appointment confirmations and after-hours messages.

Where AI changes it. A conversational AI voice agent can take the place of the menu, so callers describe their problem in their own words instead of pressing through options. The call still arrives over the same telephony stack.

When the caller is an agent. An AI agent calling an IVR has to listen to prompts, press keys or speak, and wait, turning structured data into sound and back. A published agent endpoint that accepts structured requests removes that round trip. When two voice agents meet on a phone call covers this in detail.

Neighbouring terms. Containment rate measures how many callers an IVR serves without a person. SIP carries the call itself.