The short answer

SIP is how modern voice calls are set up. It negotiates who is calling whom and which codecs both sides support, then gets out of the way while the audio flows over a separate protocol. Nearly every programmable-voice platform speaks SIP underneath its HTTP API, and SIP-level detail surfaces the moment a call fails.

In detail

SIP is a text-based request-response protocol deliberately modelled on HTTP, with methods including INVITE to start a session, ACK to confirm, BYE to end one, and REGISTER to bind a user to a network location. Responses use familiar numeric classes, so a SIP 404 means the number was not found and a SIP 486 means the far end was busy.

Crucially, SIP carries no audio. It negotiates the session using the Session Description Protocol in the message body — codecs, IP addresses, ports — and the media itself then flows over RTP directly between the endpoints or through a media server. This separation of signalling and media is why call setup and audio quality fail in completely different ways and are diagnosed with different tools.

For a voice AI platform, SIP is usually an implementation detail that becomes visible during failures. Response codes explain why a call did not connect, and codec negotiation determines what audio the recogniser actually receives. Most platforms therefore surface a mapped status — busy, no answer, failed — rather than raw SIP codes, while retaining the underlying detail for support.

Back to the glossary

All 53 glossary terms