Early Media: Does My AI Agent Hear Ringback?
Early media is audio sent before your call is answered. What SIP 180 and 183 with SDP mean, what an AI voice agent hears, and when you actually need it.
Early media is audio sent before your call is answered. What SIP 180 and 183 with SDP mean, what an AI voice agent hears, and when you actually need it.
Zero of 43 Australian place names appear in CMUdict. How to control AI voice agent pronunciation in the lexicon, phoneme, text transform and respelling layers.
Check whether a voice platform logs your API keys or caller data. Four public tracker queries, a canary test call, a sink map, and how to read what you find.
Point your own SBC at a model provider native SIP endpoint with no CPaaS in the path: the certificate, DNS, transport and media gates to clear before cutover.
Deepgram Flux, ElevenLabs Scribe and Speechmatics compared on turn detection alone: what each engine emits, the tunable ranges, and what still ends a turn.
An AI receptionist at an Australian vet clinic may book and route, but telling an owner whether a pet is an emergency is a restricted act. Where the line sits.
What an AI phone agent may handle at an Australian accounting firm, where the TFN, ATO notice and tax advice boundary sits, and which rule makes it so.
Speech-to-speech or STT plus LLM plus TTS? Choose on replaceability: the blast-radius table, provider counts per seam and the published shutdown dates.
How to make an AI voice agent remember a caller between calls: the four memory layers, identity resolution edge cases, what to store, and APP 11.2 limits.
Size a voice agent worker pool with Erlang B rather than CPU: 240 calls an hour at 4.5 minutes is 18 erlangs, 28 concurrent slots and four workers at peak.