use case
verified July 2026
TTS for AI customer-support agents.
A support agent lives or dies on turn-taking: answer inside the caller’s patience or get talked over. The synthesis leg is the one part of that budget a vendor choice can buy down — and the one part of the bill that should not scale with ticket volume.
01
The latency budget
Humans hand a conversation back and forth in about 200 milliseconds, and callers start interrupting an agent that takes much past a second. A working budget for one turn, drawn to scale:
Fig. — One turn of a support call — working budget per stage
Speech-to-text final result
150 – 300 ms
LLM first token
200 – 400 ms
TTS first audio
100 – 300 ms
▲ 107 ms measured here
Network + turn detection
100 – 200 ms
callers interrupt past here
The whole turn: 550 – 1,200 ms. Synthesis at the floor of its row hands the headroom to your LLM — the slowest stage you cannot buy down.
02
The cost shape
A desk doing 400 four-minute calls a day speaks about 48,000 minutes a month — call it six lines busy at typical concurrency, or roughly 8,000 minutes per line. Each of those lines is $240 to $800 a month on published meters, and $150 flat here. The deeper problem with a meter on a support desk is behavioral: the moment characters cost money, someone asks the agent to say less, and the agent gets worse.
03
What the desk demands of synthesis
- Stop mid-sentence and resynthesize when the caller barges in — persistent WebSocket, one socket per call.
- Speak the brand’s voice on every ticket: instant cloning from a ten-second reference, no training job.
- Survive the vendor’s bad night: standby failover lines beside your current provider, activating automatically.
04
Notes — an engineer's checklist
01How do I size lines for a support desk?
Count busy-hour simultaneous conversations, not monthly minutes. A desk stacking six concurrent calls at peak needs six lines; overflow past them spills to burst lines at $10 a line-day rather than ringing busy.
02What happens to the agent voice during an interruption?
Barge-in is one message on the open socket: playback stops mid-word, the answer is resynthesized inside the same turn budget, and the interrupted characters cost nothing because nothing is metered.
03Can the desk and the after-hours line share capacity?
Yes — a line prices concurrency, not workload. The same line that answers tickets at 2 PM can speak the after-hours prompts at midnight at zero marginal cost.
See also
Related sheets.
use case
43,200 min
TTS for AI receptionists
A 24/7 receptionist line speaks up to 43,200 minutes a month: $1,296 to $4,320 on a character meter, $150 flat. The arithmetic for always-on voice.
use case
20 ports
Replacing IVR menus with voice agents
Tearing out press-1 menus for a conversational agent changes the latency and concurrency math. What the synthesis layer must do, with measured figures.
glossary
×2
Barge-in
Barge-in is the caller interrupting the agent mid-sentence. What the TTS layer must do to support it: stop instantly, resynthesize fast, stay cheap.
glossary
107 / 108
Time to first audio (TTFA)
TTFA is the delay from sending text to the first playable audio byte. How it differs from model latency and TTFB, and how to measure it honestly.
Thirty minutes, your production script, the live latency readout — measured in front of you.
Bring this use case to the call