use case
verified July 2026
TTS for debt-collection voice agents.
Collections is outbound with a script attached by law: required identification and disclosure language on every connected attempt, at volumes where the dialer is the business. A character meter bills the disclosure every time it is spoken; a line does not care how often the law makes you say it.
01
Hear the workload
Both sides of this exchange are synthesized — rendered on this API with stock voices, published with its date. This is the tone a recovery call has to hold: firm on the terms, human on the delivery.
Audio — A collections exchange — both voices are this engine
rendered 2026-07-16
sam
“I know I'm behind on the account. I got paid Friday, but I can't do the full three hundred right now.”
agent
“Then let's not pretend you can. Split it — one fifty today, one fifty on the twenty-sixth, and the account comes current with no late mark.”
sam
“You can do that? The letters made it sound like all or nothing.”
agent
“It's already set. I've texted the plan and today's receipt — pay the second half from that link once Friday's check clears.”
both voices synthesized — leo and dane, stock
02
The cost shape
Every connected attempt opens with the same required language before the conversation proper begins — scripted characters a meter bills on every single call. On a flat line the words the law requires are not a cost input, and neither is the volume the business model requires.
Table 1 — One collections line, one month
| Connects / day | Minutes / mo | On a meter | Gandr flat |
|---|---|---|---|
| 100 | 3,000 | $90 – $300 | $150 |
| 300 | 9,000 | $270 – $900 | $150 |
| 500 | 15,000 | $450 – $1,500 | $150 |
03
What collections demands of synthesis
- A distinct cloned voice per client portfolio, from a ten-second reference — no training job per client, one key per portfolio for a clean audit trail at GET /v1/usage.
- Barge-in as one message on the socket: debtors interrupt, and the agent has to stop mid-word and deal with it.
- Refusal, not degradation, past capacity: a retryable 503 the dialer can reschedule around, never a degraded live call.
- Provenance built in: every sample carries an inaudible watermark applied at generation, so agency audio is identifiable as synthetic before a regulator asks.
See also
Related sheets.
use case
10,800 min
TTS for outbound sales agents
Outbound dialers talk six hours a day per line. That is 10,800 minutes a month — $324 to $1,080 on a character meter, $150 on a flat line.
capability
per key
Usage accounting that is not a bill
GET /v1/usage returns synthesized traffic per API key. Why usage data here is operational accounting rather than the source of your monthly invoice.
capability
never
Provenance and privacy, applied at generation
The policy surface of the engine: an inaudible watermark on every sample, reference audio that never trains, and per-key audit trails at GET /v1/usage.
glossary
100%
Audio watermarking
Audio watermarking marks synthesized speech as synthetic. How an inaudible watermark survives where metadata does not, and what a watermark cannot do.
Thirty minutes, your production script, the live latency readout — measured in front of you.
Bring this use case to the call