capability
verified July 2026
A text-to-speech API built for live calls.
One base URL, three surfaces, one price. This page is the API at a glance — the shapes you send, the figures we measure, and the bill that does not move.
Fig. 0 — voice specimens · pre-rendered on the production API
generated 2026-07-16
Leo
English
“I've placed a temporary hold on the card ending four two one nine after two charges that don't fit your pattern. If those were you, I can lift it right now. If not, I'll have a new card on its way before we hang up.”
financial services · fraud hold
11.3 s
Jenny
English
“You're all set for Thursday at ten with Doctor Okafor. I've texted the address and the two intake forms — reply here anytime and I'll move the appointment if something comes up.”
healthcare · appointment line
10.3 s
Mia
English
“Good news — your refund went through this morning, and the confirmation number is already in your inbox. It usually settles back to your card within two business days. Anything else I can take care of?”
retail · returns desk
10.4 s
Dane
English
“Your balance is about thirty days past due, and I'd much rather work it out with you than let it escalate. I can split it into two payments — would the first one landing this Friday work for you?”
collections · payment plan
9.4 s
Lewis
English
“I've opened the claim for the water damage and assigned it to an adjuster — they'll reach out within one business day. Hold on to the photos you took; they'll move everything along faster.”
insurance · first notice of loss
9.4 s
Ava
English
“Good morning, and thank you for calling the front desk. I can check you in, arrange a late checkout, or book you a table for tonight — what would help you most right now?”
hospitality · front desk
8.7 s
Dane
English
“Driver twelve is eight minutes out with two pallets, and dock four is clear. I've dropped the manifest in your inbox so you can check it before they pull in.”
logistics · dispatch
7.0 s
Leo
English
“Your technician is booked for Tuesday between nine and eleven. I'll send a text when they're about thirty minutes out, so you're not stuck waiting around all morning.”
field service · scheduling
7.7 s
Dane
English
“Ondansetron twice daily; the subpoena arrived Thursday; the charcuterie is from Worcestershire.”
hard words, first try
5.6 s
Stock voices, ready-made on the API — six of them, all multilingual. Cloning from ten seconds of your own voice runs on the call.
01
The three surfaces
- POST /v1/tts/bytes — one request, one complete WAV. Prompts, reminders, anything that is not live.
- POST /v1/tts/sse — base64 PCM chunks streaming over one HTTP response. Serverless-friendly.
- wss://tts.gandr.ai/ws — one socket per call, an utterance per turn, binary PCM back with ttfa_ms and audio_ms metadata.
curl -X POST https://tts.gandr.ai/v1/tts/bytes \
-H "authorization: Bearer gnd_yourkey" \
-H "content-type: application/json" \
-d '{
"transcript": "Your claim is open as of right now.",
"voice": {"mode": "clone",
"wav_b64": "<ten seconds of one speaker>"},
"language": "en",
"output_format": {"sample_rate": 24000}
}' --output turn.wav
# transcript · voice · output format — the voice rides in the
# request; there is no voice-management API to learn first
02
The figures and the bill
First audio is 107 ms p50 on a single stream — server time on the production API, methodology published, load behavior on its own benchmark page. The bill: a line is one concurrent call, $150 a month annual, $180 month-to-month, characters and minutes unmetered. Past about 5,000 spoken minutes a month a line beats every published meter; below that, we say so and point you at the meter.
03
Notes — an engineer's checklist
01What latency should I expect from the API?
107 ms to first audio at the 50th percentile, 108 ms at the 95th — server time on the production endpoint under call-shaped load, not model-only time. Under 16 concurrent callers on one GPU: 430 ms median, 813 ms p95, zero failed requests. The methodology is published at /benchmarks/tts-latency/ and reruns live on any demo call.
02How does authentication work?
One key on every request — x-api-key: gnd_… or Authorization: Bearer gnd_…, either header works. Keys are issued on the demo call; usage per key is readable at GET /v1/usage.
03Which streaming transports are available?
Three surfaces on one base URL: POST /v1/tts/bytes for a complete WAV, POST /v1/tts/sse for base64 PCM chunks over one HTTP response, and wss://tts.gandr.ai/ws for one socket per live call with binary PCM frames back. PCM or WAV at any output sample rate.
04What counts against my bill?
Only the number of concurrent lines you hold — $150 a month per line annual, $180 month-to-month. Characters, minutes, voices, and seats are unmetered, so retries, silence, and long answers cost nothing marginal.
05What happens when I exceed my concurrency?
Traffic past your lines spills to burst lines at $10 a line-day — no busy signals to callers — and you can cap burst in the dashboard. Past absolute capacity the API refuses fast with a retryable 503 rather than hanging a live call.
06Can I bring a custom voice?
Every request can: the voice is {mode: "clone", wav_b64} with five to ten seconds of a consenting speaker, up to ~1.5 MB. References are fingerprinted and cached, so reuse skips re-cloning. Six multilingual stock voices also ship ready-made when you do not need a clone.
07Which languages are supported?
One cloned reference speaks 23 languages. Eight carry published, dated specimens at /languages/; the rest run live on the demo call.
08What happens if Gandr has an outage?
Serving runs on three tiers — primary GPU fleet, warm spill, and an independent fallback behind one endpoint — so a fleet outage costs latency, not availability. Standby failover lines ($20 a month each) also run beside your current provider, activating automatically if it goes dark.
09Is my text or reference audio used for training?
No. Customer text and reference audio never enter a training set, and every synthesized sample carries an inaudible provenance watermark applied at generation.
10How do I evaluate without signing anything?
The specimens on this page and the homepage showcase are ungated, real output of this API with generation dates. For your own script, the 30-minute demo call synthesizes your production prompts live and you leave with the recording.
See also
Related sheets.
capability
107 ms
Streaming synthesis that keeps up with a call
Chunked PCM from the first hundred milliseconds: SSE and WebSocket streaming TTS measured at 107 ms to first audio on the production API.
capability
1 ms
Low latency, with the receipts
Low-latency text to speech with the basis stated: 107 ms p50 / 108 ms p95 server time to first audio on the production API, and how to audit it yourself.
glossary
1 shape
SSE vs WebSocket for TTS
When to stream synthesis over server-sent events and when to hold a WebSocket: connection lifetime, turn structure, and infrastructure fit.
benchmark
5 rules
How we measure TTS latency
The methodology behind every figure on this site: server-time TTFA on the production API under call-shaped load, percentile pairs, published conditions.
Your script, this API, thirty minutes — and the recording leaves with you.
Hear it on your own script