Product · Spex-TTS
Built to answer.
Spex-TTS grew out of a production live-translation product. Every feature below exists because live speech needed it, and every number beside it is published, not promised.
6 features · belowCloning
Instant voice cloning
No training job, no per-voice fee, and your clone speaks all 23 languages.
How cloning works →
Streaming
Audio starts before the sentence ends.
Sockets carry conversations, server-sent events carry long reads, and one-shot WAV covers everything else.
Streaming in depth →
Drop-in
You already know this request shape.
Changing over is a diff, not a rewrite.
The full request shape →Drop-in
# either header works
x-api-key: gnd_yourkey
{
"transcript": "Dock four is clear.",
"voice": {"mode": "id",
"id": "gandr-jenny"},
"language": "en",
"output_format": {"container": "wav",
"sample_rate": 24000}
}
→ 200 · audio/wav · 24,000 Hz
Reliability
Three tiers sit behind one endpoint.
A tier going down costs you latency.
The latency benchmark →
Languages
A cloned voice keeps its identity.
It reads all 23 languages and still sounds like one person.
Hear the languages →Languages
Mia, one clone
three scripts · production API
Privacy
We do not train on your data.
Your text and reference audio never enter a training set.
The signed clip →
Provenance
Every clip leaves signed.
An inaudible watermark rides every sample and survives a re-encode or a phone line. It marks the audio as synthetic, not whose voice it is.
voice Leo · 19.68s · 2026-08-05
Live
Hear it live.
The console on the home page casts on the production API, type a line, pick a voice, and what comes back is the engine answering, not a recording.
Open the console →Every number on this page is measured on the production API, with its basis stated. The way to check them is a key of your own.
Get a key