Text-to-Speech

Text-to-Speech Quickstart

Text in, audio out. Destesi Text-to-Speech turns any string into natural, studio-quality speech from a curated catalog of nine built-in voices, each of which speaks every supported language.

Every clip is workspace-scoped and lands in your shared Drive library, ready to reuse across the suite. You drive it from the web console at tts.destesi.io or directly over a small HTTP API.

The fastest path to first audio

The quickest way to hear it: open the console and click Generate.

1. Sign in. Open account.destesi.io, sign in (or sign up), and pick the workspace you want to work in.

2. Open Text-to-Speech. Click the Text-to-Speech tile in the 9-dot launcher, or go straight to tts.destesi.io. Single sign-on carries your session across — you land on the synthesize console already authenticated.

3. Generate a clip. On the synthesize console:

   • Pick a built-in voice (try Aria for a warm female voice, or Atlas for a deep male voice).
   • Type or paste your text.
   • Leave the language on Auto to let the model detect it, or pick a hint.
   • Optionally add a delivery direction — see Shape the delivery below.
   • Click Generate. The clip plays inline, and a Download button saves a .wav.

4. Find it later. Every generated clip appears under Jobs to replay or re-download, and is registered in your Drive library so any other Destesi product can reuse it. Group related clips into a project to keep a long piece of work together.

Shape the delivery

One per-request control changes how a line is read, not what it says: instruct, a short natural-language delivery direction — “warm and unhurried”, “urgent, like breaking news” — up to 500 characters.

The API tells you loudly rather than quietly ignoring you: send instruct where it does not apply and you get 400 unsupported_parameter instead of audio that silently ignored your direction. GET /v1/voices reports each voice’s capabilities list, so you can check before you call rather than guess.

Call it from your own code

To synthesize from a script, a backend, or another tool, authenticate with a Personal Access Token.

1. Mint a token. Open account.destesi.io/settings/api-tokens, click Create token, name it, and copy the plaintext. It is shown exactly once — store it in a secret manager or .env immediately. Tokens begin with idn_pat_. See API access & the dst CLI for how tokens work across the suite.

2. Note your workspace slug. The workspace switcher in account.destesi.io shows it. Every token call carries it in the X-Destesi-Workspace header so the request runs in the right workspace — one token works for every workspace you belong to.

3. Synthesize. A short clip renders inline with sync: true:

   export DESTESI_PAT="idn_pat_paste_your_token_here"
   export DESTESI_WORKSPACE="your-workspace-slug"
   
   curl -sS -X POST https://api.tts.destesi.io/v1/synthesize \
     -H "Authorization: Bearer $DESTESI_PAT" \
     -H "X-Destesi-Workspace: $DESTESI_WORKSPACE" \
     -H "Content-Type: application/json" \
     -d '{
       "text": "Hello from Destesi.",
       "voice_id": "voice_builtin_aria",
       "sync": true
     }'

   The response carries a signed audio_url you can download:

   {
     "job_id": "job_abc123",
     "status": "done",
     "audio_url": "https://…/jobs/job_abc123.wav?…",
     "duration_ms": 1840
   }

Tip

Inline rendering only happens when you send "sync": true, the text is 500 characters or fewer, and the engine is warm. For longer text (or if you omit sync), the call returns 202 with a job_id, and you poll GET /v1/jobs/{id} until the status is done. A 202 with "status": "queued" also arrives when the engine is still warming up — the job is safely queued, not lost. See Async response — 202 for the polling flow, or the full Reference for the request shape and limits.

Safe to retry

Send an Idempotency-Key header and a repeat of the same request returns the original job instead of synthesizing a second clip — useful when a network blip leaves you unsure whether the first call landed. Reusing a key with a different request body is a mistake, and is rejected with 409 idempotency_conflict rather than silently returning the wrong audio.

See Voices for the catalog and the languages it covers.

Where to next

ConceptsThe mental model: voices, jobs, and how speech flows into your Drive library.

ReferenceEvery endpoint, request field, voice ID, error code, and limit.

TroubleshootingThe engine is warming up, a link expired, text is too long — and how to fix each.

DriveThe one library where every clip you generate is stored.