Multilingual TTS API: 10 languages on one OpenAI-compatible endpoint — Audra

Localizing product narration usually means either one expensive multilingual vendor or a frankenstein of regional APIs. Audra Essential keeps it boring on purpose: ten languages, 72 preset voices, one OpenAI-compatible HTTP shape, and −16 LUFS mastering on the file you download.

This is for batch narration — courses, docs, support macros, marketing pages — not realtime agents or voice cloning. If you need sub-200ms conversational turns or a custom clone of a human speaker, stay on a provider built for that.

Who this is for

• Product teams shipping the same lesson or help article in multiple locales
• Docs and support orgs that want one TTS contract instead of a provider per language
• Builders already on the OpenAI SDK who can swap baseURL and keep going
• Anyone who wants to preview locale-matched voices before hard-coding production slugs

Languages on Essential

Essential covers the ten locales below on a single endpoint. Prefer a voice slug that already implies the locale (filter the catalog on /voices). Set language as a BCP-47 hint only when the slug is ambiguous — never mix languages inside one request.

Picking a voice and locale

• Filter /voices and /compare by language before locking a production slug
• Keep one locale per POST — split multilingual scripts into separate jobs
• Short UI strings → sync /v2/speech; long localized docs → jobs API
• Need denser multilingual presets than Essential? Check Plus/Pro multipliers on /pricing#engines

Example: French narration

Use POST /v2/speech when you need an explicit BCP-47 language hint. OpenAI SDK clients can still point baseURL at /v1 for simple English narration — for locale overrides, prefer the native examples below.

curl
curl https://audratalks.com/v2/speech \
  -H "Authorization: Bearer sk-live-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"audra-essential","text":"Bonjour — narration de documentation.","voice":"june","language":"fr-FR","format":"mp3"}' \
  --output out.mp3

Python
import requests

r = requests.post(
    "https://audratalks.com/v2/speech",
    headers={"Authorization": "Bearer sk-live-YOUR_KEY"},
    json={
        "model": "audra-essential",
        "text": "Bonjour — narration de documentation.",
        "voice": "june",
        "language": "fr-FR",
        "format": "mp3",
    },
    timeout=120,
)
r.raise_for_status()
open("out.mp3", "wb").write(r.content)

JavaScript
import { writeFile } from "node:fs/promises";

const res = await fetch("https://audratalks.com/v2/speech", {
  method: "POST",
  headers: {
    Authorization: "Bearer sk-live-YOUR_KEY",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "audra-essential",
    text: "Bonjour — narration de documentation.",
    voice: "june",
    language: "fr-FR",
    format: "mp3",
  }),
});

if (!res.ok) throw new Error(await res.text());
await writeFile("out.mp3", Buffer.from(await res.arrayBuffer()));

When you need an explicit language hint (or jobs/webhooks), prefer POST /v2/speech. Preview candidates on /voices, then A/B on /compare before you bake the slug into CI.

Cost sanity check

FAQ

Next steps

• Preview locale-matched voices on /voices
• A/B two candidates on /compare
• Read the API reference on /docs
• Get 25k free characters at /dashboard — no card