20 learning cards. Four languages. Ready-to-use audio.

ACTUAL API OUTPUT

Start by listening.

Here is one phrase in each language, generated with the same prepared voice. The source file includes separate text for every language; the API does not translate.

Evaluate pronunciation and similarity with your target users. These samples have technical generation and file checks, not a claim of native-speaker review in every language.

Run the complete project.

npm install
node generate.mjs --plan

node generate.mjs --run

# output/ contains MP3 files, manifest.json
# and private recovery state.json

What each request does

const job = await client.createJob({
  model: 'clone-v1',
  voice_id: readyVoiceId,
  text: card.text,
  language: card.language,
  output_format: 'mp3',
}, { idempotencyKey: savedKey });

// Save job.id, then poll until succeeded.
const { audio, chargedUSD } = await client.getJobAudio(job.id);
await writeFile(card.id + '.mp3', audio);

This is an application loop, not a native batch endpoint. Requests run sequentially to respect the current one-operation workspace concurrency. Each row may wait for capacity.

Interrupted? Continue the same work.

Keep output/state.json private. Re-run with the same folder and unchanged inputs. A hard kill can leave a lock file; confirm the earlier process is stopped before removing that specific .running lock. Retained server output expires within 24 hours, so download promptly.

Every text stays paired with its audio.

This run generated 20 files using 469 characters at $8 per million. The API returned a total charge of $0.003752. Your available trial credit may change what you pay.