AI Text-to-Speech Voiceover

AI Text-to-Speech Voiceover

Turns a block of text or a full script into a natural-sounding AI voiceover file. Pick a voice, set the speed, and choose MP3, WAV, Opus, or AAC. It's meant for the usual narration jobs: faceless videos, audiobooks, IVR prompts, explainer voiceovers.

How it works

The actor sends your text to an OpenAI-compatible TTS endpoint. Long scripts get split at sentence boundaries into chunks under ~3,500 characters, each chunk is synthesized separately, and the parts are stitched back into one file with ffmpeg (using stream copy, so there's no re-encode and no quality loss). Each finished audio file is saved to the run's key-value store and a row is pushed to the dataset.

Input

Nothing is strictly required by the schema, but in practice you need an openaiApiKey and at least one of text or texts. If neither is provided the run errors out.

Output

Each input item produces one audio file in the key-value store and one dataset record. The record includes audioKey and audioUrl (where to fetch the file), durationSeconds, characters, chunks (how many pieces the script was split into), plus the voice, model, and resolved format. Failed items get a record with ok: false and the error message instead of stopping the whole run.

Example

{
  "text": "Welcome back to the channel. Today we're looking at one of the strangest mysteries of the deep ocean.",
  "voice": "onyx",
  "model": "tts-1",
  "format": "mp3",
  "speed": 1.0,
  "openaiApiKey": "sk-..."
}

Pricing

$0.04 per voiceover — $40.00 per 1,000 — plus a $0.0000125 run-start fee each time a run begins. Flat rate: no volume tiers, no plan gates, no subscription. The OpenAI TTS usage is billed separately on your own key, so the only Apify-side cost is the per-voiceover fee above.

Batch mode is where that rate matters. Because texts accepts an array, one run can voice a whole week of scripts in a single start fee — 50 narration blocks cost $2.00 plus a fraction of a cent to start the run.

Notes

This actor calls OpenAI for synthesis, so it needs your own OpenAI API key. Individual chunks are capped at 4,000 characters before they're sent, which keeps each request within the model's per-call limit; there's no hard limit on total script length since long inputs are chunked and concatenated.