OpenAI TTS API playground
This tool calls the OpenAI audio speech endpoint directly from your browser with your own key, so you can audition voices and models on real text without writing code. Pick a voice, choose tts-1 or tts-1-hd, set the speed, and watch the character count and cost estimate update as you type — then generate and play the MP3 inline.
How it works
On generate, the playground POSTs to https://api.openai.com/v1/audio/speech
with your Authorization header, the chosen model, voice, input text, and
speed. The API returns the audio as MP3 bytes; the browser wraps them in an
object URL and feeds an inline <audio> player plus a download link. No server
sits in between and nothing is stored.
Voice character guide
OpenAI offers a fixed set of voices, each with a distinct character:
- alloy — neutral and balanced, works well for informational content
- echo — slightly masculine and crisp, suits narration
- fable — warm and expressive, good for storytelling
- onyx — deep and authoritative, popular for announcements
- nova — energetic and clear, often chosen for product demos
- shimmer — soft and conversational, fits chatbot or assistant UX
There is no voice cloning or custom voice upload — you pick from this list. The voices are not gender-locked by name, so listen before deciding; perception varies with pitch, pacing and surrounding audio context.
Comparing models: tts-1 vs tts-1-hd
Both models produce the same set of voices, but tts-1-hd applies additional processing to reduce artifacts and deliver higher fidelity at a higher cost per character. For exploratory work — testing rhythm, catching pronunciation issues, comparing voices — tts-1 is the practical choice. Switch to tts-1-hd for the final recording you ship to users.
Worked example
Paste a 200-character script, pick nova and tts-1. The cost estimate updates to show the per-character price multiplied by 200. Click Generate and the audio plays within a second or two for short text. Download the MP3 and listen in your target context — headphones, phone speaker, or embedded in a browser page — since the same voice can sound quite different across playback environments.
Tips and notes
- Iterate on tts-1, finalize on tts-1-hd. The standard model is cheaper for quick voice comparisons; switch to HD for the take you ship.
- Speed is a multiplier. Values from 0.25 to 4.0 are accepted; small nudges (0.9–1.1) sound more natural than extremes.
- Punctuation shapes pacing. Commas and periods drive pauses — edit the text, not just the speed, to fix rhythm.
- Cost scales with characters. The estimate is character-based, so trimming filler text directly reduces spend.
- Your key is never stored. It lives only in your browser’s memory for the duration of the request; closing the tab clears it entirely.