How many voices does OpenAI TTS offer?

The core OpenAI speech models ship six built-in voices — alloy, echo, fable, nova, onyx, and shimmer. They each have a distinct tone, and the same voice name works across the tts-1 and tts-1-hd models.

Which OpenAI voice is best for narration?

Onyx and alloy tend to work well for narration thanks to a steady, grounded tone, while nova is a good warmer alternative. The right pick depends on your audience and brand, so audition a couple with the same script.

Can I change the speaking speed?

Yes. The speech API accepts a speed parameter, typically from 0.25 up to 4.0, with 1.0 as normal. Slowing slightly improves clarity for instructional content, while speeding up suits skimmable summaries.

What audio formats can OpenAI TTS return?

You can request mp3, opus, aac, flac, wav, and pcm. Use mp3 or aac for general distribution, opus for low-latency streaming, and wav or flac when you need lossless audio for editing.

Is the voice the same across tts-1 and tts-1-hd?

The voice identity is the same, but tts-1-hd renders at higher audio quality with a bit more latency and cost. Use tts-1 for real-time or high-volume use and tts-1-hd for polished published audio.

OpenAI TTS Voice Picker

OpenAI TTS voice picker

OpenAI’s speech models ship six voices — alloy, echo, fable, nova, onyx, and shimmer — and the right one depends entirely on what you are making. This picker pairs each voice with its tone, best use cases, and a sample phrase, then recommends a shortlist based on your use case and tone preference.

How it works

Every voice has a consistent character regardless of model. The picker scores each one against two inputs:

Use case — narration, assistant/agent, audiobook, advertisement, or character — because a calm reader and a punchy promo voice rarely overlap.
Tone — warm, neutral, or authoritative — which biases toward the voices that carry that feel.

It then surfaces the top matches with notes on speed (the API’s speed parameter, 0.25–4.0) and audio format so you can drop the choice straight into your request.

Tips for choosing a voice

Audition with your real script. A generic “hello world” hides how a voice handles your actual sentences, names, and pacing.
Match voice to medium. Onyx and alloy read long-form well; nova and shimmer feel friendlier for assistants and welcome flows.
Set speed deliberately. Drop to ~0.9 for instructions, push to ~1.1–1.2 for snappy summaries.
Pick format by destination. mp3/aac to ship, opus to stream, wav/flac when you will edit the audio afterward.