AI voice design: create a custom voice from a text description
Design voices forCharactersNarratorsPodcastsAdsGamesAudiobooks
Describe any voice in words, hear three new ones, and keep your favorite.
10M+ files processed since 2021
- 1DesignDesign voice
- 2SaveSave voice
- 3GenerateGenerate speech
Step 1 of 3: Design voice
0 / 1,000 characters
What to include
- Age and gender
- Accent or where they're from
- Tone, texture, and pace
Each design makes 3 new voices. None is a copy of a real person’s voice.
Hear voices designed from a description
Play an example, then start from its description.
Description
A deep, gravelly male movie-trailer narrator in his late fifties. Slow and dramatic, American accent, studio-quality recording.
Designed voice
“In a world where every sound matters, one voice rises above the noise. This fall, the story you thought you knew begins again.”
Voices in these examples are AI-generated, designed on Audo.
How to create a custom AI voice from a description
- 1
Describe the voice in a sentence or two: age and gender, an accent, its pitch and texture (deep, bright, raspy, breathy), and how it moves (slow, fast, calm, excited). Or start from an example.
- 2
Press Design voices and play the three new voices Audo makes from your words. Change a word or two and design again until one fits; every earlier voice stays on the page.
- 3
Choose a voice and save it to your account with a name, then type any script and generate speech in it as an MP3, in 30+ languages.
What people make with Audo's AI voice designer
A designed voice is new, belongs to no one, and stays in your account for every script after:
AI character voice generator
Character voices for games, animation, and audio dramas
Brief the voice the way you'd brief an actor: a sly goblin merchant, a weary space captain, a cheerful robot sidekick. Save each one under the character's name, and every line reads in the same voice, with acting tags such as [whispers] or [excited] in the script.
Female and male voice generator
Any age, gender, or accent, from a sentence
Ask for a bright young woman with a Scottish accent, a deep-voiced older man, or a calm gender-neutral narrator. Age and gender come through most reliably; name an accent plainly, and design again if the first three voices don't have it yet.
Narrator and brand voices
A narrator for videos, courses, and ads that no one else uses
Stock text-to-speech voices are heard on thousands of channels. A designed voice is new, so your explainers, lessons, and ads keep a sound of their own. Put music under it in Mix audio, or lay it over a video in Replace audio in video.
Voice design for Claude, ChatGPT, and agents
Ask your assistant to design a voice and read the script
Claude and ChatGPT can't make a new voice on their own. Connect Audo's MCP server and they can: design_voice makes three previews from a description, save_voice keeps the one you pick, and clone_voice reads any script in it. Developers call the same tools over the REST API.
How AI voice design works, and the words that shape a voice
Audo's AI voice designer, a custom AI voice generator, turns a written description of 20 to 1,000 characters into three new voices, made with ElevenLabs' voice design model. Each reads a line of about 20 to 30 seconds written to suit it, so you compare them by ear rather than by settings. None is a copy of a real person: nothing is recorded or uploaded.
The voices follow some words more reliably than others. Age and gender come through almost every time, and texture (raspy, breathy, gravelly) and pace (slow, fast-talking) usually do. Accents vary the most: name the accent plainly, such as Scottish, Texan, or Irish, and design again if the first three miss it. One or two clear sentences work better than a long list of adjectives.
Every set from your visit stays on the page, so a voice from an earlier try can still be the one. The voice you choose is saved with its description and its whole preview as a sample, and from then on works like a cloned voice: pick it on Voice cloning, pass its voice_id to the API, or ask an assistant to read a script in it. Speech in a saved voice is made with Fish Drama 3 in 30+ languages, as a 192 kbps MP3, with acting tags such as [whispers], [sighs], or [excited].
| What to describe | Words that work | Example description |
|---|---|---|
| Age and gender | in her twenties, middle-aged, elderly, gender-neutral | A gender-neutral narrator in their thirties |
| Pitch and weight | deep, low, high, bright, booming, thin | A deep, booming announcer |
| Texture | raspy, gravelly, breathy, husky, smooth, nasal | A breathy, husky whisper |
| Pace and rhythm | slow, unhurried, fast-talking, measured, long pauses | Slow and unhurried, with long pauses |
| Mood and energy | warm, calm, upbeat, sly, weary, menacing | Warm and upbeat, smiling as she speaks |
| Accent or place | Scottish, Texan, Irish, light London, neutral American | A light London accent |
| Role and setting | trailer narrator, podcast host, storyteller, studio recording | A podcast host in a quiet studio |
Questions
Is Audo's AI voice designer free?
Audo designs voices free 3 times a day with no sign-up, three new voices each time. After that, a design costs $0.02, paid as you go with no subscription (see pricing). Saving a voice is free (3 voices on a free account, 10 after any purchase), and speech in a saved voice costs $0.12 per 1,000 characters, about a minute of speech.
How do I describe a voice for AI voice design?
In Audo's Voice design, describe who is speaking and how they sound in one or two plain sentences: age and gender ("a woman in her sixties"), pitch (deep, high, bright), texture (raspy, gravelly, breathy, smooth), pace (slow, fast-talking, unhurried), mood (warm, sly, weary, upbeat), and an accent or place. A role helps too, such as "trailer narrator" or "bedtime storyteller". Short and clear beats a long list of adjectives.
Are designed voices copies of real people?
No. Each voice is made new from your words, so there's nothing to record and no one's permission to ask, which voice cloning needs. Describe a style, not a person: asking for a celebrity doesn't copy their voice.
Can Claude or ChatGPT design a voice and read my script?
Not on their own: neither app makes new voices. With Audo's MCP server connected, they can ask Audo to design three voices from your description, save the one you choose, and read your script in it as an MP3. The same steps work from Cursor and other MCP clients.
Do I need to disclose an AI voice on YouTube?
YouTube asks creators to disclose AI content that seems realistic, such as making a real person appear to say something they didn't; content that isn't realistic, like a clearly animated character, doesn't need it. You disclose it under AI use when uploading in YouTube Studio, as YouTube's help page explains.
What's the difference between voice design, voice cloning, text to speech, and voice changer?
Text to speech reads a script in a ready-made studio voice. Voice cloning copies a real voice from a recording, with the speaker's permission. Voice design is an AI voice creator: it makes a new voice from a description, which you then save and use just like a cloned voice. Voice changer turns a recording of someone speaking into another voice.
More tools
- Transcribe audioSpeech to text, who said what.
- Text to SRTSync a script to audio.
- Auto captionsPut your words on the picture.
- Remove filler wordsCut the ums and uhs.
- Bleep censorBleep or mute swear words.
- Audio translatorSpeech into translated text.
- Translate subtitlesSubtitles in more languages.
- Text to speechTurn a script into a voiceover.
- SRT to speechVoice your subtitles on time.
- Voice cloningClone your voice for any script.
For AI assistants and developers
Ask ChatGPT, Claude, or a coding agent to do this for you, or call it from your own code. When an agent's free result is cut short, it gives you a link to finish the whole file, no sign-up.
Show the free API call
curl -F 'text=A warm, calm woman in her forties with a soft Irish accent, unhurried and friendly.' https://audo.ai/api/try/voice-design -D response.headers -o response.body