You can design an American AI voice with one prompt
Describe the accent, gender, age and delivery in one sentence. Gradium returns a voice id you call from the API or the SDK, with no recording and no actor licensing. The prompt below is the one that produced Desmond, the voice in the sample.
American English, US English, Standard American, GenAm and general american all describe the same voice. Write it as American English in your prompt.
Hear it
- DesmondGeneral American0:00
“Imagine a problem so stubborn that it went unsolved for decades, and then, one afternoon, a graduate student notices something nobody else had ever thought to question.”
The prompt
that produced it
An American English male voice, 55 to 65: clean, deliberate and precise, with low pitch, slow pacing and low-to-mid energy, resonant timbre and a gentle low-to-high flow. Ideal for projecting academic authority.
This exact wording produced Desmond, a English · US voice now in the Gradium library, which is live in the library as bwRhQrJel4IuvxLF.
- Pitch1/5
- lowhigh
- Pace2/5
- slowfast
- Energy2/5
- calmboisterous
General American
in four styles
A prompt is built from parts: the accent, the age, then pitch, pace, energy and what the voice is for. Change one part and you get a variation. These four are starting points to adapt, not finished prompts.
The prompt as written. Low pitch, slow pacing and resonant timbre is the documentary register, and the gentle low-to-high flow is what stops it sounding like a recorded announcement.
An American English male voice, 55 to 65: clean, deliberate and precise, with low pitch, slow pacing, low-to-mid energy, resonant timbre and a gentle low-to-high flow. Ideal for projecting academic authority.
Copy this as it stands.
Everything to the middle. An agent interrupts and gets interrupted, so extremes on any axis become obvious across hundreds of turns in a way they never are in a single demo line.
An American English male voice, 30 to 40, for a real-time voice agent: clear and easy, with mid pitch, steady natural pacing, medium energy, resonant timbre and a gentle low-to-high flow. Ideal for projecting academic authority.
In colour: the 6 parts that make this style.
American tolerates high energy far better than British does, because its intonation range is naturally wider. This is the one accent where a 5 on energy still sounds like a person.
An American English male voice, 25 to 35: bright and punchy, with mid-high pitch, fast pacing, high energy, resonant timbre and a gentle low-to-high flow. Ideal for projecting academic authority.
In colour: the 5 parts that make this style.
Slow it further than feels right and let the resonance carry the weight. Asking for more energy here is the usual mistake: trailer voices are quiet and close, not loud.
An American English male voice, 40 to 55: gravelled and deliberate, with low pitch, slow pacing, low-to-mid energy, deep chest resonance with light vocal fry and a gentle low-to-high flow. Ideal for projecting academic authority.
In colour: the 4 parts that make this style.
What makes General American
sound like itself
Each feature below is something the prompt has to encode, or something a careless prompt will accidentally undo.
- Rhotic
- Post-vocalic /r/ is fully pronounced: car, water and father all keep an audible r. This is the clearest single split from British and Australian, and it is why a prompt that borrows a British reference for warmth will drift out of the accent partway through a line.
- Father-bother merger
- PALM and LOT collapse into the same vowel, so father and bother rhyme. British keeps them apart. Scripts carrying both words expose which side of the Atlantic a candidate has landed on within a sentence.
- T-flapping
- Intervocalic /t/ becomes a voiced tap, so butter, water and city soften towards a d. Asking for crisp or precise delivery suppresses it and can make the read sound formal or non-native, so pair precision with a purpose that wants it, such as narration.
- Wide intonation range
- General American uses a wider pitch excursion than Standard British, with more terminal movement on declaratives. An energy value that reads as theatrical in British sits comfortably here, which is why porting a prompt between the two needs the energy retuned rather than copied.
- What searchers usually mean
- American in a search means General American, the broadcast-neutral accent with no regional marking. Southern, New York and Californian are separate mapped labels, and asking for them by name is what gets them.
79 General American voices
already in the library
Ready to call with no design step. Hear a few here, then open any in the studio or write a prompt of your own.
- Abigail
Adult · Female
A warm and airy American adult voice that adds a touch of magic and empathy to any story.
0:00KRo-uwfno-KcEgBM - Alexandra
Adult · Female
A joyful and smooth American adult voice ideal for reading and hosting duties.
0:004nAcNUlNhEA_Kyjo - Alexis
Adult · Female
A joyful American adult voice that delivers customer service scripts with a bright distinct tone.
0:0074asmf7CXzjfopIX - Allison
Adult · Female
A joyful high-pitched American adult voice that brings high energy to training and teaching.
0:00yU6yxQ3e8LKRwU84
Another 74 General American voices are in the catalogue. Sign in to hear them all, or design one to your own spec from a prompt.
The prompt,
and where it goes wrong
The method behind the prompt: what each clause does, what breaks it, and what it was tested on.
Each row is one part of the prompt above and what it is set to here. Change a single row and you get a variation on this voice; change several and you get a different one.
| Clause | This prompt |
|---|---|
| Accent / locale | American English |
| Gender | male |
| Age range | 55 to 65 |
| Character | clean, deliberate and precise |
| Pitch | low pitch |
| Pace | slow pacing |
| Energy | low-to-mid energy |
| Resonance / timbre | resonant timbre |
| Modifiers | a gentle low-to-high flow |
| Applications | projecting academic authority |
4 failure modes we hit repeatedly, and what fixes each one.
- Asking for crisp or precise delivery suppresses T-flapping, so butter and city harden and the read starts sounding formal or non-native rather than American.
- Only ask for precision when the purpose wants it, such as narration or compliance. For conversational work, leave the articulation alone and let the accent flap.
- Requesting a neutral accent produces something ambiguous rather than General American, because neutral is not an accent the rewriter maps.
- Name American English. It is already the broadcast-neutral variety, which is what people mean when they ask for neutral.
- Regional asks such as Boston or Midwestern are not mapped and come back approximated. Southern American, New York and Californian are mapped and do work.
- Use one of the four mapped American labels, and shape everything else with the remaining clauses.
- Porting a British prompt across without retuning leaves the energy too low, because the same value reads calmer in American than it did in British.
- Add a step or two of energy when moving a prompt from British to American. The wider intonation range absorbs it.
Descriptor words do not transfer between locales. Energetic and bright travel intact, but most descriptors resolve to a different sound depending on the language they are read in. A prompt that produces a warm American voice will not produce the equivalent warmth translated into French, because the brief has to be rewritten for the locale rather than translated.
Writing your first prompt? How to write a voice prompt covers every part a prompt can carry, and what each one changes.
Questions?
Answered
"An American English male voice, 55 to 65: clean, deliberate and precise, with low pitch, slow pacing and low-to-mid energy, resonant timbre and a gentle low-to-high flow. Ideal for projecting academic authority." That is the exact prompt that produced Desmond, the voice in the sample above.
Yes. Searches for an American voice almost always mean General American, the broadcast-neutral variety with no regional marking. Southern American, New York and Californian are separate mapped labels and you get them by naming them.
79, spread across conversational, customer service, podcast and advertising work. They need no design step: pick a voice id and call the API. The American voice library page lists all of them with their ids.
Almost always because the prompt anchors to a British reference for some other quality, such as warmth or authority. American is rhotic and British is not, so a British reference pulls the r out mid-line. Keep every reference inside the target accent.
Both. Gender is its own part of the prompt, so it changes on its own: swap male for female and everything else holds. Same accent, same age, same delivery, different voice.
Hear General American
in the studio
A designed voice is a normal voice id: same endpoints, same SDK, same latency as any library voice. Every voice we ship clears blind listener comparison and an ELO ranking first, because roughly one candidate in a thousand has it.
Open a voice, type your own script, and listen. Then write a prompt of your own to design a variant. No recording, no casting, no licensing.