Create a Custom Voice

In this tutorial you will design a custom AI voice without any audio samples. No recording or reference clip needed. By the end, you will have a saved voice in your library ready to use in Projects for audiobooks, voiceovers, podcasts, and more.

The entire process takes just a few minutes, and most of that time is spent experimenting until you land on a voice you love.

When to Use Voice Design vs. Voice Cloning

Voice Creator Pro gives you two ways to create voices:

• Voice Design - Configure the voice you want using controls or a text description. Best when you have a specific character or style in mind but no reference audio.
• Voice Cloning - Provide a 3 to 10 second audio sample. Best when you want to replicate an existing voice.

Use Design when you need a voice that does not exist yet, when you want full creative control, or when you simply do not have a recording to clone from. If you already have audio of the voice you want, the Clone a Voice tutorial is a better starting point.

Prerequisites

• Voice Creator Pro installed and running on your machine

That is it. Voice Design requires no microphone, no audio files, and no external tools.

Step-by-Step Walkthrough

Step 1: Open the Design Tab

Open Voice Creator Pro and click Lab in the left sidebar, then select the Design tab at the top. You will see three panels:

• Voice Description (left) - where you configure the voice you want
• Generate Speech (center) - where you enter text and configure settings
• Output (right) - where you listen to and download the result

Step 2: Configure Your Voice

The Voice Description panel changes depending on which model you have selected (shown as a purple badge in the Generate Speech panel). Each model takes a different approach to voice design.

OmniVoice (Dropdown Controls)

With OmniVoice selected, you configure the voice using dropdown menus:

• Gender - Select male, female, or leave on Auto
• Age - Select an age range or leave on Auto
• Pitch - Select the pitch level or leave on Auto
• English Accent - Choose a specific accent or leave on Auto
• Whisper - Toggle on to produce a whispered speaking style

Start by setting Gender and Age, then experiment with Pitch and Accent. Leaving options on Auto lets the model decide, which can produce interesting results.

Qwen3 (Text Description)

With Qwen3 selected, you describe the voice you want in a free-text field. Write a natural language description of how the voice should sound. Here is a good starting point:

A warm, friendly woman in her late 20s with a conversational tone.
She speaks at a moderate pace with a slight smile in her voice.

The more specific you are, the closer the result will match your mental image. See Writing Effective Text Descriptions below for tips and examples.

Step 3: Enter Test Text

In the Generate Speech panel, type a sentence or two that represents the kind of content this voice will eventually read. Using realistic test text helps you evaluate whether the voice fits your use case.

For example, if the voice is for a meditation app, try something like:

Take a deep breath in. Hold it for a moment. Now slowly release.
Let your shoulders drop and feel the tension leaving your body.

If it is for a product explainer video, try:

Welcome to our platform. In the next two minutes, I will walk you
through everything you need to get started.

Step 4: Generate and Listen

Click the Generate button. The Output panel will play the result once processing finishes.

Listen carefully. Ask yourself:

• Does the age sound right?
• Is the tone what I described?
• Does the pace fit the content?
• Would I want to listen to this voice for an extended period?

Step 5: Iterate

This is where Voice Design really shines. If something is off, go back to the Voice Description panel and adjust your settings.

For OmniVoice: Change one dropdown at a time. If the voice sounds too high, lower the Pitch. If the accent is wrong, try a different English Accent option. Toggle Whisper on or off to hear the difference.

For Qwen3: Adjust your text description. Here are common tweaks:

• Voice sounds too fast? Add "speaks slowly" or "deliberate, unhurried pace."
• Voice sounds flat? Add emotional cues like "expressive" or "speaks with genuine enthusiasm."
• Voice sounds too young or too old? Be more specific about age. Instead of "young," try "mid-20s."
• Wrong mood? Add a comparison like "calm and soothing, like a meditation guide."

For a detailed guide on writing effective Qwen3 voice descriptions, read the Voice Design Prompting Guide.

Each time you make a change, click Generate again and compare. It usually takes two or three rounds to get a voice you are happy with.

Step 6: Save the Voice

Once you have a voice you like, click Save Voice to add it to your Design library. Give it a clear, descriptive name so you can find it later. For example:

• "Sophia - warm narrator"
• "Deep male - corporate explainer"
• "Energetic host - podcast intro"

The voice is now stored in the Design tab's Library dropdown and is also available for use in Projects.

Note: The Design library is separate from the Clone and TTS libraries. Voices you save here will only appear in the Design tab and in Projects.

Step 7: Use the Voice in Projects

To use your new voice in a long-form project:

1. Open the Projects tab and create or open a project
2. Go to Voice Assignment
3. Select your saved voice from the list of available voices
4. Assign it to a speaker, chapter, or the entire project

Your designed voice works exactly like a cloned voice or a built-in voice in Projects. You can mix and match all three types.

Writing Effective Text Descriptions

This section applies to models that use a text description field (like Qwen3). For OmniVoice, use the dropdown controls instead.

Here are practical examples that show the difference between a vague description and a specific one.

Vague vs. Specific

Descriptions for Common Use Cases

Audiobook narrator:

A mature male voice, mid-40s, with a rich, warm tone. He reads with
clear enunciation and a steady pace, drawing listeners into the story
without being dramatic. Think classic audiobook narrator.

Podcast host:

Energetic and enthusiastic young woman in her late 20s. She speaks
with excitement and a natural, conversational flow, like she is
telling a friend about something she is genuinely passionate about.

Corporate training narration:

Professional, neutral male voice, early 30s. Clear and articulate
with a moderate pace. Friendly but not overly casual. The kind of
voice you would hear in an onboarding video.

Children's story narrator:

A warm, gentle young female voice with a soft tone. She speaks
slowly and expressively, as if reading a bedtime story to a child.

For a deeper dive into prompting techniques, common mistakes to avoid, and an iterative workflow, read the Voice Design Prompting Guide.

Tips for Best Results

• Start simple, then refine. For OmniVoice, set Gender and Age first, then experiment with Pitch and Accent. For Qwen3, write a basic description first, then add details based on what you hear.
• Try both models. OmniVoice and Qwen3 produce different voice characteristics. Generate the same text with both to see which gets closer to what you want.
• Test with realistic content. Generate speech using the kind of text this voice will actually read. A voice that sounds great reading a single sentence might not work for a five-minute narration.
• Save versions. If you are iterating and hear a voice you like along the way, save it before continuing to tweak. You can always go back to an earlier version.
• Compare with built-in voices. Switch to the TTS tab and generate the same text with a few built-in voices. This gives you a baseline to compare your designed voice against.

Next Steps

• Voice Design reference - Full details on the Design tab and its options
• Voice Design Prompting Guide - Advanced prompting techniques and examples
• Projects - Use your new voice for long-form audio production
• Clone a Voice - Learn how to clone an existing voice from audio