Setting up Voice Cloning with your own writing samples

title: "Setting up Voice Cloning with your own writing samples" description: "From upload to first humanized draft with your profile applied, in five minutes." date: "2026-05-07" tag: "Guide" author: "Inksong"

Voice Cloning is the difference between "humanized text" and "text that sounds like you wrote it." It's a Pro and Enterprise feature; the free tier ships humanization without voice profiles. This guide walks you through training your first profile and applying it to a humanization run.

Step 1 — Find a representative sample

The floor for a usable profile is 1,500 words of your own writing. 3,000+ words gets noticeably better results. Two rules:

Same domain as what you'll be humanizing. Don't train on your blog and apply the profile to sermons. The features we extract (vocabulary specificity, hedging density, sentence rhythm) are domain-dependent. Match the training sample to the target use case.
Writing you'd be happy to have your humanized work resemble. This sounds obvious — but the profile captures the patterns in the sample, including any tics you're not proud of. Pick a piece you consider representative of your good voice, not a casual draft.

For most users, a published paper, a finished blog post, or a recent newsletter issue is the right input.

Step 2 — Upload via Voice Profiles

In the dashboard:

Click Voice Profiles in the left nav.
New profile in the top right.
Give it a name (e.g., "Academic — economics", "Marketing blog voice").
Paste the text or upload a DOCX/TXT/MD file.
Save.

The text is stored alongside the extracted features. You can delete the profile (and the underlying text) at any time.

Step 3 — Training

When you save, Inksong extracts the profile features:

N-gram fingerprint (2-3 word sequences that appear unusually often)
Sentence-length distribution
Burstiness (length variance across paragraphs)
Vocabulary specificity (your lexical band)
Hedging density

Training is fast — usually finishes in a few seconds — because we're not fine-tuning anything; we're computing statistics. The profile then becomes available in the Voice Profile dropdown on any humanization run.

If you're curious about how these features get used in the rewrite, the engineering post on Voice Cloning walks through the prompt strategy.

Step 4 — Apply to a humanization run

On the home page or /dashboard, upload or paste your AI-generated draft. In the Settings panel, find the Voice Profile dropdown. Pick your profile. Set tone and domain appropriately for the target document. Hit Humanize.

The result will read closer to your training sample than a humanization without the profile. Run a diff comparison to verify.

Step 5 — Iterate

If the output still reads neutral:

Sample too short. Try a longer sample (5,000+ words).
Sample in a different domain. Match training and target domains.
Humanness too low. With voice profiles, you can usually run a higher humanness (60–70) and trust that the profile keeps the output in your voice. Without a profile, that level often reads too casual.

You can have multiple profiles — academic for papers, casual for blog posts, etc. Pro plans support multiple profiles; Enterprise plans support unlimited.

What Voice Cloning is not

It is not a copy of your voice that will write new things in your style. Inksong only humanizes text you've already drafted (with or without AI help). The profile shapes how the rewrite sounds; it doesn't generate content.

It is also not a voice classifier. We don't try to identify "this is Mike's writing" given an arbitrary text. The profile is one-way: features in, prompt guide out.

What to do next

If you write across multiple domains, train one profile per domain.
Solutions lists audience-specific tone/domain/humanness presets that pair well with Voice Profiles.
For an academic-specific walkthrough that combines voice cloning with the rest of the workflow, see Humanizing an academic paper.