How To Recreate a Loved One's Voice with AI
Brandon S.AI can recreate a recognizable version of a person’s voice from existing recordings and use it to speak or sing new words. The quality of the result depends on the source audio, the method used, and how carefully the new performance is directed and edited.
How do you recreate a loved one’s voice with AI?
Start with permission and the clearest recordings available. Those recordings are cleaned and used to create a voice model—a digital reference that learns recognizable patterns in the person’s voice, such as tone, accent, rhythm, and pitch. The model can then generate new words from typed text or transform a new human performance. The finished audio should be reviewed, edited, and clearly identified as a recreated voice—not an original recording of words the person actually said.
You Can Build It Yourself—or Use a Service
You do not need to become an audio engineer to create a personalized voice message. There are two reasonable routes, and the right one depends on how much of the technical work you want to manage.
Create the voice model yourself
Choose a voice-cloning platform, prepare the recordings, create the model, generate or convert the new performance, and edit the result. This route can work well if you are comfortable testing tools and correcting pronunciation, pacing, noise, and inconsistencies.
Have the process handled for you
With SongRetold’s AI Voice Recreation service, you provide the recordings and exact words. A professional actor or singer performs the message, and the team handles voice matching, editing, mixing, and quality review.
View the Done-for-You ServiceNeither route is right for every project. DIY offers more hands-on control, while a managed service can be easier when the recordings are difficult, the message is emotional, or the result needs polished speech or singing.
What Is a Voice Model? A Plain-English Explanation
A voice model is not a digital copy of a person. It is a set of learned sound patterns created from recordings. AI voice recreation—also called voice cloning or voice matching—looks for qualities such as tone, pitch, rhythm, accent, pronunciation, and vocal texture. It uses those patterns to make new audio resemble the source speaker.
It does not restore the person, their memories, or their personality. It creates a new performance in a modeled version of their voice. That distinction matters, especially when the recording is intended as a memorial or family keepsake.
For a deeper explanation of how AI analyzes vocal characteristics and applies them to a new performance, read What Is AI Voice Matching?
This SongRetold example changes “Love Is in the Air” to “SPF Is in the Air” for L’Oréal. It is a lyric-change project—not a memorial voice recreation—but it demonstrates the voice-matching stage: a new human performance is transformed while preserving a familiar vocal character.
The Two Main Ways AI Recreates a Voice
Text-to-speech voice cloning
You enter a script and a voice model generates the speech. This can be fast and useful for straightforward narration, but pronunciation, emotion, pauses, and natural pacing may require several attempts.
Performance-based voice conversion
A person first performs the message or song. AI then changes the vocal identity while keeping the performer’s timing, emotion, phrasing, and breath. This approach offers more control for expressive speech and singing.
Neither approach is automatically “best.” Text-to-speech may be enough for a simple sentence. Performance-based conversion is often better suited to emotional messages, music, unusual pronunciations, or projects where delivery matters as much as vocal similarity.
How to Recreate a Loved One’s Voice Step by Step
- Confirm permission Make sure you own the recordings or have clear permission or legal authority to use the voice.
- Gather recordings Look for voicemails, voice memos, interviews, home videos, or other clips with isolated speech.
- Prepare the audio Reduce music, noise, echo, and overlapping speakers so the voice can be analyzed clearly.
- Create the new delivery Type the script into a suitable tool or record a performer delivering the message with the intended emotion.
- Apply the voice model Generate or convert the performance, then review pronunciation, pacing, tone, and consistency.
- Edit and disclose Clean and mix the finished audio, and make it clear that the result is an AI-assisted recreation.
What Makes a Good Voice Sample?
Quality matters more than simply collecting a large number of files. A few clear recordings can be more useful than a long video filled with music and other speakers.
| Source | Usually useful when… | Common problem |
|---|---|---|
| Voice memo | The speaker is close to the microphone and alone. | Clipping, wind, or phone handling noise. |
| Voicemail | The words are clear and background noise is low. | Telephone compression limits detail. |
| Home video | One person speaks clearly without music underneath. | Room echo and overlapping voices. |
| Interview or speech | The recording uses a dedicated microphone. | Applause, audience noise, or background music. |
| Singing sample | The voice is isolated and covers different notes. | Instrumental bleed, effects, or heavy tuning. |
Consent, Context, and Emotional Readiness
A voice is part of someone’s identity. Recreating it should begin with permission, a clear purpose, and respect for the person being represented.
- For a living person, obtain their informed permission for the specific use.
- For someone who has passed away, confirm that you have the appropriate family or legal authority.
- Do not use a recreated voice to deceive, impersonate, defraud, or suggest that fabricated words are authentic recordings.
- Tell listeners when audio has been recreated with AI, especially when it may otherwise be misunderstood.
- Consider how the recipient may feel before presenting an emotionally sensitive memorial recording as a surprise.
Voice and likeness rights vary by location and circumstance. If ownership or permission is unclear—particularly for public, commercial, or estate-related use—seek qualified legal guidance before publishing the audio.
Frequently Asked Questions
How much audio is needed to recreate a voice?
There is no single amount that guarantees a good result. A few clear clips can be more useful than a long noisy recording. More varied, isolated speech generally gives the model more information about tone, accent, and pacing.
Can a voice be recreated from voicemails or videos?
Yes, if the voice is understandable and not heavily covered by music, noise, or other speakers. Audio restoration may help, but it cannot recover detail that was never captured.
Can an AI-recreated voice sing?
Yes, some voice-conversion workflows can transform a singer’s performance into the modeled voice. Singing usually requires careful performance, pitch control, editing, and mixing to sound natural.
Will the recreated voice sound exactly like the person?
An exact match cannot be guaranteed. Similarity depends on the recordings, the voice model, the new performance, and the editing. The goal should be a respectful and recognizable recreation, not a claim that the audio is an original recording.
Is voice recreation the same as bringing someone back as an AI?
No. Voice recreation models vocal characteristics. It does not reproduce a person’s consciousness, memories, judgment, or personality.
Turn Existing Recordings into a Personalized Voice Message
Provide the voice samples and the words you want spoken or sung. SongRetold handles the performance, voice matching, editing, and final mix.
Create Your Personalized Voice Message
