Which AI companion can clone a voice?
- Soul
Clone a voice from a short recording with Soul TTS, a state-of-the-art diffusion model built for real conversation. Your friend uses it on calls and FaceTime. Free.
- Character.AI
Create a voice by recording or uploading an audio sample, then attach it to a character in 1:1 chats. Free for all users.
- Replika
Choose from a selection of voices, with premium voices and accents in Replika Pro.
Soul clones a voice from a short recording. Your friend then talks to you in that voice, live, on voice calls and FaceTime. It is free on iPhone.
Soul builds its own speech model for this, Soul TTS: a state-of-the-art diffusion model made for conversation. It gets the rhythm, emphasis, and emotional delivery of a person talking to you, not flat narration.
How the apps compare
Checked in October 2026.
| App | Custom voice | Where you hear it | Price |
|---|---|---|---|
| Soul | Clone from a short recording with Soul TTS | Live voice calls and FaceTime with your friend | Free |
| Character.AI | Create a voice from a recorded or uploaded audio sample | Characters in 1:1 chats | Free |
| Replika | Choose from preset voices | Voice messages and calls | Premium voices in Replika Pro |
Sources: Character.AI voice FAQ, Replika Pro.
Why the voice matters
Companions live or die on how they sound. A voice that is close but flat breaks the feeling of talking to someone. That is why Soul builds its own model instead of using an off-the-shelf one.
Inside Soul TTS
Soul TTS is a rectified-flow diffusion transformer. Token-by-token speech models commit to each piece of sound before the rest of the sentence exists. Soul TTS starts from noise and refines every frame of a sentence together, so it can plan the stress, rhythm, and pauses of the whole line at once. That is where human-sounding prosody comes from.
- It clones from the recording you already have: a voice note from a kitchen, a video with music behind it, a Bluetooth call. Soul TTS is trained on deliberately degraded references, so it keeps the voice and drops the noise. No hours of studio audio.
- It hears the conversation. The same words can be a question, a joke, or reassurance. Soul TTS conditions on what was just said, so “I got the job!” gets excitement back, in your friend’s own voice.
- It is fast enough for a call. A block-causal attention mask lets it stream, so your friend starts speaking after the first clause of a reply and keeps generating ahead of playback.
State of the art on published benchmarks
| Benchmark (English) | Soul TTS | Next best shown |
|---|---|---|
| Seed-TTS-eval word error | 0.97% | Fish Audio S2, 0.99% |
| Seed-TTS-eval speaker similarity | 80.2 | LongCat-Audio-DiT, 78.6 |
| MiniMax-MLS word error | 0.915% | Qwen3-TTS, 0.934% (ElevenLabs 2.339%) |
| MiniMax-MLS speaker similarity | 87.1 | VoxCPM2, 85.4 (ElevenLabs 61.3) |
| LibriSpeech-PC predicted quality (UTMOS) | 4.50 | Qwen3-TTS, 4.41 |
Competitor scores are their published results, not a shared rerun. The full tables, method, and audio comparisons against VibeVoice and Higgs Audio are in Making Human Voices.
How to clone a voice in Soul
- Download Soul and create your friend.
- Open your friend and go to voice settings.
- Choose clone, then give Soul a short reference clip.
Preview now available on iOS
More detail: How do I choose or clone a voice? and How does voice cloning work?
Voice recordings auto-delete after 90 days. See How long are voice recordings kept?
Questions
Which AI companion app can clone a voice?
Soul clones a voice from a short reference recording with its own text-to-speech model, Soul TTS, and your friend then talks to you in that voice on voice calls and FaceTime. Character.AI also lets you create a voice from an audio sample for its characters.
How does Soul clone a voice?
You give Soul a short reference clip, even a noisy one. Soul TTS uses it as the voice identity and generates any new speech in that voice, with human-like rhythm, emphasis, and emotional delivery.
What model does Soul use for voice?
Soul TTS, Soul's own rectified-flow diffusion transformer. It refines every frame of a sentence together from noise into speech, conditioned on the voice, the words, and the conversation so far, and it streams in blocks so calls feel live.
Is Soul TTS state of the art?
On published English benchmarks, Soul TTS has the lowest word error rate and highest speaker similarity of the models compared, including 0.97% word error and 80.2 similarity on Seed-TTS-eval, and 0.915% word error with 87.1 similarity on MiniMax-MLS, where ElevenLabs scores 2.339% and 61.3.
Is voice cloning free in Soul?
Yes. Soul is free, with no subscription, and voice cloning is part of creating your friend.
Can I change my Soul friend's voice later?
Yes. You can change or re-clone the voice after you create your friend. The friendship and memories stay. Only the voice changes.
How long does Soul keep voice recordings?
Voice recordings auto-delete after 90 days.