---
type: source
id: barrington-2025-voice-clones
title: People are poorly equipped to detect AI-powered voice clones
authors:
- Sarah Barrington
- Emily A. Cooper
- Hany Farid
year: 2025
venue: Scientific Reports
volume: '15'
pages: '11004'
doi: 10.1038/s41598-025-94170-3
peer_review: peer-reviewed-journal
evidence_tier: B
design: 'Online perceptual studies: identity study (same speaker? for real-real, real-clone and different-speaker pairs) and naturalness study (real vs AI classification of single clips); small exploratory follow-up'
modality:
- audio
stimuli: ElevenLabs Instant Voice Cloning clones of 220 real speakers from the DeepSpeak dataset (32 prompts per speaker); identity batches of 44 stimuli
n_participants: 634
n_note: Identity study n=304; naturalness study n=300 recruited (294 analysed); exploratory follow-up n=30
topics:
- human-detection
- voice
- voice-clone
- identity
- vishing
summary: Listeners judged a commercial voice clone to be the same person as the real speaker about 80% of the time (median 83.3% of real-clone pairs) and identified clones as AI only 60.8% of the time.
results:
- metric: 'Identity: real-real same-speaker pairs judged same'
  value: median 100.0% (IQR 85.7-100.0)
  note: ''
  source_location: null
- metric: 'Identity: real-clone pairs judged same speaker'
  value: median 83.3% (IQR 69.2-100.0)
  note: 'abstract: about 80% of the time; 26.6% of listeners said ''same'' on every real-clone trial'
  source_location: null
- metric: 'Identity: different real speakers judged same'
  value: median 11.1%
  note: ''
  source_location: null
- metric: 'Naturalness: real clips correctly classified'
  value: 67.4% (SD 14.8)
  note: ''
  source_location: null
- metric: 'Naturalness: AI clones correctly classified'
  value: 60.8% (SD 16.7)
  note: 21.0% of listeners at or below chance on clones
  source_location: null
moderators:
- 'Exploratory follow-up: unscripted clips easier than scripted (median 76.7% vs 56.7%)'
limitations:
- Unfamiliar speakers
- Single commercial cloning service
- Short clips, online listening
verification:
  status: verified-fulltext
  confidence: high
  checked: 2026-10-12
  via: nature.com full text, doi:10.1038/s41598-025-94170-3
url: null
pmid: null
pmcid: null
arxiv_id: null
openalex_id: null
oa_url: null
cited_in:
- MR-0001
- MR-0003
updated: '2026-10-12'
---

# People are poorly equipped to detect AI-powered voice clones

> **Source record** `barrington-2025-voice-clones` · 2025 · Scientific Reports  
> **Evidence tier:** B (peer-reviewed-journal) · **Verification:** verified-fulltext  
> **Bottom line:** Listeners judged a commercial voice clone to be the same person as the real speaker about 80% of the time (median 83.3% of real-clone pairs) and identified clones as AI only 60.8% of the time.

## Citation {#citation}

Sarah Barrington, Emily A. Cooper, Hany Farid (2025). People are poorly equipped to detect AI-powered voice clones. *Scientific Reports*, 15, 11004. https://doi.org/10.1038/s41598-025-94170-3

Verification entry (identifiers, resolvers, claims with source locations): [/data/citations.json#barrington-2025-voice-clones](/data/citations.json#barrington-2025-voice-clones)

## Study design {#design}

| Field | Value |
|---|---|
| Design | Online perceptual studies: identity study (same speaker? for real-real, real-clone and different-speaker pairs) and naturalness study (real vs AI classification of single clips); small exploratory follow-up |
| Modality | audio |
| Stimuli | ElevenLabs Instant Voice Cloning clones of 220 real speakers from the DeepSpeak dataset (32 prompts per speaker); identity batches of 44 stimuli |
| Participants (N) | 634 |
| N detail | Identity study n=304; naturalness study n=300 recruited (294 analysed); exploratory follow-up n=30 |
| Topics | human-detection, voice, voice-clone, identity, vishing |

## Results {#results}

| Metric | Value | Note |
|---|---|---|
| Identity: real-real same-speaker pairs judged same | median 100.0% (IQR 85.7-100.0) |  |
| Identity: real-clone pairs judged same speaker | median 83.3% (IQR 69.2-100.0) | abstract: about 80% of the time; 26.6% of listeners said 'same' on every real-clone trial |
| Identity: different real speakers judged same | median 11.1% |  |
| Naturalness: real clips correctly classified | 67.4% (SD 14.8) |  |
| Naturalness: AI clones correctly classified | 60.8% (SD 16.7) | 21.0% of listeners at or below chance on clones |

## Moderators {#moderators}

- Exploratory follow-up: unscripted clips easier than scripted (median 76.7% vs 56.7%)

## Limitations {#limitations}

- Unfamiliar speakers
- Single commercial cloning service
- Short clips, online listening

## Cited in {#cited-in}

- [MR-0001](/metareports/human-detection-accuracy.md): Human Detection of Deepfakes: Accuracy, Bias, Confidence, and Training
- [MR-0003](/metareports/voice-cloning-human-detection.md): Voice Clones and Synthetic Speech: How Well Do People Detect Them?

## Verification {#verification}

| Field | Value |
|---|---|
| Status | `verified-fulltext` |
| Compiler confidence | high |
| Checked on | 2026-10-12 |
| Via | nature.com full text, doi:10.1038/s41598-025-94170-3 |