Evidence · first-party tested

On noisy source audio, one generated variant was only about 20% similar to the original voice and showed a significant deviation in tone and vocal characteristics, so the clone could miss the speaker identity badly on a weak output.

✗ Failed20Heygen
Output — unretouched
tool output
Output — unretouched
tool output
The observation

On noisy source audio, one generated variant was only about 20% similar to the original voice and showed a significant deviation in tone and vocal characteristics, so the clone could miss the speaker identity badly on a weak output.

Criterion: Voice match accuracy · Scenario: Low-Quality Voice Sample (group: voice-cloning-inputs)
Query this
get_evidence({
  tool: "heygen",
  scenario: "voice-cloning-inputs"
})
MCP · mcp.aidemos.com/api/mcp
Free with attribution.
Real outputs, no retouching · every cell queryable via API & MCP · aidemos.com