← All published sample evidence

Obama-labeled voice sample pair

These are two public demonstration WAVs labeled locally as synthetic and human speech. Their filenames and labels do not establish that either recording depicts Obama, came from a particular event, or is an authenticated original. The source registry has no pinned upstream record for either file.

The figures below are uncalibrated Worker decision scores, not probabilities or evidence of speaker identity. A score at or above 0.5 triggers the synthetic verdict for this frozen detector. Two selected examples cannot estimate real-world accuracy.

Locally labeled synthetic speech

File
deepfake_obama_clone.wav
SHA-256
c5728684ac685c2484089e3ca6c3b60e7ef89c10ba42c0753194efff894cf8eb
Worker verdict
Synthetic detected (technical enum: SYNTHETIC_DETECTED)
Decision score
0.9933 at the fixed 0.5 threshold

Source status: unverified upstream record. The label is a local assertion, not independently authenticated provenance.

Locally labeled human speech

File
human_real_obama_speech.wav
SHA-256
0350bfe30ed2a6a4ce37f19eeac746f5f382fdbdae69234a8daa53dcfb52861c
Worker verdict
Human-like signal (technical enum: AUTHENTIC_HUMAN)
Decision score
0.0219 at the fixed 0.5 threshold

Source status: unverified upstream record. The label is a local assertion, not independently authenticated provenance.

Verify the published evidence

Download the WAVs and compare their SHA-256 digests above with the 29-sample registry. The Worker score ledger records the displayed decisions. Release provenance links the exact public DSP source snapshot and its hash. The selected corpus report shows the detector's wider limitations.

Neither the local label nor a detector score verifies a public figure's identity or the origin of an alleged recording.