Synthetic media forensics glossary

Which questions belong in this field?

Begin by stating the proposition. “This recording contains generated speech” differs from “this person authorized the recording” and from “this file came from this service.” Each proposition calls for different evidence and can reach a different level of support.

Identify the modality and the actual object under examination. A screenshot of a video, a compressed messaging copy, and an original camera file are different examination inputs. Findings about one should not silently become findings about the others.

Evidence sources and examination context

Preserve the received file and document how it was obtained. Look for available originals, editing records, related messages, and provenance information. Inspect container metadata and encoding history with tools whose behavior you understand.

Use detection results as measurements with a defined scope. Record the detector version, input preparation, operating threshold, and relevant validation conditions. NIST’s synthetic-content overview describes detection, watermarking, and provenance as different approaches; none supplies a universal answer to every authenticity question. NIST AI 100-4: Reducing Risks Posed by Synthetic Content.

Consider transformations such as cropping, re-encoding, resizing, and playback recording. Compare test conditions with the case object rather than treating a published benchmark as a guarantee for a redistributed copy.

A worked evidence example

Hypothetical example: two copies of a voice clip produce different detector scores. One is the received attachment; the other has been converted for analysis. The immediate result is a disagreement between measurements on different inputs, not proof that either score is the true probability of synthesis.

Retain both objects, record the conversion, and examine whether the tool supports the relevant format and conditions. The report can explain the disagreement and still leave the source of the voice unresolved.

What this evidence can and cannot establish

A method can support a bounded finding about observed characteristics or documented provenance. A detector label does not establish the identity of a creator, the truth of the scene, or a complete editing history.

Missing Content Credentials do not establish that media is generated. Valid credentials do not establish that a depicted event happened. Keep media assessment, provenance validation, and factual corroboration connected but distinct.

Essential terminology

• Synthetic media — Media generated or modified through computational synthesis, including AI-generated or AI-assisted image, audio, video, and text.
• Deepfake — A commonly used term for synthetic or manipulated media representing a person or event, often through AI-based generation or alteration.
• Audio spoofing — Presentation of manipulated, replayed, or generated audio to misrepresent a source or defeat an audio check.
• Watermark — A scheme-dependent signal embedded in or associated with content to support a particular identification or labeling claim.
• False positive — A positive classification when the tested proposition is absent, measured relative to defined ground truth and evaluation conditions.
• Lossy compression — Compression that discards some information according to an encoding method.
• Transcoding — Conversion of media to another encoding or container, potentially changing metadata and method-specific signals.
• Detection threshold — A chosen decision boundary used to classify a measurement or score.

All 49 terms in this evidence field

• Audio spoofing
• Audio synthesis
• Bit depth
• Codec
• Deepfake
• Detection threshold
• Diffusion model
• EXIF metadata
• Face reenactment
• Face swap
• False negative
• False positive
• Frame rate
• Generation seed
• Generative adversarial network
• Generative AI
• Guidance scale
• Image synthesis
• Image-to-image generation
• Inpainting
• JPEG quantization
• Keyframe
• Latent representation
• Lip synchronization
• Lossless compression
• Lossy compression
• Media lineage
• Multimodal
• Negative prompt
• Outpainting
• Perceptual hash
• Replay attack
• Sample rate
• Sampler
• Sampling steps
• Speaker verification
• Spectrogram
• Speech-to-speech conversion
• Super-resolution
• Synthetic media
• Synthetic media forensics
• Temporal artifact
• Text-to-image generation
• Text-to-speech
• Transcoding
• Upscaling
• Voice conversion
• Watermark
• Waveform

Continue with an examination reference

• ai detection limitations
• metadata
• synthetic media
• synthetic media forensics

Citation and editorial context

This is an editorial reference. Source checks are separate from independent expert review; no independent reviewer is recorded.

Examples are hypothetical. See the editorial policy, methodology, and correction process.

Shelley, Jesse (ed.). “Synthetic media forensics glossary.” Synthetic Forensics, 1.1.0-glossary-library-1, 2026-10-09. https://syntheticforensics.org/glossary/media/ Accessed: add the date you consulted this page.

The complete vocabulary

Browse the full glossary of 300 terms or download the machine-readable vocabulary.

Field assignment is the project’s organizational taxonomy; related evidence can span several fields.