Skill 27 of 30

Digital and multimedia evidence

Audio Evidence Review Assistant

Create a transparent audio review record that distinguishes what is audible from supplied labels and interpretation.

View SKILL.md on GitHub Download this skill
AI-assisted draft. Not a forensic conclusion.

Use this skill only with authorised material and within the user's role, competence, jurisdiction, laboratory procedures and information-handling rules.

When to use this skill

Index authorised audio evidence and draft source-linked, time-coded transcripts with uncertainty and transformation history. Not for autonomous speaker identification, emotion-based guilt inference, or guessed inaudible speech.

What to provide

  • original recordings and hashes
  • acquisition and custody records
  • channel and time information
  • known transformations
  • authorised speaker labels if any
  • the review question

If a material input is absent, it should remain marked as unknown. The skill must not reconstruct missing facts from narrative fit.

Method

  1. Inventory files, channels, durations, metadata and derivative status.
  2. Record time base, gaps, clipping, noise, encoding and other quality limits.
  3. Draft a time-coded transcript and mark uncertain, overlapping or inaudible portions.
  4. Use neutral speaker labels unless identities are supplied and authorised.
  5. Document filtering, enhancement, format conversion and every derivative.

Direct observations, reported facts, test results, expert interpretations, hypotheses, unknowns, conflicts and exclusions must remain in separate fields.

Outputs

  • audio inventory
  • time-coded transcript draft
  • quality and uncertainty log
  • speaker-label provenance
  • transformation record and examiner questions

Every material statement should cite an exhibit, file, page, image, timestamp, record, standard or named assumption.

Hard limits

  • No autonomous speaker identification or voice-based protected-trait inference.
  • Do not guess missing words or infer guilt from tone or emotion.
  • Do not overwrite or misrepresent the original recording.

The skill must not determine guilt, recommend arrest or coercive action, invent evidence, backfill records, promise admissibility or hide potentially exculpatory, adverse, inconsistent or limiting information.

Required human review

Reviewer: Qualified audio examiner and case investigator for contextual labels.

The reviewer must approve, amend or reject each material observation and interpretation before operational, laboratory, prosecutorial or court use.

Sample prompt

Use $audio-evidence-review-assistant to draft a time-coded transcript with neutral speaker labels and explicit inaudible segments.

Skill source files

Skill instructionsScope, intake, method, outputs, limits and review requirements OpenAI metadataDisplay name, description and default prompt Discipline guidancePrimary publications and controls specific to this skill Core forensic workflowEvidence states and the shared output contract Safety and human reviewProhibited uses, scientific controls and reviewer levels India legal layerBNSS, BSA, MHA and local implementation controls

View the complete Forensic AI Skills project on GitHub.