The collections we are built to handle.

The challenge

Historical sound collections often span decades of format changes, recording conditions and documentation gaps. Legacy recordings may be degraded, unevenly preserved and recorded long before digital retrieval was a consideration.

Processing this material for digital access requires more than automated text extraction. It requires traceable handling, documented provenance and outputs suitable for institutional archival workflows.

What R2 Mechanics provides

• Offline processing — source material is not sent to public cloud transcription services by default
• Speaker-attributed, time-stamped transcripts with chapter-level navigation
• Structured outputs designed for long-term archival use
• Documented processing history for each collection item
• Separate forensic baseline and optional structured reading version

The challenge

Academic and oral history recordings present distinctive analytical demands. Speaker changes carry interpretive significance. The distinction between the spoken record and later interpretation matters. Timestamps must be reliable enough to cite.

Researchers need a transcript that preserves the structure and evidence of the original — not one optimised for reading smoothness at the cost of fidelity.

What R2 Mechanics provides

• Speaker attribution handled as a separate, dedicated analysis
• Time-aligned transcripts suitable for citation and qualitative analysis
• Separate forensic and editorial versions — fidelity preserved alongside readability
• Structured HTML output for searchable, navigable review
• Export formats for research workflows (DOCX, JSON, Markdown)

The challenge

Hearings, institutional interviews, court proceedings and sensitive historical recordings require processing under conditions for which public cloud transcription services may be inappropriate. Confidentiality, local processing and documented handling are not optional — they are the baseline requirement.

What R2 Mechanics provides

• Processing on our own infrastructure — not via third-party AI transcription services by default
• Written project scope covering material handling, output and retention conditions
• Traceable, versioned and reproducible results
• Operator-reviewed selection rather than automatic finalisation
• Outputs structured for institutional use and long-term record-keeping

The challenge

Long interview material, archival recordings and multi-source productions require transcripts that support editorial work — not raw text that requires manual reformatting.

Chapter navigation, speaker identification and searchability are not conveniences. For investigative and documentary production, they determine whether a 90-minute recording can actually be used efficiently.

What R2 Mechanics provides

• Speaker-labelled, chapter-structured transcripts with reliable timestamps
• Interactive HTML output navigable by chapter, speaker and full-text search
• Export formats for editorial workflows (DOCX, Markdown, structured JSON)
• Separate forensic and production-adapted versions
• Long-form and multi-source recordings treated as normal project conditions, not edge cases.

The challenge

Recordings where languages change between speakers, between passages or within a single conversation present a compound problem that standard transcription tools are not designed to solve.

Many standard multilingual transcription systems perform best when language changes are relatively predictable and recording conditions remain consistent. They are less well suited to the combination of language change, speaker change, inconsistent recording conditions, code-switching and archival source quality found in real collections.

The problem is not recognising individual languages. It is maintaining speaker continuity, language attribution and time alignment across a recording that behaves like an unpredictable source — because real archival material rarely behaves predictably.

What R2 Mechanics provides

• Processing strategies adapted to the specific language and speaker conditions of your material
• Speaker structure and language attribution are treated as separate, reviewable dimensions.
• Documented handling of code-switching and language transitions
• Project-specific assessment of language conditions before processing begins
• Reviewable outputs that preserve language and speaker context throughout