# Meridian account media · v1

One synthetic narration reads the eight complete paragraphs of Observer A's retained first written account, verbatim. It is not an interview, original voice, new account or independent witness. The two pseudonymous witnesses and three original account records remain unchanged. Audio time starts at the beginning of the generated reading, not the field session.

The authored audio uses the installed macOS Samantha voice at 165 words per minute. Each paragraph is synthesized separately, converted with FFmpeg to mono 16-bit PCM at 16 kHz, and joined with 250 ms silence between paragraphs. Cue boundaries come from exact PCM frame counts, with a retained JSON transcript and millisecond WebVTT projection. Every cue matches its original Markdown paragraph and line. Synthetic pronunciation is illustrative; the source text remains authoritative.

Three annotations on each retained account illustration connect a visible feature to an exact source paragraph. Pixel rectangles and zoom are application presentation metadata against the 720 × 440 authored SVG, not standard measurement selectors. The drawings are derived illustrations, not witness-made sketches, measured trajectories or physical reconstructions. Both illustrations keep their original hashes and processing records.

## Verify and rebuild

Keep the sibling `reader-r2` folder and use Node 22 with the built canonical checkout at `88bcbfcc8b17af30837313aff596cec8e1bde43a`:

```sh
node build.mjs /path/to/canonical
```

The verifier checks every retained hash, PCM duration, cue bounds, paragraph wording, annotation bounds and source paragraph mapping. It validates observation processing, archival entities and exact audio-time edition citations through the existing standard contracts. It reproduces the packet and viewer projection without requiring speech software or network access.

## Explicit authoring

```sh
python3 synthesize.py ../reader-r2/sources/account-a.md .
```

This requires macOS with Samantha installed and FFmpeg on PATH. Synthesized speech bytes can change with OS, voice or resampler versions; this is an authoring recipe, not a cross-platform bitwise reproduction promise. Changes require explicit artifact/manifest review. The packet build never silently synthesizes or rewrites its manifest. WAV is a browser playback example, not a general imported-audio adapter.

The generated audio and transcript have their own artifact identities, processing inputs and source edition. Original text citations remain original text citations. No field-event timestamp, historical recording date, authenticated custody or scientific validity is inferred from playback or integrity checks. No new standard contracts and no npm publication.

Fictional source text and authored annotation metadata: CC0-1.0. Example code: MIT. The synthetic reading is an illustrative speech rendering of that text and does not represent any real person's voice.
