# Original synthetic rendering examples

Created for TheMomentra's SEO implementation on 27 October 2026. The text, source title-board graphics and synthetic examples are original demonstration material. Speech was generated locally with the open-source eSpeak NG 1.52.0 speech synthesizer. No sampled customer voice, customer media, paid model API or Apple system voice was used. eSpeak NG is GPL-3.0-or-later; its program/voice files are not redistributed in this directory. Tool/source: https://github.com/espeak-ng/espeak-ng.

Hindi and Marathi audio was synthesized separately for each sentence, then assembled. Captions were supplied from those scripts and phrase durations. Hinglish spellings were supplied as demonstration text. These are not automatic-transcription, word-alignment, translation, diarization or speaker-tracking test results. Speech pronunciation and your own source material still require human review. Do not claim a measured accuracy percentage from these examples.

The actual local TheMomentra FFmpeg renderer produced the vertical outputs. `manifest.json` records source text, edit parameters, file hashes and native dimensions/duration. A full landscape frame is intentionally fitted into a vertical canvas; no face tracking is claimed. VTT files contain phrase timings for accessibility. The frame images are extracted from the corresponding output files.

The Devanagari font is Noto Sans Devanagari, under the SIL Open Font License; its font and license are in the repository's assets/fonts directory. Latin render fonts have their own existing OFL notices there.

Reproduce with `scripts/build-seo-demos.ts`, eSpeak NG and the FFmpeg/ffprobe versions configured for your local environment. The script writes only these example assets and a temporary working directory; it does not upload, spend credits or publish anything.

The separate existing NASA podcast demo remains under `/demo/` and has its own provenance in that directory's README. NASA is not affiliated with or an endorser of TheMomentra.

The screen example uses three captured states of the actual Hinglish converter component, with enlarged recording typography and original input text. These are assembled into a narrated source video; they are not a continuous interaction recording. Narration is synthetic English, captions are supplied from its script, and the render uses full-frame fitting with no presenter/tracking. See recording/capture.json for the actual input/output pairs.