Heirloom Voices

Inspiration

Many grandparents leave years of voice notes on the family NAS, often in Vietnamese, next to tens of thousands of photos. The grandchildren read English, nobody has time to transcribe and find the matching pictures, and uploading a parent's voice and the family's faces to the cloud feels wrong.

What it does

Heirloom Voices turns one person's existing voice memos into a private story book on the UGREEN AI NAS. It transcribes the recordings, splits them into stories, and keeps up to three word-for-word quotes per story with a source tag such as [memo_0412.m4a 02:15]. It finds candidate photos by person, year, place and text in the photo, checks each one with image understanding, and labels it high or maybe with reasons. Stories with no match stay "voice only". After review, and only with separate approvals, it saves chapters and an index, creates one album per story, translates quotes, and tags the source recordings. Family members can then ask "What did Bà say about her shop?" and get the exact quote, its source and the photos.

How we built it

A single SKILL.md that follows the Uliya Skill Development Standard and delegates every NAS action to registered experts (file, knowledge, photo, document, media). Every capability was checked against the official capability catalog, and the package passes the official checker with 0 errors and 0 warnings.

Challenges

Making every delegation self-contained, keeping all writes behind approvals (translation, folder creation, albums, tag creation and tagging are approval-gated), and never inventing quotes or photo matches.

What's next

Round 2 testing on a real iDX6011 Pro with a fully synthetic test set: a fictional grandmother, six Vietnamese memos, 30 photos and 5 decoys, including a 1992 sign that must stay out of a 1972 story.

Disclosure

AI tools: Anthropic Claude helped research the rules and catalog, draft and revise the Skill files and proposal, run the official checker and proofread. Third-party components: none. External services and data flows: none; everything stays on the user's NAS.

Built With

• markdown
• ugreen
• uliya