Inspiration
Family voice recordings and digitised cassettes often sit unheard beside thousands of disconnected photos. Linking stories to images manually means hours of listening and searching.
What it does
Spoken Memories Album turns a family voice recording into illustrated story albums. It transcribes the recording, identifies each story and the people, places and years mentioned, finds candidate photos on the NAS, and creates one album per story plus a Story Book containing timestamped verbatim quotes. Unmatched names and undated stories are surfaced rather than invented.
How it is built
This is a Uliya orchestration Skill for UGREEN AI NAS, developed with the UGOS Skill Creation Tool. Ten steps delegate to four registered experts: file-agent (ug-filesystem), knowledge-agent (ug-knowledge), document-agent (ug-document), and photo-agent (ug-photo). It combines file search, audio transcription, document extraction, named faces, photo search, local vision re-ranking, and album creation. Each step passes verified upstream results. The user confirms candidate photos before albums are created. Transcript, album and Story Book writes go through workflow approval; nothing is deleted, moved, tagged or shared. Models run locally on the NAS, with no external data flows.
Development status and challenges
Implemented: SKILL.md with ten-step orchestration, all five delegation states, empty and partial results, and four references. It passes the official uliya-skill-kit 0.7.3 static checker with zero errors and warnings. Live NAS validation is still pending for transcription quality, OCR on old scans, and album item format. Transcript translation is not implemented. Accuracy depends on language, accent, named faces and photo metadata; long recordings may need splitting.
Validation plan and next steps
Test an English recording naming two library people, a city and a year; verify correct albums and verbatim quotes. Test scanned prints discoverable only through caption OCR. Test vague dates, duplicate first names and cancelled approval: the Skill must ask rather than guess and record an album as not created when approval is cancelled.
Track and disclosure
Understand Your Life. Individual project from Georgia. The package contains original Markdown files and uses documented UGREEN AI NAS capabilities, with no external services, APIs, libraries, models or datasets. AI assistance was used for drafting and checking.
Built With
- ai-assistant
- ugos-skill-creation-tool
- ugreen-ai-nas
