Inspiration
We wanted to make it easier to find something you remember hearing in a meeting, interview, or voice note stored on your UGREEN NAS.
What it does
Spoken Audio Search lets you search a folder of recordings by exact phrase or by describing what you remember. It returns the recording name, a short quote, and a playback timestamp when available.
How we built it
We built a Uliya skill that transcribes recordings on the NAS and uses Uliya’s local model to match your query against the transcripts. You can save transcripts in a library so future searches only need to transcribe new or changed recordings.
Challenges we ran into
We had to handle recordings that couldn’t be transcribed, segments with missing timestamps, and matches where the speaker used different words from the search query. The skill reports which recordings it could search and what it had to skip.
Accomplishments that we're proud of
The skill passed static validation. It can start with a folder of audio and no existing transcripts, and it shows the original wording so users can judge whether a match is relevant.
What we learned
Context matters when searching speech. A few nearby sentences can tell you whether someone proposed moving a deadline or whether the team actually agreed to move it.
What's next for NAS Scout — Spoken Audio Search
We’ll test the skill on a live UGREEN NAS, measure transcription and search times on larger audio libraries, and check how reliably it finds relevant moments when users search in their own words.
Built With
- markdown
- uliya