Catalog a collection with object photos and transcribed voice memos
At a glance
A Lobsters contributor builds a phone-friendly art-glass catalog that turns each object's spoken description into text with whisper.cpp. This workflow is listed under Personal collections, Transcription, Background jobs.
- Agents / tools
- whisper.cpp, Rails, Amazon S3, Turbo Streams
SourceBuilding a tool for these workflows? See weekly rates.
What they did
1) Prepare a Rails application with photo and voice-memo capture, Amazon S3 upload access, a background-job runner, and a working whisper.cpp installation on the worker host. This is a custom application to build, not a ready-to-install catalog. 2) Create an item record from a smartphone, photographing one object and recording its description as a short voice memo. 3) Upload the recording to S3 and trigger a background job that invokes whisper.cpp to transcribe it. Return the transcript to the item's page through Turbo Streams when processing finishes. 4) Have the catalog administrator read the text and add structured item details. Retain audio playback to resolve ambiguous transcriptions.
How it went
The contributor reports CPU-based operation on a small DigitalOcean droplet, with transcripts appearing in roughly 5–10 seconds in production.
Worth knowing
The reported workload involved short recordings and only a couple of users; imperfect transcripts were acceptable because the originals remained playable.