image001Seba — a Personal Museum Guide

What is it?

People often come to museums seeking cultural knowledge.  Being immersed in a quiet room filled with thematically connected exhibits can foster a level of engagement that helps learning.  However, many gallery visitors today simply glide past the exhibits, barely viewing each for a few seconds.  When they are attracted to an exhibit, they find very little material that explains the exhibit's significance and details.  Each visitor would have different questions about the exhibits, depending on their level of knowledge and interest. The limited description typically does not provide answers. A trained tour guide could help by answering questions, while tuning their responses to each visitor's level.  Different responses are needed for a child vs. an adult, or a knowledgeable person vs. a novice.  Unfortunately, it is not practical to assign a human guide for each visitor.

Seba is a personalized AI based guide that addresses this need.  It is knowledgeable about all the exhibits in the gallery and can verbally converse with each visitor to explain and answer their unique questions.  It learns about each visitor’s interests and presents information in a way that makes it interesting.  Seba runs as a web page on the visitor's phone – no app to install.  For training, it is given details of each exhibit, similar to the text that may be published at the venue's website.  This material is used to generate a short introduction to the exhibit, to suggest follow up topics and to answer specific questions.  The process starts with Seba identifying the exhibit.  Currently, this is done by asking the visitor to aim their phone camera at the exhibit.  Seba uses the image to identify the exhibit and start the conversation.

Web App

An advanced prototype has been built.  It is set up with more than 100 exhibits at the Asian Art Collection at the Princeton University Art Museum.  More exhibits can be included by simply adding their descriptions.  Other venues can be set up just as easily.

SebaStarRot20Seba Visitor’s Guide — Click to the left to see how to use this App.

Prototype — A web app implements a personal guide to the Asian Art Collection at the Princeton University Art Museum.

On your mobile phone, scan or click on the QRC to the left or on “Prototype”.  An app will appear. Touch the yellow button labeled “Touch to Scan”, frame the exhibit, then touch the “Take photo” button.  This tells the app that you wish to discuss that exhibit.  The app will audibly describe the exhibit.  On iOS, you may need to press the “Press to hear” button.  You can ask questions verbally by touching the “Touch to Talk” button.  For more instructions, see the Seba Visitor’s Guide.

To try out the app at a site away from the museum, you will need a few pictures of exhibits.   Using a computer (or another phone), click on the following to display a list of images.  Seba can scan these images.

Exhibit4Exhibit8Exhibit9Exhibit1

Selected Exhibit Images — A few pictures of exhibits that can be used to demonstrate the app.

Gallery Director’s Guide — What is involved in setting up Seba for a Gallery.

Exhibit Manager’s Guide — How to add and update exhibits using the Management app below.

Exhibit Management app — Feel free to make changes, everything will be restored each night.

POCMed Another venue – the PUAM collection of Ancient Mediterranean Art (but only 10 exhibits)

Details

Introduction — Read more about what it is.

Sample Conversation — A sample transcript of a visitor’s session.

Sample Exhibit Description — A sample of training data.

Architecture — How it works, for techies.

Tech Note: Exhibit Recognition — How Seba uses image matching

Tech Note: Model Selection — How Seba dynamically picks an AI model

Project Plan — What next…

Seba vs. Musa — Compare functionalities

Seba vs. Smartify — Compare functionalities.

Tests (for techies only)

Text recognition – Scan image for text.

Audio — Record and play back speech.

Text to Speech — Multilanguage, but needs translation API.

Speech to Text — Multilanguage.

Identify Image – Image match tests.

Exhibit Management app —Needs admin key.

POC1 —Exhibits – Old version, uses OCR to read exhibit descriptions