Building accessible experiences with AI

on YouTube ↗46m

" summary

Demonstrates concrete AI components that make interfaces accessible: Azure Custom Vision for training classifiers (recommendation: about 50 images per class, object detection used ~545 images for Simpsons), speech-to-text transcription and neural text-to-speech for real-time translation, Computer Vision/Face APIs for descriptions, bounding boxes, celebrity or custom face identification, and OCR including handwriting. Shows a live pipeline with Azure Media Services + Video Analyzer, produces captions and editable transcripts, and notes a streaming cost example of €4/hour streaming and measured prediction confidences like ~64–90%.

AI-written summary · about

" chapters (15)

1. 0:00Introduction to talk
2. 1:07Live translation demo setup
3. 2:30Why accessibility matters
4. 8:16Microsoft AI services overview
5. 9:18Computer vision demo start
6. 11:48Custom vision training demo
7. 17:53Object detection with custom vision
8. 20:51Face API demo and training
9. 23:39Text and handwriting recognition
10. 25:48Speech recognition and neural voices
11. 30:19Azure Media Services for captions
12. 35:42Video Analyzer for media demos
13. 38:07Seeing AI app overview
14. 38:53Key accessibility questions
15. 42:26Q&A session

" related talks

• 44mThe age of applied AIHenk Boelman · Build Stuff 2020
• 57mEasier AI allows everyone to build smart appsDavid Burela, Azadeh Khojandi · NDC Sydney 2018
• 53mAI for the betterDennie Declercq · NDC Sydney 2020
• 01h 00m5 Levels of AI in Microsoft AzureAlan Smith · CloudBurst 2022
• 01h 03mEasier AI allows everyone to build smart appsDavid Burela, Azadeh Khojandi · NDC London 2018
• 01h 05mMicrosoft Cognitive Services: Making AI EasyJennifer Marsman · NDC London 2017
• 01h 10mSolutions prototyping using Azure AI StudioPawel Ekk-Cierniakowski · DSC Europe 24
• 47mAccessibility in the era of generative AIPaulette Lee, Marco Casalaina, Ronak Chokshi, Ailsa Leen, Josh Loebner, Neelay Thaker · Microsoft Build 2024