Power-Up Your App With Apple’s Machine Learning Frameworks
summary
Shows how Apple frameworks chain together: Speech framework (1,000 requests/device/hour, 1 minute max) transcribes audio, NSLinguisticTagger stems and tags, then Core ML runs models bundled with the app and can offload to GPU but Core ML bundles pre-trained models and cannot train incrementally on-device. Prototype uses Keras -> Core ML export: bag‑of‑words MLP and an LSTM for joint intent + entity tagging, yields ~99.5% on seen data and ~98% on unseen, and drives a working voice CRUD demo (table add 0.99, record intent 0.87).
AI-written summary · about
chapters (28)
- 0:00About Me
- 1:07What is a Conversational UI?
- 1:19Can We Build Our Own Siri?
- 2:46The Ghost In The Shell
- 3:21Some of the Parts
- 5:03The SpeechFramework
- 5:44Permissions
- 6:27Start Your Engine
- 6:46The Good and the Not-So
- 7:55Code...
- 8:21Detect Language
- 8:36Extract Word Stems
- 8:58What Else Can You Extract?
- 10:35Bag of What??
- 11:12What're The Bags Saying?
- 11:47Histogram of Words
- 13:13Training Utterances & Intents
- 14:00Converting to BoW inside Python
- 15:46Using CoreML in Swift
- 17:04The Parts One Last Time
- 17:23All The Abbreviations...
- 18:12Have Fun Tagging
- 18:59How LSTM Works
- 19:57Let's Open The Box
- 20:29Training Data
- 22:19Overall Design
- 28:30I'm Almost Done!!!
- 30:03The End!
related talks
22mCore ML and Your App
25mGetting Started with Machine Learning in your iOS apps
17mExploring Natural Language Processing
23mWhy MLX
40mNavigating the Machine Learning Landscape for Apple Devices
34mVoice Search Using NLP & CoreML
26mThe Blue Ocean of Machine Learning: Core ML
47mBuilding smarter apps with Machine Learning, from magic to reality