How I Helped My Daughter Learn to Read with Machine Learning
summary
A web app was built to help children practice reading by classifying spoken syllables: it uses the Speech Commands dataset plus 30–50 child recordings per ~30 target syllables, performs audio augmentation and converts waveforms to spectrograms. The pipeline fine-tunes a TensorFlow model in Python and exports it to TensorFlow.js for real-time, offline syllable-confidence feedback in the browser; results are promising but need more data and tuning.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 2:22Genesis of the idea
- 5:21Definition of "ready"
- 6:20Learning to read
- 8:21Syllabic fusion
- 9:37Richness of French
- 11:59AI at Orange
- 13:09Machine learning
- 17:07Speech-to-text at Orange
- 20:58Model building
- 22:16Transfer learning
- 23:38Sound classification
- 27:19Data collection
- 28:49Data preparation
- 30:04Model training
- 33:44Model export
- 34:30Importing the model into JavaScript
- 35:19Using the model in JavaScript
- 38:29Demo: word syllable by syllable
- 39:47Perspectives
related talks
51mHow I Helped My Daughter Learn to Read with Machine Learning
28mHow to use Deep Learning to process audio data and classify sounds.
11mVOICE Summit 2019 | VED Talks | Melissa Hammersley | July 23
20mExploring the hidden potential of sound data
31m“Sorry, Could you repeat that again?” - Speech Recognition with Python
01h 14mExploring the hidden potential of sound data
25mLearning tools for Hearing impaired children using OS web applications
22mMachine learning in Education