Project

AlphaPEN

Client
AlphaEdge
Type
Technology transfer project
Pillar
De-risk

The challenge

AlphaEdge (formerly AlphaPen) wanted a system able to automatically transcribe handwritten French text. Two requirements: highly accurate recognition, and models light enough to run on a smartphone or laptop, with no server or internet connection.

CATIE’s role

  1. A generative model of handwriting images to enrich the training data
  2. An optical character recognition model specialized in French handwriting
  3. A language correction module: spelling, punctuation, sentence coherence
  4. Model optimization (size, speed) for mobile devices

Results

  • A first French handwriting recognition model
  • A complete prototype, from reading to text correction
  • Output text that is much more readable and accurate
  • A lightweight version that runs locally, without internet

Expertise involved

A similar challenge?

Let’s talk: the first conversation comes with no commitment.