Text-to-Handwriting Generation Using Multi-Layer LSTM with Soft Window Attention Mechanism

🔎 Project Description

This project aims to develop an AI model that generates human-like handwriting from typed text. Based on the LSTM architecture proposed in Alex Graves' paper, the model learns to map character-level text input into continuous pen stroke sequences.

🎮 Demo Application

Try the live demo here:

• React + FastAPI Web App: Handwriting Generator Demo (Vercel)
  Type any text and see the handwriting appear on canvas in real-time

• Gradio Demo (alternative interface): Hugging Face Spaces
  Simple interface to test handwriting generation and model response

📄 Dataset

The IAM Online Handwriting Database contains pen stroke sequences and corresponding transcriptions from over 200 writers, used for handwriting generation and recognition tasks.

🔗 References

1. https://arxiv.org/abs/1308.0850
2. https://medium.com/geekculture/understanding-the-paper-generating-sequences-with-rnns-by-alex-graves-18635cdd32be
3. https://github.com/swechhasingh/Handwriting-synthesis
4. https://www.calligrapher.ai/
5. https://magenta.tensorflow.org/assets/sketch_rnn_demo/index.html