Related Experiment Video
Updated: Jul 4, 2026
09:09
Foreign Accent and Forensic Speaker Identification in Voice Lineups: The Influence of Acoustic Features Based on Prosody
Published on: September 27, 2024
Offline recognition of unconstrained handwritten texts using HMMs and statistical language models
Alessandro Vinciarelli1, Samy Bengio, Horst Bunke
1Dalle Molle Institute for Perceptual Artificial Intelligence, Rue du Simplon 4, 1920 Martigny, Switzerland. vincia@idiap.ch
Summary
This study introduces a system for recognizing unconstrained handwritten English text, significantly improving accuracy with statistical language models. The system reduces error rates by up to 50% for single-writer data and 25% for multiple-writer data.
Area of Science:
- Computer Science
- Artificial Intelligence
- Pattern Recognition
Background:
- Large vocabulary unconstrained handwritten text recognition is challenging.
- Existing systems often struggle with variability in handwriting styles and lexicon size.
Purpose of the Study:
- To develop and evaluate a system for offline recognition of large vocabulary unconstrained handwritten English text.
- To demonstrate the effectiveness of Statistical Language Models (SLMs) in improving recognition accuracy.
Main Methods:
- Developed a recognition system for unconstrained handwritten English text.
- Applied Statistical Language Models to enhance performance.
- Conducted experiments with single and multiple writer data using lexica of varying sizes (10,000-50,000 words).
Main Results:
- The integration of language models significantly improved system accuracy.
- With a 50,000-word lexicon, error rates decreased by approximately 50% for single-writer data.
- Error rates decreased by approximately 25% for multiple-writer data.
Conclusions:
- Statistical Language Models are crucial for improving the accuracy of handwritten text recognition systems.
- The proposed system and experimental setup offer a robust approach for unconstrained handwritten text recognition.