Improvement of automatic speech recognition models in regards to the recognition of proper names through the creation of a named entity dataset via speech synthesis

• Grant number: 25/06244-6
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: August 01, 2025
• End date: July 31, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Principal Investigator: Sandra Maria Aluísio
• Grantee: Rodrigo de Freitas Lima
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Company: Universidade de São Paulo (USP). Centro de Inovação da USP (INOVA)
• Associated research grant: 19/07665-4 - Center for Artificial Intelligence, AP.eScience.CPE

Abstract

In recent years, deep neural network approaches have been established as the most effective way to implement automatic speech recognition (ASR) systems, with all state-of-the-art models relying on this strategy. However, systems based on artificial neural networks (ANNs) still face challenges in recognizing named entities, especially in underrepresented languages such as Brazilian Portuguese. This scientific initiation project aims to improve the recognition of proper names by ASR systems for Brazilian Portuguese. To achieve this goal, the project proposes the use of speech synthesis tools to automatically generate audio datasets focused on named entities, thereby increasing the availability of such data for training automatic transcription models and enabling neural networks to better recognize proper names.