Conformer Transducer
Conformer Transducer models are a leading architecture in automatic speech recognition (ASR), aiming to improve accuracy and efficiency, particularly in challenging scenarios like child speech or streaming applications. Current research focuses on enhancing these models through techniques like incorporating contextual information (e.g., from previous turns in a conversation or language models), addressing limitations in streaming performance via improved normalization methods, and developing model compression strategies for resource-constrained devices. These advancements hold significant promise for improving the accuracy and accessibility of ASR systems across diverse applications and hardware platforms.
Papers
February 9, 2024
January 14, 2024
November 7, 2023
July 20, 2023