The pipeline provides a fully open and modular approach, with a focus on leveraging models available through the Transformers library on the Hugging Face hub. The code is designed for easy ...

Abstract: Recently, audio-visual speech recognition has attracted increasing attention. However, most existing works only focused on scenarios with two speakers. In this work, we study the effect of ...

Abstract: This letter presents a new target speech recognition problem, where the target speech is defined by a keyword. For instance, when a person speaks “Hey Google” or “Help Me”, we hope the model ...