Active Shadowing

Active shadowing encompasses techniques where a system replicates or simulates the behavior of a native speaker listening to and repeating non-native speech, primarily for evaluating speech intelligibility in language learning or for manipulating perceptions in human-robot interaction. Current research focuses on using voice conversion and latent speech representations, particularly self-supervised speech representations, to create virtual shadowing systems that accurately reflect native speaker responses. These methods offer improvements over traditional automatic speech recognition approaches and provide valuable insights into speech assessment and human-computer interaction, with applications in language learning and robotics.