The autoregressive (AR) models, such as attention-based encoder-decoder models and RNN-Transducer, have achieved great success in speech recognition. They predict the output sequence conditioned on ...

The auditory selection framework with attention and memory (ASAM), which has an attention mechanism, embedding generator, generated embedding array, and life-long memory, is used to deal with mixed ...