I want to know, is the encoder shared between ASR and AST task and decoder shared between MT and AST task when we run the LibriSpeech Multitask recipe. Or the ASR, AST and MT tasks use different encoders and decoders and update the parameters turn by turn.
I want to know, is the encoder shared between ASR and AST task and decoder shared between MT and AST task when we run the LibriSpeech Multitask recipe. Or the ASR, AST and MT tasks use different encoders and decoders and update the parameters turn by turn.