We propose an unsupervised adaptation framework, Self-T Aught Recognizer (ST AR), which leverages unlabeled data to enhance the robustness of automatic speech recognition (ASR) systems in diverse target domains, such as noise and accents. ST AR is developed for prevalent speech foundation models based on Transformer-related architecture with auto-regressive decoding (e.g., Whisper, Canary; SeamlessM4T).

We have conducted the experiments of replacing proposal generator, including MaskFormer [3] and RPN in Mask R-CNN combined with class-agnostic segmentation head [6, 7] (denote as RPN+Seghead). We also conduct the results for generating different numbers of proposals (N) with Mask2Former. Note that the original setting of MicroSeg is Mask2Former (N = 100).

For class incremental semanticsegmentation, suchaphenomenon oftenbecomesmuchworseduetothe background shift,i.e., some concepts learned at previous stages are assigned to the background class at the current training stage, therefore, significantly reducing the performance of these old concepts. To address this issue, we propose a simple yet effective method in this paper, namedMining unseenClasses via RegionalObjectness forSegmentation (MicroSeg).

Let Cat( ),Bernoulli( )denotethe parameter (2 R1 C 0 ).

Here, we further conduct an experiment on the MPII validation set to demonstrate this.