The recent large-scale text-to-speech (TTS) systems are usually grouped as autoregressive and non-autoregressive systems. The autoregressive systems implicitly model duration but exhibit certain deficiencies in robustness and lack of duration controllability. Non-autoregressive systems require explicit alignment information between text and speech during training and predict durations for linguistic units (e.g. phone), which may compromise their naturalness. In this paper, we introduce $\textbf{Mask}$ed $\textbf{G}$enerative $\textbf{C}$odec $\textbf{T}$ransformer (MaskGCT), a fully non-autoregressive TTS model that eliminates the need for explicit alignment information between text and speech supervision, as well as phone-level duration prediction. MaskGCT is a two-stage model: in the first stage, the model uses text to predict semantic tokens extracted from a speech self-supervised learning (SSL) model, and in the second stage, the model predicts acoustic tokens conditioned on these semantic tokens. MaskGCT follows the mask-and-predict learning paradigm. During training, MaskGCT learns to predict masked semantic or acoustic tokens based on given conditions and prompts. During inference, the model generates tokens of a specified length in a parallel manner. Experiments with 100K hours of in-the-wild speech demonstrate that MaskGCT outperforms the current state-of-the-art zero-shot TTS systems in terms of quality, similarity, and intelligibility. Audio samples are available at https://maskgct.github.io/. We release our code and model checkpoints at https://github.com/open-mmlab/Amphion/blob/main/models/tts/maskgct.

Muhammad Arslan Manzoor, Ruihong Zeng, Dilshod Azizov, Preslav Nakov, Shangsong Liang. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Graph neural networks (GNNs) have gained significant attention for their impressive results on different graph-based tasks. The essential mechanism of GNNs is the message-passing framework, whereby node representations are aggregated from local neighborhoods. Recently, Transformer-based GNNs have been introduced to learn the long-range dependencies, enhancing performance. However, their quadratic computational complexity, due to the attention computation, has constrained their applicability on large-scale graphs. To address this issue, we propose MGIGNN ( M emorized G lobal I nformation G raph N eural N etwork), an innovative approach that leverages memorized global information to enhance existing GNNs in both transductive and inductive scenarios. Specifically, MGIGNN captures long-range dependencies by identifying and incorporating global similar nodes, which are defined as nodes exhibiting similar features, structural patterns and label information within a graph. To alleviate the computational overhead associated with computing embeddings for all nodes, we introduce an external memory module to facilitate the retrieval of embeddings and optimize performance on large graphs. To enhance the memory-efficiency, MGIGNN selectively retrieves global similar nodes from a small set of candidate nodes. These candidate nodes are selected from the training nodes based on a sparse node selection distribution with a Dirichlet prior. This selecting approach not only reduces the memory size required but also ensures efficient utilization of computational resources. Through comprehensive experiments conducted on ten widely-used and real-world datasets, including seven homogeneous datasets and three heterogeneous datasets, we demonstrate that our MGIGNN can generally improve the performance of existing GNNs on node classification tasks under both inductive and transductive settings.