ABSTRACT In real-world applications, users often require both translations and transcriptions of speech to enhance their comprehension, particularly in streaming scenarios where incremental generation is necessary. This paper introduces a streaming Transformer-Transducer that jointly generates automatic Figure 1. To produce ASR and ST content effectively with minimal latency, we propose a joint token-level serialized output training method that interleaves source and target while incrementally receiving additional speech content. Experiments particular, only Weller et al., 2021 [10] proposed a unifieddecoder in monolingual (it-en) and multilingual ({de,es,it}- solution for real-time applications that, however, en) settings demonstrate that our approach achieves the best leverages a fully attention-based encoder-decoder (AED) architecture quality-latency balance. With an average ASR latency of 1s [11], which is theoretically not well suited for and ST latency of 1.3s, our model shows no degradation or the streaming scenario [12], and adopts the re-translation even improves output quality compared to separate ASR and approach [13], which is well-known to be affected by the ST models, yielding an average improvement of 1.1 WER and flickering problem [14].

Graph clustering is the process of grouping vertices into densely connected sets called clusters. We tailor two mathematical programming formulations from the literature, to this problem. In doing so, we obtain a heuristic approximation to the intra-cluster density maximization problem. We use two variations of a Boltzmann machine heuristic to obtain numerical solutions. For benchmarking purposes, we compare solution quality and computational performances to those obtained using a commercial solver, Gurobi. We also compare clustering quality to the clusters obtained using the popular Louvain modularity maximization method. Our initial results clearly demonstrate the superiority of our problem formulations. They also establish the superiority of the Boltzmann machine over the traditional exact solver. In the case of smaller less complex graphs, Boltzmann machines provide the same solutions as Gurobi, but with solution times that are orders of magnitude lower. In the case of larger and more complex graphs, Gurobi fails to return meaningful results within a reasonable time frame. Finally, we also note that both our clustering formulations, the distance minimization and $K$-medoids, yield clusters of superior quality to those obtained with the Louvain algorithm.