Multimodal Large Language Models (MLLMs) have achieved great success in Speech-to-Text Translation (S2TT) tasks. However, current research is constrained by two key challenges: language coverage and efficiency. Most of the popular S2TT datasets are substantially English-centric, which restricts the scaling-up of MLLMs' many-to-many translation capabilities. Moreover, the inference speed of MLLMs degrades dramatically when the speech is converted into long sequences (e.g., 750 tokens). To address these limitations, we propose a Multilingual Cost-effective Accelerated Speech-to-Text Translator (MCAT) framework, which includes two innovations. First, a language scaling method that leverages curriculum learning and a data balancing strategy is introduced to extend the language coverage supported by MLLMs to 70 languages and achieve mutual translation among these languages. Second, an optimized speech adapter module is designed to reduce the length of the speech sequence to only 30 tokens. Extensive experiments were conducted on MLLMs of different scales (9B and 27B). The experimental results demonstrate that MCAT not only surpasses state-of-the-art end-to-end models on the FLEURS dataset across 70x69 directions but also enhances batch inference efficiency. This is achieved with only ~100M trainable parameters and by using only 10 hours of S2TT data per language. Furthermore, we have released MCAT as open-source to promote the development of MLLMs for robust S2TT capabilities. The code and models are released at https://github.com/yxduir/m2m-70.

Abstract--Human-robot collaboration towards a shared goal requires robots to understand human action and interaction with the surrounding environment. This paper focuses on human-robot interaction (HRI) based on human-robot dialogue that relies on the robot action confirmation and action step generation using multimodal scene understanding. The state-of-the-art approach uses multimodal transformers to generate robot action steps aligned with robot action confirmation from a single clip showing a task composed of multiple micro steps. Although actions towards a long-horizon task depend on each other throughout an entire video, the current approaches mainly focus on clip-level processing and do not leverage long-context information. This paper proposes a long-context Q-former incorporating left and right context dependency in full videos. Furthermore, this paper proposes a text-conditioning approach to feed text embeddings directly into the LLM decoder to mitigate the high abstraction of the information in text by Q-former . Experiments with the Y ouCook2 corpus show that the accuracy of confirmation generation is a major factor in the performance of action planning. Furthermore, we demonstrate that the long-context Q-former improves the confirmation and action planning by integrating VideoLLaMA3.

Speech Large Language Models (SpeechLLMs) have achieved breakthroughs in multilingual speech-to-text translation (S2TT). However, existing approaches often overlook semantic commonalities across source languages, leading to biased translation performance. In this work, we propose \textbf{POTSA} (Parallel Optimal Transport for Speech Alignment), a new framework based on cross-lingual parallel speech pairs and Optimal Transport (OT), designed to bridge high- and low-resource translation gaps. First, we introduce a Bias Compensation module to coarsely align initial speech representations across languages. Second, we impose token-level OT constraints on a Q-Former using parallel speech pairs to establish fine-grained consistency of representations. Then, we apply a layer scheduling strategy to focus OT constraints on the most semantically beneficial layers. Experiments on the FLEURS dataset show that our method achieves SOTA performance, with +0.93 BLEU on average over five common languages and +5.05 BLEU on zero-shot languages, using only 10 hours of parallel speech per source language.

Large-scale pre-training and instruction tuning have been successful at creating general-purpose language models with broad competence. However, building general-purpose vision-language models is challenging due to the rich input distributions and task diversity resulting from the additional visual input.

The rapid progress of Multimodal Large Language Models(MLLMs) has transformed the AI landscape. These models combine pre-trained LLMs with various modality encoders. This integration requires a systematic understanding of how different modalities connect to the language backbone. Our survey presents an LLM-centric analysis of current approaches. We examine methods for transforming and aligning diverse modal inputs into the language embedding space. This addresses a significant gap in existing literature. We propose a classification framework for MLLMs based on three key dimensions. First, we examine architectural strategies for modality integration. This includes both the specific integration mechanisms and the fusion level. Second, we categorize representation learning techniques as either joint or coordinate representations. Third, we analyze training paradigms, including training strategies and objective functions. By examining 125 MLLMs developed between 2021 and 2025, we identify emerging patterns in the field. Our taxonomy provides researchers with a structured overview of current integration techniques. These insights aim to guide the development of more robust multimodal integration strategies for future models built on pre-trained foundations.

Current video-language models struggle with long-video understanding due to limited context lengths and reliance on sparse frame subsampling, often leading to information loss. This paper introduces $\infty$-Video, which can process arbitrarily long videos through a continuous-time long-term memory (LTM) consolidation mechanism. Our framework augments video Q-formers by allowing them to process unbounded video contexts efficiently and without requiring additional training. Through continuous attention, our approach dynamically allocates higher granularity to the most relevant video segments, forming "sticky" memories that evolve over time. Experiments with Video-LLaMA and VideoChat2 demonstrate improved performance in video question-answering tasks, showcasing the potential of continuous-time LTM mechanisms to enable scalable and training-free comprehension of long videos.

Existing Singlish spoken corpora have primarily focused on linguistic analysis and speech recognition Speech technologies have evolved over decades, tasks (Deterding and Low, 2001; Chen et al., progressing from modularized solutions for speech 2010; Lyu et al., 2010; Tan, 2019). Given the relatively recognition (Povey et al., 2011; Radford et al., small population of Singlish speakers, estimated 2023), speaker identification (Togneri and Pullella, at just a few million, resources for Singlish 2011), and gender recognition (Hechmi et al., speech corpora are significantly more limited compared 2021) with modularized toolkits like Kaldi (Povey to major languages like English, Chinese, et al., 2011) and ESPnet (Watanabe et al., 2018) French, and Spanish. Singapore's government to advanced solutions integrating large language agency, IMDA, has open-sourced the largest available models for multimodal understanding in an allencompassing, Singlish corpus, known as the National Speech omni-style approach (Team et al., Corpus (Koh et al., 2019).

Unlike traditional Automatic Speech Recognition (ASR), Audio-Visual Speech Recognition (AVSR) takes audio and visual signals simultaneously to infer the transcription. Recent studies have shown that Large Language Models (LLMs) can be effectively used for Generative Error Correction (GER) in ASR by predicting the best transcription from ASR-generated N-best hypotheses. However, these LLMs lack the ability to simultaneously understand audio and visual, making the GER approach challenging to apply in AVSR. In this work, we propose a novel GER paradigm for AVSR, termed AVGER, that follows the concept of ``listening and seeing again''. Specifically, we first use the powerful AVSR system to read the audio and visual signals to get the N-Best hypotheses, and then use the Q-former-based Multimodal Synchronous Encoder to read the audio and visual information again and convert them into an audio and video compression representation respectively that can be understood by LLM. Afterward, the audio-visual compression representation and the N-Best hypothesis together constitute a Cross-modal Prompt to guide the LLM in producing the best transcription. In addition, we also proposed a Multi-Level Consistency Constraint training criterion, including logits-level, utterance-level and representations-level, to improve the correction accuracy while enhancing the interpretability of audio and visual compression representations. The experimental results on the LRS3 dataset show that our method outperforms current mainstream AVSR systems. The proposed AVGER can reduce the Word Error Rate (WER) by 24% compared to them. Code and models can be found at: https://github.com/CircleRedRain/AVGER.

This paper investigates a novel approach to end-to-end speech translation (ST) based on aligning frozen pre-trained automatic speech recognition (ASR) and machine translation (MT) models via a small connector module (Q-Former, our Subsampler-Transformer Encoder). This connector bridges the gap between the speech and text modalities, transforming ASR encoder embeddings into the latent representation space of the MT encoder while being the only part of the system optimized during training. Experiments are conducted on the How2 English-Portuguese dataset as we investigate the alignment approach in a small-scale scenario focusing on ST. While keeping the size of the connector module constant and small in comparison ( < 5% of the size of the larger aligned models), increasing the size and capability of the foundation ASR and MT models universally improves translation results. We also find that the connectors can serve as domain adapters for the foundation MT models, significantly improving translation performance in the aligned ST setting. We conclude that this approach represents a viable and scalable approach to training end-to-end ST systems.