分享

Concurrent Speaker Detection (CSD), the task of identifying the presence and overlap of active speakers in an audio signal, is crucial for many audio tasks such as meeting transcription, speaker diarization, and speech separation. This study introduces a multimodal deep learning approach that leverages both audio and visual information. The proposed model employs an early fusion strategy combining audio and visual features through cross-modal attention mechanisms, with a learnable [CLS] token capturing the relevant audio-visual relationships. The model is extensively evaluated on two real-world datasets, AMI and the recently introduced EasyCom dataset. Experiments validate the effectiveness of the multimodal fusion strategy. Ablation studies further support the design choices and the training procedure of the model. As this is the first work reporting CSD results on the challenging EasyCom dataset, the findings demonstrate the potential of the proposed multimodal approach for CSD in real-world scenarios.

Language models are capable of iteratively improving their outputs based on natural language feedback, thus enabling in-context optimization of user preference. In place of human users, a second language model can be used as an evaluator, providing feedback along with numerical ratings which the generator attempts to optimize. However, because the evaluator is an imperfect proxy of user preference, this optimization can lead to reward hacking, where the evaluator's ratings improve while the generation quality remains stagnant or even decreases as judged by actual user preference. The concern of reward hacking is heightened in iterative self-refinement where the generator and the evaluator use the same underlying language model, in which case the optimization pressure can drive them to exploit shared vulnerabilities. Using an essay editing task, we show that iterative self-refinement leads to deviation between the language model evaluator and human judgment, demonstrating that reward hacking can occur spontaneously in-context with the use of iterative self-refinement. In addition, we study conditions under which reward hacking occurs and observe two factors that affect reward hacking severity: model size and context sharing between the generator and the evaluator.

Prompt engineering, as an efficient and effective way to leverage Large Language Models (LLM), has drawn a lot of attention from the research community. The existing research primarily emphasizes the importance of adapting prompts to specific tasks, rather than specific LLMs. However, a good prompt is not solely defined by its wording, but also binds to the nature of the LLM in question. In this work, we first quantitatively demonstrate that different prompts should be adapted to different LLMs to enhance their capabilities across various downstream tasks in NLP. Then we novelly propose a model-adaptive prompt optimizer (MAPO) method that optimizes the original prompts for each specific LLM in downstream tasks. Extensive experiments indicate that the proposed method can effectively refine prompts for an LLM, leading to significant improvements over various downstream tasks.

Serialized Output Training (SOT) has showcased state-of-the-art performance in multi-talker speech recognition by sequentially decoding the speech of individual speakers. To address the challenging label-permutation issue, prior methods have relied on either the Permutation Invariant Training (PIT) or the time-based First-In-First-Out (FIFO) rule. This study presents a model-based serialization strategy that incorporates an auxiliary module into the Attention Encoder-Decoder architecture, autonomously identifying the crucial factors to order the output sequence of the speech components in multi-talker speech. Experiments conducted on the LibriSpeech and LibriMix databases reveal that our approach significantly outperforms the PIT and FIFO baselines in both 2-mix and 3-mix scenarios. Further analysis shows that the serialization module identifies dominant speech components in a mixture by factors including loudness and gender, and orders speech components based on the dominance score.

This paper presents NT-Java-1.1B, an open-source specialized code language model built on StarCoderBase-1.1B, designed for coding tasks in Java programming. NT-Java-1.1B achieves state-of-the-art performance, surpassing its base model and majority of other models of similar size on MultiPL-E Java code benchmark. While there have been studies on extending large, generic pre-trained models to improve proficiency in specific programming languages like Python, similar investigations on small code models for other programming languages are lacking. Large code models require specialized hardware like GPUs for inference, highlighting the need for research into building small code models that can be deployed on developer desktops. This paper addresses this research gap by focusing on the development of a small Java code model, NT-Java-1.1B, and its quantized versions, which performs comparably to open models around 1.1B on MultiPL-E Java code benchmarks, making them ideal for desktop deployment. This paper establishes the foundation for specialized models across languages and sizes for a family of NT Models.

Discretizing speech into tokens and generating them by a decoder-only model have been a promising direction for text-to-speech (TTS) and spoken language modeling (SLM). To shorten the sequence length of speech tokens, acoustic byte-pair encoding (BPE) has emerged in SLM that treats speech tokens from self-supervised semantic representations as characters to further compress the token sequence. But the gain in TTS has not been fully investigated, and the proper choice of acoustic BPE remains unclear. In this work, we conduct a comprehensive study on various settings of acoustic BPE to explore its effectiveness in decoder-only TTS models with semantic speech tokens. Experiments on LibriTTS verify that acoustic BPE uniformly increases the intelligibility and diversity of synthesized speech, while showing different features across BPE settings. Hence, acoustic BPE is a favorable tool for decoder-only TTS.

Alignment of Large Language Models (LLMs) remains an unsolved problem. Human preferences are highly distributed and can be captured at multiple levels of abstraction, from the individual to diverse populations. Organisational preferences, represented by standards and principles, are defined to mitigate reputational risk or meet legislative obligations. In this paper, we present ABC Align, a novel alignment methodology for LLMs that enables integration of the standards and preferences of a large media organisation into the LLM itself. We combine a set of data and methods that build on recent breakthroughs in synthetic data generation, preference optimisation, and post-training model quantisation. Our unified approach mitigates bias and improves accuracy, while preserving reasoning capability, as measured against standard benchmarks.

Visual geo-localization demands in-depth knowledge and advanced reasoning skills to associate images with precise real-world geographic locations. Existing image database retrieval methods are limited by the impracticality of storing sufficient visual records of global landmarks. Recently, Large Vision-Language Models (LVLMs) have demonstrated the capability of geo-localization through Visual Question Answering (VQA), enabling a solution that does not require external geo-tagged image records. However, the performance of a single LVLM is still limited by its intrinsic knowledge and reasoning capabilities. To address these challenges, we introduce smileGeo, a novel visual geo-localization framework that leverages multiple Internet-enabled LVLM agents operating within an agent-based architecture. By facilitating inter-agent communication, smileGeo integrates the inherent knowledge of these agents with additional retrieved information, enhancing the ability to effectively localize images. Furthermore, our framework incorporates a dynamic learning strategy that optimizes agent communication, reducing redundant interactions and enhancing overall system efficiency. To validate the effectiveness of the proposed framework, we conducted experiments on three different datasets, and the results show that our approach significantly outperforms current state-of-the-art methods. The source code is available at https://anonymous.4open.science/r/ViusalGeoLocalization-F8F5.

Multimodal Large Language Models (MLLMs) have recently demonstrated remarkable perceptual and reasoning abilities, typically comprising a Vision Encoder, an Adapter, and a Large Language Model (LLM). The adapter serves as the critical bridge between the visual and language components. However, training adapters with image-level supervision often results in significant misalignment, undermining the LLMs' capabilities and limiting the potential of Multimodal LLMs. To address this, we introduce Supervised Embedding Alignment (SEA), a token-level alignment method that leverages vision-language pre-trained models, such as CLIP, to align visual tokens with the LLM's embedding space through contrastive learning. This approach ensures a more coherent integration of visual and language representations, enhancing the performance and interpretability of multimodal LLMs while preserving their inherent capabilities. Extensive experiments show that SEA effectively improves MLLMs, particularly for smaller models, without adding extra data or inference computation. SEA also lays the groundwork for developing more general and adaptable solutions to enhance multimodal systems.

Large Language Models (LLMs) have demonstrated significant potential in clinical applications through prompt engineering, which enables the generation of flexible and diverse clinical predictions. However, they pose challenges in producing prediction probabilities, which are essential for transparency and allowing clinicians to apply flexible probability thresholds in decision-making. While explicit prompt instructions can lead LLMs to provide prediction probability numbers through text generation, LLMs' limitations in numerical reasoning raise concerns about the reliability of these text-generated probabilities. To assess this reliability, we compared explicit probabilities derived from text generation to implicit probabilities calculated based on the likelihood of predicting the correct label token. Experimenting with six advanced open-source LLMs across five medical datasets, we found that the performance of explicit probabilities was consistently lower than implicit probabilities with respect to discrimination, precision, and recall. Moreover, these differences were enlarged on small LLMs and imbalanced datasets, emphasizing the need for cautious interpretation and applications, as well as further research into robust probability estimation methods for LLMs in clinical contexts.

Generating images with accurately represented text, especially in non-Latin languages, poses a significant challenge for diffusion models. Existing approaches, such as the integration of hint condition diagrams via auxiliary networks (e.g., ControlNet), have made strides towards addressing this issue. However, diffusion models often fall short in tasks requiring controlled text generation, such as specifying particular fonts or producing text in small fonts. In this paper, we introduce a novel approach for multilingual visual text creation, named JoyType, designed to maintain the font style of text during the image generation process. Our methodology begins with assembling a training dataset, JoyType-1M, comprising 1 million pairs of data. Each pair includes an image, its description, and glyph instructions corresponding to the font style within the image. We then developed a text control network, Font ControlNet, tasked with extracting font style information to steer the image generation. To further enhance our model's ability to maintain font style, notably in generating small-font text, we incorporated a multi-layer OCR-aware loss into the diffusion process. This enhancement allows JoyType to direct text rendering using low-level descriptors. Our evaluations, based on both visual and accuracy metrics, demonstrate that JoyType significantly outperforms existing state-of-the-art methods. Additionally, JoyType can function as a plugin, facilitating the creation of varied image styles in conjunction with other stable diffusion models on HuggingFace and CivitAI. Our project is open-sourced on https://jdh-algo.github.io/JoyType/.

We introduce LingGen, a novel approach for controlled text generation that offers precise control over a wide array of linguistic attributes, even as the number of attributes varies. LingGen employs a dynamic P-MASKING strategy, which samples masking rates from a power law distribution during training. This innovative approach enables the model to develop robust representations and adapt its attribute control capabilities across a variable number of attributes, from a single attribute to multiple complex configurations. The P-MASKING technique enhances LingGen's ability to manage different levels of attribute visibility, resulting in superior performance in multi-attribute generation tasks. Our experiments demonstrate that LingGen surpasses current state-of-the-art models in both attribute control accuracy and text fluency, particularly excelling in scenarios with varying attribute demands. Additionally, our ablation studies highlight the effectiveness of P-MASKING and the influence of different base language models on performance. These findings demonstrate LingGen's potential for applications requiring precise and adaptable control over multiple linguistic attributes in text generation.