分享

In this paper, we aim to generate clean speech frame by frame from a live video stream and a noisy audio stream without relying on future inputs. To this end, we propose RT-LA-VocE, which completely re-designs every component of LA-VocE, a state-of-the-art non-causal audio-visual speech enhancement model, to perform causal real-time inference with a 40ms input frame. We do so by devising new visual and audio encoders that rely solely on past frames, replacing the Transformer encoder with the Emformer, and designing a new causal neural vocoder C-HiFi-GAN. On the popular AVSpeech dataset, we show that our algorithm achieves state-of-the-art results in all real-time scenarios. More importantly, each component is carefully tuned to minimize the algorithm latency to the theoretical minimum (40ms) while maintaining a low end-to-end processing latency of 28.15ms per frame, enabling real-time frame-by-frame enhancement with minimal delay.

LiDAR-camera 3D representation pretraining has shown significant promise for 3D perception tasks and related applications. However, two issues widely exist in this framework: 1) Solely keyframes are used for training. For example, in nuScenes, a substantial quantity of unpaired LiDAR and camera frames remain unutilized, limiting the representation capabilities of the pretrained network. 2) The contrastive loss erroneously distances points and image regions with identical semantics but from different frames, disturbing the semantic consistency of the learned presentations. In this paper, we propose a novel Vision-Foundation-Model-driven sample exploring module to meticulously select LiDAR-Image pairs from unexplored frames, enriching the original training set. We utilized timestamps and the semantic priors from VFMs to identify well-synchronized training pairs and to discover samples with diverse content. Moreover, we design a cross- and intra-modal conflict-aware contrastive loss using the semantic mask labels of VFMs to avoid contrasting semantically similar points and image regions. Our method consistently outperforms existing state-of-the-art pretraining frameworks across three major public autonomous driving datasets: nuScenes, SemanticKITTI, and Waymo on 3D semantic segmentation by +3.0\%, +3.0\%, and +3.3\% in mIoU, respectively. Furthermore, our approach exhibits adaptable generalization to different 3D backbones and typical semantic masks generated by non-VFM models.

This technical report presents the implementation of a state-of-the-art video encoder for video-text modal alignment and a video conversation framework called HiLight, which features dual visual towers. The work is divided into two main parts: 1.alignment of video and text modalities; 2.convenient and efficient way to interact with users. Our goal is to address the task of video comprehension in the context of billiards. The report includes a discussion of the concepts and the final solution developed during the task's implementation.

As experts in voice modification, trans-feminine gender-affirming voice teachers have unique perspectives on voice that confound current understandings of speaker identity. To demonstrate this, we present the Versatile Voice Dataset (VVD), a collection of three speakers modifying their voices along gendered axes. The VVD illustrates that current approaches in speaker modeling, based on categorical notions of gender and a static understanding of vocal texture, fail to account for the flexibility of the vocal tract. Utilizing publicly-available speaker embeddings, we demonstrate that gender classification systems are highly sensitive to voice modification, and speaker verification systems fail to identify voices as coming from the same speaker as voice modification becomes more drastic. As one path towards moving beyond categorical and static notions of speaker identity, we propose modeling individual qualities of vocal texture such as pitch, resonance, and weight.

In this work, we introduce LokiLM, a 1.4B parameter large language model trained on 500B tokens. Our model performs strongly in natural language reasoning tasks and achieves state-of-the-art performance among models with 1.5B parameters or less. LokiLM is trained using multi-teacher knowledge distillation and high-quality training data to achieve benchmark results competitive with larger models trained on significantly more tokens. We support these findings by introducing steps to avoid benchmark contamination and overfitting throughout our development process. Despite its promising performance, LokiLM exhibits a concerning amount of hallucinations and scores poorly on the TruthfulQA benchmark, so we do not release the model publicly.

This study applies Natural Language Processing techniques, including Latent Dirichlet Allocation, to analyse anonymised maternity incident investigation reports from the Healthcare Safety Investigation Branch. The reports underwent preprocessing, annotation using the Safety Intelligence Research taxonomy, and topic modelling to uncover prevalent topics and detect differences in maternity care across ethnic groups. A combination of offline and online methods was utilised to ensure data protection whilst enabling advanced analysis, with offline processing for sensitive data and online processing for non-sensitive data using the `Claude 3 Opus' language model. Interactive topic analysis and semantic network visualisation were employed to extract and display thematic topics and visualise semantic relationships among keywords. The analysis revealed disparities in care among different ethnic groups, with distinct focus areas for the Black, Asian, and White British ethnic groups. The study demonstrates the effectiveness of topic modelling and NLP techniques in analysing maternity incident investigation reports and highlighting disparities in care. The findings emphasise the crucial role of advanced data analysis in improving maternity care quality and equity.

Culture and language evolve together. The old literary form of Tamil is used commonly for writing and the contemporary colloquial Tamil is used for speaking. Human-computer interaction applications require Colloquial Tamil (CT) to make it more accessible and easy for the everyday user and, it requires Literary Tamil (LT) when information is needed in a formal written format. Continuing the use of LT alongside CT in computer aided language learning applications will both preserve LT, and provide ease of use via CT, at the same time. Hence there is a need for the conversion between LT and CT dialects, which demands as a first step, dialect identification. Dialect Identification (DID) of LT and CT is an unexplored area of research. In the current work, keeping the nuances of both these dialects in mind, five methods are explored which include two implicit methods - Gaussian Mixture Model (GMM) and Convolutional Neural Network (CNN); two explicit methods - Parallel Phone Recognition (PPR) and Parallel Large Vocabulary Continuous Speech Recognition (P-LVCSR); two versions of the proposed explicit Unified Phone Recognition method (UPR-1 and UPR-2). These methods vary based on: the need for annotated data, the size of the unit, the way in which modelling is carried out, and the way in which the final decision is made. Even though the average duration of the test utterances is less - 4.9s for LT and 2.5s for CT - the systems performed well, offering the following identification accuracies: 87.72% (GMM), 93.97% (CNN), 89.24% (PPR), 94.21% (P-LVCSR), 88.57% (UPR-1), 93.53% (UPR-1 with P-LVCSR), 94.55% (UPR-2), and 95.61% (UPR-2 with P-LVCSR).

In recent years, the field of Natural Language Generation (NLG) has been boosted by the recent advances in deep learning technologies. Nonetheless, these new data-intensive methods introduce language-dependent disparities in NLG as the main training data sets are in English. Also, most neural NLG systems use decoder-only (causal) transformer language models, which work well for English, but were not designed with other languages in mind. In this work we depart from the hypothesis that they may introduce generation bias in target languages with less rigid word ordering, subject omission, or different attachment preferences for relative clauses, so that for these target languages other language generation strategies may be more desirable. This paper first compares causal and non-causal language modeling for English and Spanish, two languages with different grammatical structures and over 1.5 billion and 0.5 billion speakers, respectively. For this purpose, we define a novel metric of average causal and non-causal context-conditioned entropy of the grammatical category distribution for both languages as an information-theoretic a priori approach. The evaluation of natural text sources (such as training data) in both languages reveals lower average non-causal conditional entropy in Spanish and lower causal conditional entropy in English. According to this experiment, Spanish is more predictable than English given a non-causal context. Then, by applying a conditional relative entropy metric to text generation experiments, we obtain as insights that the best performance is respectively achieved with causal NLG in English, and with non-causal NLG in Spanish. These insights support further research in NLG in Spanish using bidirectional transformer language models.

Spiking neural networks (SNNs) are gaining popularity in the computational simulation and artificial intelligence fields owing to their biological plausibility and computational efficiency. This paper explores the historical development of SNN and concludes that these two fields are intersecting and merging rapidly. Following the successful application of Dynamic Vision Sensors (DVS) and Dynamic Audio Sensors (DAS), SNNs have found some proper paradigms, such as continuous visual signal tracking, automatic speech recognition, and reinforcement learning for continuous control, that have extensively supported their key features, including spike encoding, neuronal heterogeneity, specific functional circuits, and multiscale plasticity. Compared to these real-world paradigms, the brain contains a spiking version of the biology-world paradigm, which exhibits a similar level of complexity and is usually considered a mirror of the real world. Considering the projected rapid development of invasive and parallel Brain-Computer Interface (BCI), as well as the new BCI-based paradigms that include online pattern recognition and stimulus control of biological spike trains, SNNs naturally leverage their advantages in energy efficiency, robustness, and flexibility. The biological brain has inspired the present study of SNNs and effective SNN machine-learning algorithms, which can help enhance neuroscience discoveries in the brain by applying them to the new BCI paradigm. Such two-way interactions with positive feedback can accelerate brain science research and brain-inspired intelligence technology.

Autonomous driving holds great potential to transform road safety and traffic efficiency by minimizing human error and reducing congestion. A key challenge in realizing this potential is the accurate estimation of steering angles, which is essential for effective vehicle navigation and control. Recent breakthroughs in deep learning have made it possible to estimate steering angles directly from raw camera inputs. However, the limited available navigation data can hinder optimal feature learning, impacting the system's performance in complex driving scenarios. In this paper, we propose a shared encoder trained on multiple computer vision tasks critical for urban navigation, such as depth, pose, and 3D scene flow estimation, as well as semantic, instance, panoptic, and motion segmentation. By incorporating diverse visual information used by humans during navigation, this unified encoder might enhance steering angle estimation. To achieve effective multi-task learning within a single encoder, we introduce a multi-scale feature network for pose estimation to improve depth learning. Additionally, we employ knowledge distillation from a multi-backbone model pretrained on these navigation tasks to stabilize training and boost performance. Our findings demonstrate that a shared backbone trained on diverse visual tasks is capable of providing overall perception capabilities. While our performance in steering angle estimation is comparable to existing methods, the integration of human-like perception through multi-task learning holds significant potential for advancing autonomous driving systems. More details and the pretrained model are available at https://hi-computervision.github.io/uni-encoder/.

This paper investigates scaling laws for local SGD in LLM training, a distributed optimization algorithm that facilitates training on loosely connected devices. Through extensive experiments, we show that local SGD achieves competitive results compared to conventional methods, given equivalent model parameters, datasets, and computational resources. Furthermore, we explore the application of local SGD in various practical scenarios, including multi-cluster setups and edge computing environments. Our findings elucidate the necessary conditions for effective multi-cluster LLM training and examine the potential and limitations of leveraging edge computing resources in the LLM training process. This demonstrates its viability as an alternative to single large-cluster training.

Large language models (LLMs) have shown potential in biomedical applications, leading to efforts to fine-tune them on domain-specific data. However, the effectiveness of this approach remains unclear. This study evaluates the performance of biomedically fine-tuned LLMs against their general-purpose counterparts on a variety of clinical tasks. We evaluated their performance on clinical case challenges from the New England Journal of Medicine (NEJM) and the Journal of the American Medical Association (JAMA) and on several clinical tasks (e.g., information extraction, document summarization, and clinical coding). Using benchmarks specifically chosen to be likely outside the fine-tuning datasets of biomedical models, we found that biomedical LLMs mostly perform inferior to their general-purpose counterparts, especially on tasks not focused on medical knowledge. While larger models showed similar performance on case tasks (e.g., OpenBioLLM-70B: 66.4% vs. Llama-3-70B-Instruct: 65% on JAMA cases), smaller biomedical models showed more pronounced underperformance (e.g., OpenBioLLM-8B: 30% vs. Llama-3-8B-Instruct: 64.3% on NEJM cases). Similar trends were observed across the CLUE (Clinical Language Understanding Evaluation) benchmark tasks, with general-purpose models often performing better on text generation, question answering, and coding tasks. Our results suggest that fine-tuning LLMs to biomedical data may not provide the expected benefits and may potentially lead to reduced performance, challenging prevailing assumptions about domain-specific adaptation of LLMs and highlighting the need for more rigorous evaluation frameworks in healthcare AI. Alternative approaches, such as retrieval-augmented generation, may be more effective in enhancing the biomedical capabilities of LLMs without compromising their general knowledge.