Large language model (LLM) based zero-shot text-to-speech (TTS) methods tend to preserve the acoustic environment of the audio prompt, leading to degradation in synthesized speech quality when the audio prompt contains noise. In this paper, we propose a novel neural codec-based speech denoiser and integrate it with the advanced LLM-based TTS model, LauraTTS, to achieve noise-robust zero-shot TTS. The proposed codec denoiser consists of an audio codec, a token denoiser, and an embedding refiner. The token denoiser predicts the first two groups of clean acoustic tokens from the noisy ones, which can serve as the acoustic prompt for LauraTTS to synthesize high-quality personalized speech or be converted to clean speech waveforms through the embedding refiner and codec decoder. Experimental results show that our proposed codec denoiser outperforms state-of-the-art speech enhancement (SE) methods, and the proposed noise-robust LauraTTS surpasses the approach using additional SE models.
Large language models (LLMs) exhibit remarkable capabilities in handling natural language tasks; however, they may struggle to consistently follow complex instructions including those involve multiple constraints. Post-training LLMs using supervised fine-tuning (SFT) is a standard approach to improve their ability to follow instructions. In addressing complex instruction following, existing efforts primarily focus on data-driven methods that synthesize complex instruction-output pairs for SFT. However, insufficient attention allocated to crucial sub-contexts may reduce the effectiveness of SFT. In this work, we propose transforming sequentially structured input instruction into multiple parallel instructions containing subcontexts. To support processing this multi-input, we propose MISO (Multi-Input Single-Output), an extension to currently dominant decoder-only transformer-based LLMs. MISO introduces a mixture-of-contexts paradigm that jointly considers the overall instruction-output alignment and the influence of individual sub-contexts to enhance SFT effectiveness. We apply MISO fine-tuning to complex instructionfollowing datasets and evaluate it with standard LLM inference. Empirical results demonstrate the superiority of MISO as a fine-tuning method for LLMs, both in terms of effectiveness in complex instruction-following scenarios and its potential for training efficiency.
Multi-hop Question Answering (MHQA) adds layers of complexity to question answering, making it more challenging. When Language Models (LMs) are prompted with multiple search results, they are tasked not only with retrieving relevant information but also employing multi-hop reasoning across the information sources. Although LMs perform well on traditional question-answering tasks, the causal mask can hinder their capacity to reason across complex contexts. In this paper, we explore how LMs respond to multi-hop questions by permuting search results (retrieved documents) under various configurations. Our study reveals interesting findings as follows: 1) Encoder-decoder models, such as the ones in the Flan-T5 family, generally outperform causal decoder-only LMs in MHQA tasks, despite being significantly smaller in size; 2) altering the order of gold documents reveals distinct trends in both Flan T5 models and fine-tuned decoder-only models, with optimal performance observed when the document order aligns with the reasoning chain order; 3) enhancing causal decoder-only models with bi-directional attention by modifying the causal mask can effectively boost their end performance. In addition to the above, we conduct a thorough investigation of the distribution of LM attention weights in the context of MHQA. Our experiments reveal that attention weights tend to peak at higher values when the resulting answer is correct. We leverage this finding to heuristically improve LMs' performance on this task. Our code is publicly available at https://github.com/hwy9855/MultiHopQA-Reasoning.
OpenAI's Whisper has achieved significant success in Automatic Speech Recognition. However, it has consistently been found to exhibit hallucination issues, particularly in non-speech segments, which limits its broader application in complex industrial settings. In this paper, we introduce a novel method to reduce Whisper's hallucination on non-speech segments without using any pre- or post-possessing techniques. Specifically, we benchmark the contribution of each self-attentional head in the Whisper-large-v3 decoder to the hallucination problem by performing a head-wise mask. Our findings reveal that only 3 of the 20 heads account for over 75% of the hallucinations on the UrbanSound dataset. We then fine-tune these three crazy heads using a collection of non-speech data. The results show that our best fine-tuned model, namely Calm-Whisper, achieves over 80% reduction in non-speech hallucination with only less than 0.1% WER degradation on LibriSpeech test-clean and test-other.
The use of audio recordings of human speech to train LLMs poses privacy concerns due to these models' potential to generate outputs that closely resemble artifacts in the training data. In this study, we propose a speaker privacy-preserving representation learning method through the Universal Speech Codec (USC), a computationally efficient encoder-decoder model that disentangles speech into: (i) privacy-preserving semantically rich representations, capturing content and speech paralinguistics, and (ii) residual acoustic and speaker representations that enables high-fidelity reconstruction. Extensive evaluations presented show that USC's semantic representation preserves content, prosody, and sentiment, while removing potentially identifiable speaker attributes. Combining both representations, USC achieves state-of-the-art speech reconstruction. Additionally, we introduce an evaluation methodology for measuring privacy-preserving properties, aligning with perceptual tests. We compare USC against other codecs in the literature and demonstrate its effectiveness on privacy-preserving representation learning, illustrating the trade-offs of speaker anonymization, paralinguistics retention and content preservation in the learned semantic representations. Audio samples are shared in https://www.amazon.science/usc-samples.
Deep generative models are promising tools for science and engineering, but their reliance on abundant, high-quality data limits applicability. We present a novel framework for generative modeling of random fields (probability distributions over continuous functions) that incorporates domain knowledge to supplement limited, sparse, and indirect data. The foundation of the approach is latent flow matching, where generative modeling occurs on compressed function representations in the latent space of a pre-trained variational autoencoder (VAE). Innovations include the adoption of a function decoder within the VAE and integration of physical/statistical constraints into the VAE training process. In this way, a latent function representation is learned that yields continuous random field samples satisfying domain-specific constraints when decoded, even in data-limited regimes. Efficacy is demonstrated on two challenging applications: wind velocity field reconstruction from sparse sensors and material property inference from a limited number of indirect measurements. Results show that the proposed framework achieves significant improvements in reconstruction accuracy compared to unconstrained methods and enables effective inference with relatively small training datasets that is intractable without constraints.
Open-vocabulary semantic segmentation aims to segment images into distinct semantic regions for both seen and unseen categories at the pixel level. Current methods utilize text embeddings from pre-trained vision-language models like CLIP but struggle with the inherent domain gap between image and text embeddings, even after extensive alignment during training. Additionally, relying solely on deep text-aligned features limits shallow-level feature guidance, which is crucial for detecting small objects and fine details, ultimately reducing segmentation accuracy. To address these limitations, we propose a dual prompting framework, DPSeg, for this task. Our approach combines dual-prompt cost volume generation, a cost volume-guided decoder, and a semantic-guided prompt refinement strategy that leverages our dual prompting scheme to mitigate alignment issues in visual prompt generation. By incorporating visual embeddings from a visual prompt encoder, our approach reduces the domain gap between text and image embeddings while providing multi-level guidance through shallow features. Extensive experiments demonstrate that our method significantly outperforms existing state-of-the-art approaches on multiple public datasets.
We report on the application of a high-capacity semantic segmentation pipeline to the GOOSE 2D Semantic Segmentation Challenge for unstructured off-road environments. Using a FlashInternImage-B backbone together with a UPerNet decoder, we adapt established techniques, rather than designing new ones, to the distinctive conditions of off-road scenes. Our training recipe couples strong photometric distortion augmentation (to emulate the wide lighting variations of outdoor terrain) with an Exponential Moving Average (EMA) of weights for better generalization. Using only the GOOSE training dataset, we achieve 88.8\% mIoU on the validation set.
Denoising diffusion models achieved impressive results on several image generation tasks often outperforming GAN based models. Recently, the generative capabilities of diffusion models have been employed for perceptual image compression, such as in CDC. A major drawback of these diffusion-based methods is that, while producing impressive perceptual quality images they are dropping in fidelity/increasing the distortion to the original uncompressed images when compared with other traditional or learned image compression schemes aiming for fidelity. In this paper, we propose a hybrid compression scheme optimized for perceptual quality, extending the approach of the CDC model with a decoder network in order to reduce the impact on distortion metrics such as PSNR. After using the decoder network to generate an initial image, optimized for distortion, the latent conditioned diffusion model refines the reconstruction for perceptual quality by predicting the residual. On standard benchmarks, we achieve up to +2dB PSNR fidelity improvements while maintaining comparable LPIPS and FID perceptual scores when compared with CDC. Additionally, the approach is easily extensible to video compression, where we achieve similar results.
Modern video generation frameworks based on Latent Diffusion Models suffer from inefficiencies in tokenization due to the Frame-Proportional Information Assumption. Existing tokenizers provide fixed temporal compression rates, causing the computational cost of the diffusion model to scale linearly with the frame rate. The paper proposes the Duration-Proportional Information Assumption: the upper bound on the information capacity of a video is proportional to the duration rather than the number of frames. Based on this insight, the paper introduces VFRTok, a Transformer-based video tokenizer, that enables variable frame rate encoding and decoding through asymmetric frame rate training between the encoder and decoder. Furthermore, the paper proposes Partial Rotary Position Embeddings (RoPE) to decouple position and content modeling, which groups correlated patches into unified tokens. The Partial RoPE effectively improves content-awareness, enhancing the video generation capability. Benefiting from the compact and continuous spatio-temporal representation, VFRTok achieves competitive reconstruction quality and state-of-the-art generation fidelity while using only 1/8 tokens compared to existing tokenizers.
Despite the prominence of decoder-only language models, encoders remain crucial for resource-constrained applications. We introduce ModernGBERT (134M, 1B), a fully transparent family of German encoder models trained from scratch, incorporating architectural innovations from ModernBERT. To evaluate the practical trade-offs of training encoders from scratch, we also present LL\"aMmlein2Vec (120M, 1B, 7B), a family of encoders derived from German decoder-only models via LLM2Vec. We benchmark all models on natural language understanding, text embedding, and long-context reasoning tasks, enabling a controlled comparison between dedicated encoders and converted decoders. Our results show that ModernGBERT 1B outperforms prior state-of-the-art German encoders as well as encoders adapted via LLM2Vec, with regard to performance and parameter-efficiency. All models, training data, checkpoints and code are publicly available, advancing the German NLP ecosystem with transparent, high-performance encoder models.
With the rapid rise of large models, copyright protection for generated image content has become a critical security challenge. Although deep learning watermarking techniques offer an effective solution for digital image copyright protection, they still face limitations in terms of visual quality, robustness and generalization. To address these issues, this paper proposes an adaptive robust iterative watermarking framework (ARIW-Framework) that achieves high-quality watermarked images while maintaining exceptional robustness and generalization performance. Specifically, we introduce an iterative approach to optimize the encoder for generating robust residuals. The encoder incorporates noise layers and a decoder to compute robustness weights for residuals under various noise attacks. By employing a parallel optimization strategy, the framework enhances robustness against multiple types of noise attacks. Furthermore, we leverage image gradients to determine the embedding strength at each pixel location, significantly improving the visual quality of the watermarked images. Extensive experiments demonstrate that the proposed method achieves superior visual quality while exhibiting remarkable robustness and generalization against noise attacks.
Recent text-to-image systems face limitations in handling multimodal inputs and complex reasoning tasks. We introduce MindOmni, a unified multimodal large language model that addresses these challenges by incorporating reasoning generation through reinforcement learning. MindOmni leverages a three-phase training strategy: i) design of a unified vision language model with a decoder-only diffusion module, ii) supervised fine-tuning with Chain-of-Thought (CoT) instruction data, and iii) our proposed Reasoning Generation Policy Optimization (RGPO) algorithm, utilizing multimodal feedback to effectively guide policy updates. Experimental results demonstrate that MindOmni outperforms existing models, achieving impressive performance on both understanding and generation benchmarks, meanwhile showcasing advanced fine-grained reasoning generation capabilities, especially with mathematical reasoning instruction. All codes will be made public at \href{https://github.com/EasonXiao-888/MindOmni}{https://github.com/EasonXiao-888/MindOmni}.
Accurate prediction of traffic flow parameters and real time identification of congestion states are essential for the efficient operation of intelligent transportation systems. This paper proposes a Periodic Pattern Transformer Network (PPTNet) for traffic flow prediction, integrating periodic pattern extraction with the Transformer architecture, coupled with a fuzzy inference method for real-time congestion identification. Firstly, a high-precision traffic flow dataset (Traffic Flow Dataset for China's Congested Highways and Expressways, TF4CHE) suitable for congested highway scenarios in China is constructed based on drone aerial imagery data. Subsequently, the proposed PPTNet employs Fast Fourier Transform to capture multi-scale periodic patterns and utilizes two-dimensional Inception convolutions to efficiently extract intra and inter periodic features. A Transformer decoder dynamically models temporal dependencies, enabling accurate predictions of traffic density and speed. Finally, congestion probabilities are calculated in real-time using the predicted outcomes via a Mamdani fuzzy inference-based congestion identification module. Experimental results demonstrate that the proposed PPTNet significantly outperforms mainstream traffic prediction methods in prediction accuracy, and the congestion identification module effectively identifies real-time road congestion states, verifying the superiority and practicality of the proposed method in real-world traffic scenarios. Project page: https://github.com/ADSafetyJointLab/PPTNet.
Code-mixed languages, characterized by frequent within-sentence language transitions, present structural challenges that standard language models fail to address. In this work, we propose CMLFormer, an enhanced multi-layer dual-decoder Transformer with a shared encoder and synchronized decoder cross-attention, designed to model the linguistic and semantic dynamics of code-mixed text. CMLFormer is pre-trained on an augmented Hinglish corpus with switching point and translation annotations with multiple new objectives specifically aimed at capturing switching behavior, cross-lingual structure, and code-mixing complexity. Our experiments show that CMLFormer improves F1 score, precision, and accuracy over other approaches on the HASOC-2021 benchmark under select pre-training setups. Attention analyses further show that it can identify and attend to switching points, validating its sensitivity to code-mixed structure. These results demonstrate the effectiveness of CMLFormer's architecture and multi-task pre-training strategy for modeling code-mixed languages.
In this paper, an unsupervised Recurrent Wavelet Probabilistic Neural Network (RWPNN) is proposed, which aims at detecting anomalies in non-stationary environments by modelling the temporal features using a nonparametric density estimation network. The novel framework consists of two components, a Stacked Recurrent Encoder-Decoder (SREnc-Dec) module that captures temporal features in a latent space, and a Multi-Receptive-field Wavelet Probabilistic Network (MRWPN) that creates an ensemble probabilistic model to characterise the latent space. This formulation extends the standard wavelet probabilistic networks to wavelet deep probabilistic networks, which can handle higher data dimensionality. The MRWPN module can adapt to different rates of data variation in different datasets without imposing strong distribution assumptions, resulting in a more robust and accurate detection for Time Series Anomaly Detection (TSAD) tasks in the non-stationary environment. We carry out the assessment on 45 real-world time series datasets from various domains, verify the performance of RWPNN in TSAD tasks with several constraints, and show its ability to provide early warnings for anomalous events.
Variational Autoencoders (VAEs) are well-established as a principled approach to probabilistic unsupervised learning with neural networks. Typically, an encoder network defines the parameters of a Gaussian distributed latent space from which we can sample and pass realizations to a decoder network. This model is trained to reconstruct its inputs and is optimized through the evidence lower bound. In recent years, discrete latent spaces have grown in popularity, suggesting that they may be a natural choice for many data modalities (e.g. text). In this tutorial, we provide a rigorous, yet practical, introduction to discrete variational autoencoders -- specifically, VAEs in which the latent space is made up of latent variables that follow a categorical distribution. We assume only a basic mathematical background with which we carefully derive each step from first principles. From there, we develop a concrete training recipe and provide an example implementation, hosted at https://github.com/alanjeffares/discreteVAE.
State-of-the-art upper limb myoelectric prostheses often use pattern recognition (PR) control systems that translate electromyography (EMG) signals into desired movements. As prosthesis movement complexity increases, users often struggle to produce sufficiently distinct EMG patterns for reliable classification. Existing training typically involves heuristic, trial-and-error user adjustments to static decoder boundaries. Goal: We introduce the Reviewer, a 3D visual interface projecting EMG signals directly into the decoder's classification space, providing intuitive, real-time insight into PR algorithm behavior. This structured feedback reduces cognitive load and fosters mutual, data-driven adaptation between user-generated EMG patterns and decoder boundaries. Methods: A 10-session study with 12 able-bodied participants compared PR performance after motor-based training and updating using the Reviewer versus conventional virtual arm visualization. Performance was assessed using a Fitts law task that involved the aperture of the cursor and the control of orientation. Results: Participants trained with the Reviewer achieved higher completion rates, reduced overshoot, and improved path efficiency and throughput compared to the standard visualization group. Significance: The Reviewer introduces decoder-informed motor training, facilitating immediate and consistent PR-based myoelectric control improvements. By iteratively refining control through real-time feedback, this approach reduces reliance on trial-and-error recalibration, enabling a more adaptive, self-correcting training framework. Conclusion: The 3D visual feedback significantly improves PR control in novice operators through structured training, enabling feedback-driven adaptation and reducing reliance on extensive heuristic adjustments.
For 3D face modeling, the recently developed 3D-aware neural rendering methods are able to render photorealistic face images with arbitrary viewing directions. The training of the parametric controllable 3D-aware face models, however, still relies on a large-scale dataset that is lab-collected. To address this issue, this paper introduces "StyleMorpheus", the first style-based neural 3D Morphable Face Model (3DMM) that is trained on in-the-wild images. It inherits 3DMM's disentangled controllability (over face identity, expression, and appearance) but without the need for accurately reconstructed explicit 3D shapes. StyleMorpheus employs an auto-encoder structure. The encoder aims at learning a representative disentangled parametric code space and the decoder improves the disentanglement using shape and appearance-related style codes in the different sub-modules of the network. Furthermore, we fine-tune the decoder through style-based generative adversarial learning to achieve photorealistic 3D rendering quality. The proposed style-based design enables StyleMorpheus to achieve state-of-the-art 3D-aware face reconstruction results, while also allowing disentangled control of the reconstructed face. Our model achieves real-time rendering speed, allowing its use in virtual reality applications. We also demonstrate the capability of the proposed style-based design in face editing applications such as style mixing and color editing. Project homepage: https://github.com/ubc-3d-vision-lab/StyleMorpheus.
Large-scale machine learning models deliver strong performance across a wide range of tasks but come with significant computational and resource constraints. To mitigate these challenges, local smaller models are often deployed alongside larger models, relying on routing and deferral mechanisms to offload complex tasks. However, existing approaches inadequately balance the capabilities of these models, often resulting in unnecessary deferrals or sub-optimal resource usage. In this work we introduce a novel loss function called Gatekeeper for calibrating smaller models in cascade setups. Our approach fine-tunes the smaller model to confidently handle tasks it can perform correctly while deferring complex tasks to the larger model. Moreover, it incorporates a mechanism for managing the trade-off between model performance and deferral accuracy, and is broadly applicable across various tasks and domains without any architectural changes. We evaluate our method on encoder-only, decoder-only, and encoder-decoder architectures. Experiments across image classification, language modeling, and vision-language tasks show that our approach substantially improves deferral performance.