Feiyang Chen

Shengpeng Ji, Jialong Zuo, Minghui Fang et al.

Recently, there has been a growing interest in the field of controllable Text-to-Speech (TTS). While previous studies have relied on users providing specific style factor values based on acoustic knowledge or selecting reference speeches that meet certain requirements, generating speech solely from natural text prompts has emerged as a new challenge for researchers. This challenge arises due to the scarcity of high-quality speech datasets with natural text style prompt and the absence of advanced text-controllable TTS models. In light of this, 1) we propose TextrolSpeech, which is the first large-scale speech emotion dataset annotated with rich text attributes. The dataset comprises 236,220 pairs of style prompt in natural text descriptions with five style factors and corresponding speech samples. Through iterative experimentation, we introduce a multi-stage prompt programming approach that effectively utilizes the GPT model for generating natural style descriptions in large volumes. 2) Furthermore, to address the need for generating audio with greater style diversity, we propose an efficient architecture called Salle. This architecture treats text controllable TTS as a language model task, utilizing audio codec codes as an intermediate representation to replace the conventional mel-spectrogram. Finally, we successfully demonstrate the ability of the proposed model by showing a comparable performance in the controllable TTS task. Audio samples are available at https://sall-e.github.io/

Madhuri Shanbhogue, Zhe Li, Shanfeng Zhang et al.

We introduce Gemini Embedding 2, a native multimodal embedding model that allows embedding video, audio, image, and text modalities in a unified representation space. We leverage the multimodal capabilities of Gemini to produce embeddings for arbitrary combinations of interleaved inputs across all these modalities that generalize well across a wide variety of tasks. Applying large-scale contrastive learning in a multi-task multi-stage training setup, we achieve state-of-the-art performance on key embedding benchmarks including unimodal, cross-modal, and multimodal retrieval spanning a diverse set of tasks. We show that our embedding model demonstrates strong performance (with a score of 62.9 R@1 on MSCOCO, 68.8 NDCG@10 on Vatex, 69.9 on MTEB multilingual and 84.0 on MTEB Code) across a variety of tasks surpassing the performance of specialized models. These unified capabilities make Gemini Embedding 2 a promising candidate for downstream use cases such as RAG, recommendation and search. Furthermore, its robust zero-shot performance across distinct fields - from astronomy and bioscience to fine arts and the culinary arts - establishes it as a highly reliable, out-of-the-box representation even for specialized domains.

Suma Bailis, Jane Friedhoff, Feiyang Chen

This paper introduces Werewolf Arena, a novel framework for evaluating large language models (LLMs) through the lens of the classic social deduction game, Werewolf. In Werewolf Arena, LLMs compete against each other, navigating the game's complex dynamics of deception, deduction, and persuasion. The framework introduces a dynamic turn-taking system based on bidding, mirroring real-world discussions where individuals strategically choose when to speak. We demonstrate the framework's utility through an arena-style tournament featuring Gemini and GPT models. Our results reveal distinct strengths and weaknesses in the models' strategic reasoning and communication. These findings highlight Werewolf Arena's potential as a challenging and scalable LLM benchmark.

Feiyang Chen, Yu Cheng, Lei Wang et al.

Transformers and large language models (LLMs) have revolutionized machine learning, with attention mechanisms at the core of their success. As the landscape of attention variants expands, so too do the challenges of optimizing their performance, particularly across different hardware platforms. Current optimization strategies are often narrowly focused, requiring extensive manual intervention to accommodate changes in model configurations or hardware environments. In this paper, we introduce AttentionEngine, a comprehensive framework designed to streamline the optimization of attention mechanisms across heterogeneous hardware backends. By decomposing attention computation into modular operations with customizable components, AttentionEngine enables flexible adaptation to diverse algorithmic requirements. The framework further automates kernel optimization through a combination of programmable templates and a robust cross-platform scheduling strategy. Empirical results reveal performance gains of up to 10x on configurations beyond the reach of existing methods. AttentionEngine offers a scalable, efficient foundation for developing and deploying attention mechanisms with minimal manual tuning. Our code has been open-sourced and is available at https://github.com/microsoft/AttentionEngine.

Jinglin Liu, Chengxi Li, Yi Ren et al.

Singing voice synthesis (SVS) systems are built to synthesize high-quality and expressive singing voice, in which the acoustic model generates the acoustic features (e.g., mel-spectrogram) given a music score. Previous singing acoustic models adopt a simple loss (e.g., L1 and L2) or generative adversarial network (GAN) to reconstruct the acoustic features, while they suffer from over-smoothing and unstable training issues respectively, which hinder the naturalness of synthesized singing. In this work, we propose DiffSinger, an acoustic model for SVS based on the diffusion probabilistic model. DiffSinger is a parameterized Markov chain that iteratively converts the noise into mel-spectrogram conditioned on the music score. By implicitly optimizing variational bound, DiffSinger can be stably trained and generate realistic outputs. To further improve the voice quality and speed up inference, we introduce a shallow diffusion mechanism to make better use of the prior knowledge learned by the simple loss. Specifically, DiffSinger starts generation at a shallow step smaller than the total number of diffusion steps, according to the intersection of the diffusion trajectories of the ground-truth mel-spectrogram and the one predicted by a simple mel-spectrogram decoder. Besides, we propose boundary prediction methods to locate the intersection and determine the shallow step adaptively. The evaluations conducted on a Chinese singing dataset demonstrate that DiffSinger outperforms state-of-the-art SVS work. Extensional experiments also prove the generalization of our methods on text-to-speech task (DiffSpeech). Audio samples: https://diffsinger.github.io. Codes: https://github.com/MoonInTheRiver/DiffSinger. The old title of this work: "Diffsinger: Diffusion acoustic model for singing voice synthesis".

Jinhyuk Lee, Feiyang Chen, Sahil Dua et al.

In this report, we introduce Gemini Embedding, a state-of-the-art embedding model leveraging the power of Gemini, Google's most capable large language model. Capitalizing on Gemini's inherent multilingual and code understanding capabilities, Gemini Embedding produces highly generalizable embeddings for text spanning numerous languages and textual modalities. The representations generated by Gemini Embedding can be precomputed and applied to a variety of downstream tasks including classification, similarity, clustering, ranking, and retrieval. Evaluated on the Massive Multilingual Text Embedding Benchmark (MMTEB), which includes over one hundred tasks across 250+ languages, Gemini Embedding substantially outperforms prior state-of-the-art models, demonstrating considerable improvements in embedding quality. Achieving state-of-the-art performance across MMTEB's multilingual, English, and code benchmarks, our unified model demonstrates strong capabilities across a broad selection of tasks and surpasses specialized domain-specific models.

Yu Zhang, Rongjie Huang, Ruiqi Li et al.

Style transfer for out-of-domain (OOD) singing voice synthesis (SVS) focuses on generating high-quality singing voices with unseen styles (such as timbre, emotion, pronunciation, and articulation skills) derived from reference singing voice samples. However, the endeavor to model the intricate nuances of singing voice styles is an arduous task, as singing voices possess a remarkable degree of expressiveness. Moreover, existing SVS methods encounter a decline in the quality of synthesized singing voices in OOD scenarios, as they rest upon the assumption that the target vocal attributes are discernible during the training phase. To overcome these challenges, we propose StyleSinger, the first singing voice synthesis model for zero-shot style transfer of out-of-domain reference singing voice samples. StyleSinger incorporates two critical approaches for enhanced effectiveness: 1) the Residual Style Adaptor (RSA) which employs a residual quantization module to capture diverse style characteristics in singing voices, and 2) the Uncertainty Modeling Layer Normalization (UMLN) to perturb the style attributes within the content representation during the training phase and thus improve the model generalization. Our extensive evaluations in zero-shot style transfer undeniably establish that StyleSinger outperforms baseline models in both audio quality and similarity to the reference singing voice samples. Access to singing voice samples can be found at https://aaronz345.github.io/StyleSingerDemo/.

Feiyang Chen, Ziqian Luo, Lisang Zhou et al.

Vision Transformers (ViT) have marked a paradigm shift in computer vision, outperforming state-of-the-art models across diverse tasks. However, their practical deployment is hampered by high computational and memory demands. This study addresses the challenge by evaluating four primary model compression techniques: quantization, low-rank approximation, knowledge distillation, and pruning. We methodically analyze and compare the efficacy of these techniques and their combinations in optimizing ViTs for resource-constrained environments. Our comprehensive experimental evaluation demonstrates that these methods facilitate a balanced compromise between model accuracy and computational efficiency, paving the way for wider application in edge computing devices.

Henrique Schechter Vera, Sahil Dua, Biao Zhang et al.

We introduce EmbeddingGemma, a new lightweight, open text embedding model based on the Gemma 3 language model family. Our innovative training recipe strategically captures knowledge from larger models via encoder-decoder initialization and geometric embedding distillation. We improve model robustness and expressiveness with a spread-out regularizer, and ensure generalizability by merging checkpoints from varied, optimized mixtures. Evaluated on the Massive Text Embedding Benchmark (MTEB) across multilingual, English, and code domains, EmbeddingGemma (300M) achieves state-of-the-art results. Notably, it outperforms prior top models, both proprietary and open, with fewer than 500M parameters, and provides performance comparable to models double its size, offering an exceptional performance-to-cost ratio. Remarkably, this lead persists when quantizing model weights or truncating embedding outputs. This makes EmbeddingGemma particularly well-suited for low-latency and high-throughput use cases such as on-device applications. We provide ablation studies exploring our key design choices. We release EmbeddingGemma to the community to promote further research.

Siyu Jiang, Feiyang Chen, Xiaojin Zhang et al.

Despite the significant progress of Multimodal Large Language Models (MLLMs) across diverse tasks, hallucination -- corresponding to the generation of visually inconsistent objects, attributes, or relations -- remains a major obstacle to their reliable deployment. Unlike pure language models, MLLMs must ground their generation process in visual inputs. However, existing models often suffer from semantic drift during decoding, causing outputs to diverge from visual facts as the sequence length increases. To address this issue, we propose KVSmooth, a training-free and plug-and-play method that mitigates hallucination by performing attention-entropy-guided adaptive smoothing on hidden states. Specifically, KVSmooth applies an exponential moving average (EMA) to both keys and values in the KV-Cache, while dynamically quantifying the sink degree of each token through the entropy of its attention distribution to adaptively adjust the smoothing strength. Unlike computationally expensive retraining or contrastive decoding methods, KVSmooth operates efficiently during inference without additional training or model modification. Extensive experiments demonstrate that KVSmooth significantly reduces hallucination ($\mathit{CHAIR}_{S}$ from $41.8 \rightarrow 18.2$) while improving overall performance ($F_1$ score from $77.5 \rightarrow 79.2$), achieving higher precision and recall simultaneously. In contrast, prior methods often improve one at the expense of the other, validating the effectiveness and generality of our approach.

Zhenghao Lin, Zihao Tang, Xiao Liu et al.

We introduce Sigma, an efficient large language model specialized for the system domain, empowered by a novel architecture including DiffQKV attention, and pre-trained on our meticulously collected system domain data. DiffQKV attention significantly enhances the inference efficiency of Sigma by optimizing the Query (Q), Key (K), and Value (V) components in the attention mechanism differentially, based on their varying impacts on the model performance and efficiency indicators. Specifically, we (1) conduct extensive experiments that demonstrate the model's varying sensitivity to the compression of K and V components, leading to the development of differentially compressed KV, and (2) propose augmented Q to expand the Q head dimension, which enhances the model's representation capacity with minimal impacts on the inference speed. Rigorous theoretical and empirical analyses reveal that DiffQKV attention significantly enhances efficiency, achieving up to a 33.36% improvement in inference speed over the conventional grouped-query attention (GQA) in long-context scenarios. We pre-train Sigma on 6T tokens from various sources, including 19.5B system domain data that we carefully collect and 1T tokens of synthesized and rewritten data. In general domains, Sigma achieves comparable performance to other state-of-arts models. In the system domain, we introduce the first comprehensive benchmark AIMicius, where Sigma demonstrates remarkable performance across all tasks, significantly outperforming GPT-4 with an absolute improvement up to 52.5%.

Feiyang Chen, Hongpeng Pan, Haonan Xu et al.

Diffusion distillation is central to accelerating image and video generation, yet existing methods are fundamentally limited by the denoising process, where step reduction has largely saturated. Partial timestep low-resolution generation can further accelerate inference, but it suffers noticeable quality degradation due to cross-resolution distribution gaps. We propose Cross-Resolution Distribution Matching Distillation (RMD), a novel distillation framework that bridges cross-resolution distribution gaps for high-fidelity, few-step multi-resolution cascaded inference. Specifically, RMD divides the timestep intervals for each resolution using logarithmic signal-to-noise ratio (logSNR) curves, and introduces logSNR-based mapping to compensate for resolution-induced shifts. Distribution matching is conducted along resolution trajectories to reduce the gap between low-resolution generator distributions and the teacher's high-resolution distribution. In addition, a predicted-noise re-injection mechanism is incorporated during upsampling to stabilize training and improve synthesis quality. Quantitative and qualitative results show that RMD preserves high-fidelity generation while accelerating inference across various backbones. Notably, RMD achieves up to 33.4X speedup on SDXL and 25.6X on Wan2.1-14B, while preserving high visual fidelity.

Yixin Song, Zhenliang Xue, Dongliang Wei et al.

While frontier large language models (LLMs) continue to push capability boundaries, their deployment remains confined to GPU-powered cloud infrastructure. We challenge this paradigm with SmallThinker, a family of LLMs natively designed - not adapted - for the unique constraints of local devices: weak computational power, limited memory, and slow storage. Unlike traditional approaches that mainly compress existing models built for clouds, we architect SmallThinker from the ground up to thrive within these limitations. Our innovation lies in a deployment-aware architecture that transforms constraints into design principles. First, We introduce a two-level sparse structure combining fine-grained Mixture-of-Experts (MoE) with sparse feed-forward networks, drastically reducing computational demands without sacrificing model capacity. Second, to conquer the I/O bottleneck of slow storage, we design a pre-attention router that enables our co-designed inference engine to prefetch expert parameters from storage while computing attention, effectively hiding storage latency that would otherwise cripple on-device inference. Third, for memory efficiency, we utilize NoPE-RoPE hybrid sparse attention mechanism to slash KV cache requirements. We release SmallThinker-4B-A0.6B and SmallThinker-21B-A3B, which achieve state-of-the-art performance scores and even outperform larger LLMs. Remarkably, our co-designed system mostly eliminates the need for expensive GPU hardware: with Q4_0 quantization, both models exceed 20 tokens/s on ordinary consumer CPUs, while consuming only 1GB and 8GB of memory respectively. SmallThinker is publicly available at hf.co/PowerInfer/SmallThinker-4BA0.6B-Instruct and hf.co/PowerInfer/SmallThinker-21BA3B-Instruct.

Rongjie Huang, Feiyang Chen, Yi Ren et al.

High-fidelity multi-singer singing voice synthesis is challenging for neural vocoder due to the singing voice data shortage, limited singer generalization, and large computational cost. Existing open corpora could not meet requirements for high-fidelity singing voice synthesis because of the scale and quality weaknesses. Previous vocoders have difficulty in multi-singer modeling, and a distinct degradation emerges when conducting unseen singer singing voice generation. To accelerate singing voice researches in the community, we release a large-scale, multi-singer Chinese singing voice dataset OpenSinger. To tackle the difficulty in unseen singer modeling, we propose Multi-Singer, a fast multi-singer vocoder with generative adversarial networks. Specifically, 1) Multi-Singer uses a multi-band generator to speed up both training and inference procedure. 2) to capture and rebuild singer identity from the acoustic feature (i.e., mel-spectrogram), Multi-Singer adopts a singer conditional discriminator and conditional adversarial training objective. 3) to supervise the reconstruction of singer identity in the spectrum envelopes in frequency domain, we propose an auxiliary singer perceptual loss. The joint training approach effectively works in GANs for multi-singer voices modeling. Experimental results verify the effectiveness of OpenSinger and show that Multi-Singer improves unseen singer singing voices modeling in both speed and quality over previous methods. The further experiment proves that combined with FastSpeech 2 as the acoustic model, Multi-Singer achieves strong robustness in the multi-singer singing voice synthesis pipeline. Samples are available at https://Multi-Singer.github.io/

Rongjie Huang, Chenye Cui, Feiyang Chen et al.

Deep generative models have achieved significant progress in speech synthesis to date, while high-fidelity singing voice synthesis is still an open problem for its long continuous pronunciation, rich high-frequency parts, and strong expressiveness. Existing neural vocoders designed for text-to-speech cannot directly be applied to singing voice synthesis because they result in glitches and poor high-frequency reconstruction. In this work, we propose SingGAN, a generative adversarial network designed for high-fidelity singing voice synthesis. Specifically, 1) to alleviate the glitch problem in the generated samples, we propose source excitation with the adaptive feature learning filters to expand the receptive field patterns and stabilize long continuous signal generation; and 2) SingGAN introduces global and local discriminators at different scales to enrich low-frequency details and promote high-frequency reconstruction; and 3) To improve the training efficiency, SingGAN includes auxiliary spectrogram losses and sub-band feature matching penalty loss. To the best of our knowledge, SingGAN is the first work designed toward high-fidelity singing voice vocoding. Our evaluation of SingGAN demonstrates the state-of-the-art results with higher-quality (MOS 4.05) samples. Also, SingGAN enables a sample speed of 50x faster than real-time on a single NVIDIA 2080Ti GPU. We further show that SingGAN generalizes well to the mel-spectrogram inversion of unseen singers, and the end-to-end singing voice synthesis system SingGAN-SVS enjoys a two-stage pipeline to transform the music scores into expressive singing voices. Audio samples are available at \url{https://SingGAN.github.io/}

Chenye Cui, Yi Ren, Jinglin Liu et al.

Recently, there has been an increasing interest in neural speech synthesis. While the deep neural network achieves the state-of-the-art result in text-to-speech (TTS) tasks, how to generate a more emotional and more expressive speech is becoming a new challenge to researchers due to the scarcity of high-quality emotion speech dataset and the lack of advanced emotional TTS model. In this paper, we first briefly introduce and publicly release a Mandarin emotion speech dataset including 9,724 samples with audio files and its emotion human-labeled annotation. After that, we propose a simple but efficient architecture for emotional speech synthesis called EMSpeech. Unlike those models which need additional reference audio as input, our model could predict emotion labels just from the input text and generate more expressive speech conditioned on the emotion embedding. In the experiment phase, we first validate the effectiveness of our dataset by an emotion classification task. Then we train our model on the proposed dataset and conduct a series of subjective evaluations. Finally, by showing a comparable performance in the emotional speech synthesis task, we successfully demonstrate the ability of the proposed model.

Yifan Wu, Min Gao, Min Zeng et al.

Motivation: Exploring drug-protein interactions (DPIs) work as a pivotal step in drug discovery. The fast expansion of available biological data enables computational methods effectively assist in experimental methods. Among them, deep learning methods extract features only from basic characteristics, such as protein sequences, molecule structures. Others achieve significant improvement by learning from not only sequences/molecules but the protein-protein and drug-drug associations (PPAs and DDAs). The PPAs and DDAs are generally obtained by using computational methods. However, existing computational methods have some limitations, resulting in low-quality PPAs and DDAs that hamper the prediction performance. Therefore, we hope to develop a novel supervised learning method to learn the PPAs and DDAs effectively and thereby improve the prediction performance of the specific task of DPI. Results: In this research, we propose a novel deep learning framework, namely BridgeDPI. BridgeDPI introduces a class of nodes named hyper-nodes, which bridge different proteins/drugs to work as PPAs and DDAs. The hyper-nodes can be supervised learned for the specific task of DPI since the whole process is an end-to-end learning. Consequently, such a model would improve prediction performance of DPI. In three real-world datasets, we further demonstrate that BridgeDPI outperforms state-of-the-art methods. Moreover, ablation studies verify the effectiveness of the hyper-nodes. Last, in an independent verification, BridgeDPI explores the candidate bindings among COVID-19's proteins and various antiviral drugs. And the predictive results accord with the statement of the World Health Organization and Food and Drug Administration, showing the validity and reliability of BridgeDPI.

Feiyang Chen, Ziqian Luo, Yanyan Xu et al.

Sentiment analysis, mostly based on text, has been rapidly developing in the last decade and has attracted widespread attention in both academia and industry. However, the information in the real world usually comes from multiple modalities, such as audio and text. Therefore, in this paper, based on audio and text, we consider the task of multimodal sentiment analysis and propose a novel fusion strategy including both multi-feature fusion and multi-modality fusion to improve the accuracy of audio-text sentiment analysis. We call it the DFF-ATMF (Deep Feature Fusion - Audio and Text Modality Fusion) model, which consists of two parallel branches, the audio modality based branch and the text modality based branch. Its core mechanisms are the fusion of multiple feature vectors and multiple modality attention. Experiments on the CMU-MOSI dataset and the recently released CMU-MOSEI dataset, both collected from YouTube for sentiment analysis, show the very competitive results of our DFF-ATMF model. Furthermore, by virtue of attention weight distribution heatmaps, we also demonstrate the deep features learned by using DFF-ATMF are complementary to each other and robust. Surprisingly, DFF-ATMF also achieves new state-of-the-art results on the IEMOCAP dataset, indicating that the proposed fusion strategy also has a good generalization ability for multimodal emotion recognition.

Feiyang Chen, Ziqian Luo

Audio Sentiment Analysis is a popular research area which extends the conventional text-based sentiment analysis to depend on the effectiveness of acoustic features extracted from speech. However, current progress on audio sentiment analysis mainly focuses on extracting homogeneous acoustic features or doesn't fuse heterogeneous features effectively. In this paper, we propose an utterance-based deep neural network model, which has a parallel combination of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) based network, to obtain representative features termed Audio Sentiment Vector (ASV), that can maximally reflect sentiment information in an audio. Specifically, our model is trained by utterance-level labels and ASV can be extracted and fused creatively from two branches. In the CNN model branch, spectrum graphs produced by signals are fed as inputs while in the LSTM model branch, inputs include spectral features and cepstrum coefficient extracted from dependent utterances in audio. Besides, Bidirectional Long Short-Term Memory (BiLSTM) with attention mechanism is used for feature fusion. Extensive experiments have been conducted to show our model can recognize audio sentiment precisely and quickly, and demonstrate our ASV is better than traditional acoustic features or vectors extracted from other deep learning models. Furthermore, experimental results indicate that the proposed model outperforms the state-of-the-art approach by 9.33\% on Multimodal Opinion-level Sentiment Intensity dataset (MOSI) dataset.

Feiyang Chen, Nan Chen, Hanyang Mao et al.

Although the image recognition has been a research topic for many years, many researchers still have a keen interest in it[1]. In some papers[2][3][4], however, there is a tendency to compare models only on one or two datasets, either because of time restraints or because the model is tailored to a specific task. Accordingly, it is hard to understand how well a certain model generalizes across image recognition field[6]. In this paper, we compare four neural networks on MNIST dataset[5] with different division. Among them, three are Convolutional Neural Networks (CNN)[7], Deep Residual Network (ResNet)[2] and Dense Convolutional Network (DenseNet)[3] respectively, and the other is our improvement on CNN baseline through introducing Capsule Network (CapsNet)[1] to image recognition area. We show that the previous models despite do a quite good job in this area, our retrofitting can be applied to get a better performance. The result obtained by CapsNet is an accuracy rate of 99.75\%, and it is the best result published so far. Another inspiring result is that CapsNet only needs a small amount of data to get excellent performance. Finally, we will apply CapsNet's ability to generalize in other image recognition field in the future.