Xuchen Song

Peiyu Wang, Yichen Wei, Yi Peng et al.

We present Skywork R1V2, a next-generation multimodal reasoning model and a major leap forward from its predecessor, Skywork R1V. At its core, R1V2 introduces a hybrid reinforcement learning paradigm that jointly leverages the Mixed Preference Optimization (MPO) and the Group Relative Policy Optimization (GRPO), which harmonizes reward-model guidance with rule-based strategies, thereby addressing the long-standing challenge of balancing sophisticated reasoning capabilities with broad generalization. To further enhance training efficiency, we propose the Selective Sample Buffer (SSB) mechanism, which effectively addresses the vanishing advantages dilemma inherent in GRPO by prioritizing high-value samples throughout the optimization process. Notably, we observe that excessive reinforcement signals can induce visual hallucinations--a phenomenon we systematically monitor and mitigate through calibrated reward thresholds throughout the training process. Empirical results affirm the exceptional capability of R1V2, with benchmark-leading performances such as 62.6 on OlympiadBench, 78.9 on AIME2024, 63.6 on LiveCodeBench, and 73.6 on MMMU. These results underscore R1V2's superiority over existing open-source models and demonstrate significant progress in closing the performance gap with premier proprietary systems, including Gemini 2.5 and OpenAI-o4-mini. The Skywork R1V2 model weights have been publicly released to promote openness and reproducibility https://huggingface.co/Skywork/Skywork-R1V2-38B.

Yifan Zhang, Liang Hu, Haofeng Sun et al.

Despite recent progress in multimodal agentic systems, existing approaches often treat image manipulation and web search as disjoint capabilities, rely heavily on costly reinforcement learning, and lack planning grounded in real tool-execution traces. To address these limitations, we present Skywork-R1V4, a 30B (A3B) parameter multimodal agentic model that unifies multimodal planning, active image manipulation ("thinking with images"), deep multimodal search, and, most critically, interleaved reasoning that dynamically alternates between visual operations and external knowledge retrieval. Trained solely via supervised fine-tuning on fewer than 30,000 high-quality, planning-execution-consistent trajectories and validated through stepwise consistency filtering, Skywork-R1V4 achieves state-of-the-art results across perception and multimodal search benchmarks: it scores 66.1 on MMSearch and 67.2 on FVQA, surpassing Gemini 2.5 Flash on all 11 metrics. Skywork-R1V4 exhibits emergent long-horizon reasoning at inference time, successfully orchestrating more than 10 tool calls to solve complex, multi-step tasks. Our results demonstrate that sophisticated agentic multimodal intelligence can be achieved through carefully curated supervised learning alone, without any reliance on reinforcement learning.

Qiqi He, Xuchen Song, Weituo Hao et al.

Does popular music from the 60s sound different than that of the 90s? Prior study has shown that there would exist some variations of patterns and regularities related to instrumentation changes and growing loudness across multi-decadal trends. This indicates that perceiving the era of a song from musical features such as audio and artist information is possible. Music era information can be an important feature for playlist generation and recommendation. However, the release year of a song can be inaccessible in many circumstances. This paper addresses a novel task of music era recognition. We formulate the task as a music classification problem and propose solutions based on supervised contrastive learning. An audio-based model is developed to predict the era from audio. For the case where the artist information is available, we extend the audio-based model to take multimodal inputs and develop a framework, called MultiModal Contrastive (MMC) learning, to enhance the training. Experimental result on Million Song Dataset demonstrates that the audio-based model achieves 54% in accuracy with a tolerance of 3-years range; incorporating the artist information with the MMC framework for training leads to 9% improvement further.

Qiuqiang Kong, Bochen Li, Xuchen Song et al.

Automatic music transcription (AMT) is the task of transcribing audio recordings into symbolic representations. Recently, neural network-based methods have been applied to AMT, and have achieved state-of-the-art results. However, many previous systems only detect the onset and offset of notes frame-wise, so the transcription resolution is limited to the frame hop size. There is a lack of research on using different strategies to encode onset and offset targets for training. In addition, previous AMT systems are sensitive to the misaligned onset and offset labels of audio recordings. Furthermore, there are limited researches on sustain pedal transcription on large-scale datasets. In this article, we propose a high-resolution AMT system trained by regressing precise onset and offset times of piano notes. At inference, we propose an algorithm to analytically calculate the precise onset and offset times of piano notes and pedal events. We show that our AMT system is robust to the misaligned onset and offset labels compared to previous systems. Our proposed system achieves an onset F1 of 96.72% on the MAESTRO dataset, outperforming previous onsets and frames system of 94.80%. Our system achieves a pedal onset F1 score of 91.86\%, which is the first benchmark result on the MAESTRO dataset. We have released the source code and checkpoints of our work at https://github.com/bytedance/piano_transcription.

Xiaokun Wang, Peiyu Wang, Jiangbo Pei et al.

We propose Skywork-VL Reward, a multimodal reward model that provides reward signals for both multimodal understanding and reasoning tasks. Our technical approach comprises two key components: First, we construct a large-scale multimodal preference dataset that covers a wide range of tasks and scenarios, with responses collected from both standard vision-language models (VLMs) and advanced VLM reasoners. Second, we design a reward model architecture based on Qwen2.5-VL-7B-Instruct, integrating a reward head and applying multi-stage fine-tuning using pairwise ranking loss on pairwise preference data. Experimental evaluations show that Skywork-VL Reward achieves state-of-the-art results on multimodal VL-RewardBench and exhibits competitive performance on the text-only RewardBench benchmark. Furthermore, preference data constructed based on our Skywork-VL Reward proves highly effective for training Mixed Preference Optimization (MPO), leading to significant improvements in multimodal reasoning capabilities. Our results underscore Skywork-VL Reward as a significant advancement toward general-purpose, reliable reward models for multimodal alignment. Our model has been publicly released to promote transparency and reproducibility.

Max W. Y. Lam, Qiao Tian, Tang Li et al.

Recent progress in music generation has been remarkably advanced by the state-of-the-art MusicLM, which comprises a hierarchy of three LMs, respectively, for semantic, coarse acoustic, and fine acoustic modelings. Yet, sampling with the MusicLM requires processing through these LMs one by one to obtain the fine-grained acoustic tokens, making it computationally expensive and prohibitive for a real-time generation. Efficient music generation with a quality on par with MusicLM remains a significant challenge. In this paper, we present MeLoDy (M for music; L for LM; D for diffusion), an LM-guided diffusion model that generates music audios of state-of-the-art quality meanwhile reducing 95.7% or 99.6% forward passes in MusicLM, respectively, for sampling 10s or 30s music. MeLoDy inherits the highest-level LM from MusicLM for semantic modeling, and applies a novel dual-path diffusion (DPD) model and an audio VAE-GAN to efficiently decode the conditioning semantic tokens into waveform. DPD is proposed to simultaneously model the coarse and fine acoustics by incorporating the semantic information into segments of latents effectively via cross-attention at each denoising step. Our experimental results suggest the superiority of MeLoDy, not only in its practical advantages on sampling speed and infinitely continuable generation, but also in its state-of-the-art musicality, audio quality, and text correlation. Our samples are available at https://Efficient-MeLoDy.github.io/.

Wei-Tsung Lu, Ju-Chiang Wang, Minz Won et al.

Transformers have drawn attention in the MIR field for their remarkable performance shown in natural language processing and computer vision. However, prior works in the audio processing domain mostly use Transformer as a temporal feature aggregator that acts similar to RNNs. In this paper, we propose SpecTNT, a Transformer-based architecture to model both spectral and temporal sequences of an input time-frequency representation. Specifically, we introduce a novel variant of the Transformer-in-Transformer (TNT) architecture. In each SpecTNT block, a spectral Transformer extracts frequency-related features into the frequency class token (FCT) for each frame. Later, the FCTs are linearly projected and added to the temporal embeddings (TEs), which aggregate useful information from the FCTs. Then, a temporal Transformer processes the TEs to exchange information across the time axis. By stacking the SpecTNT blocks, we build the SpecTNT model to learn the representation for music signals. In experiments, SpecTNT demonstrates state-of-the-art performance in music tagging and vocal melody extraction, and shows competitive performance for chord recognition. The effectiveness of SpecTNT and other design choices are further examined through ablation studies.

Ju-Chiang Wang, Jordan B. L. Smith, Wei-Tsung Lu et al.

Music structure analysis (MSA) methods traditionally search for musically meaningful patterns in audio: homogeneity, repetition, novelty, and segment-length regularity. Hand-crafted audio features such as MFCCs or chromagrams are often used to elicit these patterns. However, with more annotations of section labels (e.g., verse, chorus, and bridge) becoming available, one can use supervised feature learning to make these patterns even clearer and improve MSA performance. To this end, we take a supervised metric learning approach: we train a deep neural network to output embeddings that are near each other for two spectrogram inputs if both have the same section type (according to an annotation), and otherwise far apart. We propose a batch sampling scheme to ensure the labels in a training pair are interpreted meaningfully. The trained model extracts features that can be used in existing MSA algorithms. In evaluations with three datasets (HarmonixSet, SALAMI, and RWC), we demonstrate that using the proposed features can improve a traditional MSA algorithm significantly in both intra- and cross-dataset scenarios.

Ju-Chiang Wang, Jordan B. L. Smith, Jitong Chen et al.

This paper presents a novel supervised approach to detecting the chorus segments in popular music. Traditional approaches to this task are mostly unsupervised, with pipelines designed to target some quality that is assumed to define "chorusness," which usually means seeking the loudest or most frequently repeated sections. We propose to use a convolutional neural network with a multi-task learning objective, which simultaneously fits two temporal activation curves: one indicating "chorusness" as a function of time, and the other the location of the boundaries. We also propose a post-processing method that jointly takes into account the chorus and boundary predictions to produce binary output. In experiments using three datasets, we compare our system to a set of public implementations of other segmentation and chorus-detection algorithms, and find our approach performs significantly better.

Jiawen Huang, Ju-Chiang Wang, Jordan B. L. Smith et al.

A music mashup combines audio elements from two or more songs to create a new work. To reduce the time and effort required to make them, researchers have developed algorithms that predict the compatibility of audio elements. Prior work has focused on mixing unaltered excerpts, but advances in source separation enable the creation of mashups from isolated stems (e.g., vocals, drums, bass, etc.). In this work, we take advantage of separated stems not just for creating mashups, but for training a model that predicts the mutual compatibility of groups of excerpts, using self-supervised and semi-supervised methods. Specifically, we first produce a random mashup creation pipeline that combines stem tracks obtained via source separation, with key and tempo automatically adjusted to match, since these are prerequisites for high-quality mashups. To train a model to predict compatibility, we use stem tracks obtained from the same song as positive examples, and random combinations of stems with key and/or tempo unadjusted as negative examples. To improve the model and use more data, we also train on "average" examples: random combinations with matching key and tempo, where we treat them as unlabeled data as their true compatibility is unknown. To determine whether the combined signal or the set of stem signals is more indicative of the quality of the result, we experiment on two model architectures and train them using semi-supervised learning technique. Finally, we conduct objective and subjective evaluations of the system, comparing them to a standard rule-based system.

Xuchen Song, Qiuqiang Kong, Xingjian Du et al.

Music source separation (MSS) is the task of separating a music piece into individual sources, such as vocals and accompaniment. Recently, neural network based methods have been applied to address the MSS problem, and can be categorized into spectrogram and time-domain based methods. However, there is a lack of research of using complementary information of spectrogram and time-domain inputs for MSS. In this article, we propose a CatNet framework that concatenates a UNet separation branch using spectrogram as input and a WavUNet separation branch using time-domain waveform as input for MSS. We propose an end-to-end and fully differentiable system that incorporate spectrogram calculation into CatNet. In addition, we propose a novel mix-audio data augmentation method that randomly mix audio segments from the same source as augmented audio segments for training. Our proposed CatNet MSS system achieves a state-of-the-art vocals separation source distortion ratio (SDR) of 7.54 dB, outperforming MMDenseNet of 6.57 dB evaluated on the MUSDB18 dataset.

Dongyang Dai, Li Chen, Yuping Wang et al.

With the popularity of deep neural network, speech synthesis task has achieved significant improvements based on the end-to-end encoder-decoder framework in the recent days. More and more applications relying on speech synthesis technology have been widely used in our daily life. Robust speech synthesis model depends on high quality and customized data which needs lots of collecting efforts. It is worth investigating how to take advantage of low-quality and low resource voice data which can be easily obtained from the Internet for usage of synthesizing personalized voice. In this paper, the proposed end-to-end speech synthesis model uses both speaker embedding and noise representation as conditional inputs to model speaker and noise information respectively. Firstly, the speech synthesis model is pre-trained with both multi-speaker clean data and noisy augmented data; then the pre-trained model is adapted on noisy low-resource new speaker data; finally, by setting the clean speech condition, the model can synthesize the new speaker's clean voice. Experimental results show that the speech generated by the proposed approach has better subjective evaluation results than the method directly fine-tuning pre-trained multi-speaker speech synthesis model with denoised new speaker data.

Qiuqiang Kong, Yuxuan Wang, Xuchen Song et al.

Source separation is the task to separate an audio recording into individual sound sources. Source separation is fundamental for computational auditory scene analysis. Previous work on source separation has focused on separating particular sound classes such as speech and music. Many of previous work require mixture and clean source pairs for training. In this work, we propose a source separation framework trained with weakly labelled data. Weakly labelled data only contains the tags of an audio clip, without the occurrence time of sound events. We first train a sound event detection system with AudioSet. The trained sound event detection system is used to detect segments that are mostly like to contain a target sound event. Then a regression is learnt from a mixture of two randomly selected segments to a target segment conditioned on the audio tagging prediction of the target segment. Our proposed system can separate 527 kinds of sound classes from AudioSet within a single system. A U-Net is adopted for the separation system and achieves an average SDR of 5.67 dB over 527 sound classes in AudioSet.