Chaofan Ding

Junjie Zheng, Huixin Xue, Shihong Ren et al.

Text-to-speech (TTS) and singing voice synthesis (SVS) both aim to generate human vocal audio from symbolic inputs, but they impose different requirements on the generation process. Speech generation relies on flexible, language-driven prosody, whereas singing generation requires explicit melody control and accurate rhythmic alignment. This mismatch makes it challenging to train a single model that can generate both natural speech and controllable singing, since melody-related conditions should strongly constrain singing but should not restrict speech prosody. We present UniVoice, a unified speech and singing voice generation framework based on conditional flow matching. Instead of using a single undifferentiated conditioning representation, UniVoice factorizes the condition into content, melody, and timbre, which are encoded by modality-appropriate encoders and consumed by a shared Diffusion Transformer (DiT) backbone. For singing, the melody condition is represented by MIDI note sequences; for speech, it is replaced with a learned null melody token, allowing the model to infer prosody from linguistic and acoustic context. This design preserves explicit melody control for singing while avoiding the need to impose melody constraints on speech. We further analyze the null melody token as an approximation to melody marginalization in the conditional flow. Trained on 30k hours of speech and 35k hours of singing data, UniVoice achieves a speech PER of 5.26\%, comparable to dedicated TTS systems such as F5-TTS (5.21\%) and CosyVoice3 (5.30\%). On singing generation, UniVoice achieves a PER of 16.22\%, outperforming the unified baseline Vevo1.5 (24.72\%).

Gongyu Chen, Xiaoyu Zhang, Zhenqiang Weng et al.

Singing voice conversion (SVC) aims to render the target singer's timbre while preserving melody and lyrics. However, existing zero-shot SVC systems remain fragile in real songs due to harmony interference, F0 errors, and the lack of inductive biases for singing. We propose YingMusic-SVC, a robust zero-shot framework that unifies continuous pre-training, robust supervised fine-tuning, and Flow-GRPO reinforcement learning. Our model introduces a singing-trained RVC timbre shifter for timbre-content disentanglement, an F0-aware timbre adaptor for dynamic vocal expression, and an energy-balanced rectified flow matching loss to enhance high-frequency fidelity. Experiments on a graded multi-track benchmark show that YingMusic-SVC achieves consistent improvements over strong open-source baselines in timbre similarity, intelligibility, and perceptual naturalness, especially under accompanied and harmony-contaminated conditions, demonstrating its effectiveness for real-world SVC deployment.

Yunming Liang, Zihao Chen, Chaofan Ding et al.

Currently, high-quality, synchronized audio is synthesized from video and optional text inputs using various multi-modal joint learning frameworks. However, the precise alignment between the visual and generated audio domains remains far from satisfactory. One key factor is the lack of sufficient temporal and semantic alignment annotations in open-source video-audio and text-audio benchmarks. Therefore, we propose a framework for audio generation from videos, leveraging the internal chain-of-thought (CoT) of a multi-modal large language model (MLLM) to enable step-by-step reasoning without requiring additional annotations. Additionally, a corresponding multi-modal reasoning dataset is constructed to facilitate the learning of initial reasoning in audio generation. In the experiments, we demonstrate the effectiveness of the proposed framework in reducing misalignment (voice-over) in generated audio and achieving competitive performance compared to various state-of-the-art models. The evaluation results show that the proposed method outperforms state-of-the-art approaches across multiple metrics. Specifically, the F DP aSST indicator is reduced by up to 10.07%, the F DP AN N s indicator by up to 11.62%, and the F DV GG indicator by up to 38.61%. Furthermore, the IS indicator improves by up to 4.95%, the IB-score indicator increases by up to 6.39%, and the DeSync indicator is reduced by up to 0.89%.

Xinhan Di, Zihao Chen, Yunming Liang et al.

Large-scale text-to-speech (TTS) models have made significant progress recently.However, they still fall short in the generation of Chinese dialectal speech. Toaddress this, we propose Bailing-TTS, a family of large-scale TTS models capable of generating high-quality Chinese dialectal speech. Bailing-TTS serves as a foundation model for Chinese dialectal speech generation. First, continual semi-supervised learning is proposed to facilitate the alignment of text tokens and speech tokens. Second, the Chinese dialectal representation learning is developed using a specific transformer architecture and multi-stage training processes. With the proposed design of novel network architecture and corresponding strategy, Bailing-TTS is able to generate Chinese dialectal speech from text effectively and efficiently. Experiments demonstrate that Bailing-TTS generates Chinese dialectal speech towards human-like spontaneous representation. Readers are encouraged to listen to demos at \url{https://c9412600.github.io/bltts_tech_report/index.html}.

Huan Yang, Jiahui Chen, Chaofan Ding et al.

Gestures are pivotal in enhancing co-speech communication. While recent works have mostly focused on point-level motion transformation or fully supervised motion representations through data-driven approaches, we explore the representation of gestures in co-speech, with a focus on self-supervised representation and pixel-level motion deviation, utilizing a diffusion model which incorporates latent motion features. Our approach leverages self-supervised deviation in latent representation to facilitate hand gestures generation, which are crucial for generating realistic gesture videos. Results of our first experiment demonstrate that our method enhances the quality of generated videos, with an improvement from 2.7 to 4.5% for FGD, DIV, and FVD, and 8.1% for PSNR, 2.5% for SSIM over the current state-of-the-art methods.

Junjie Zheng, Chunbo Hao, Guobin Ma et al.

Singing Voice Synthesis (SVS) remains constrained in practical deployment due to its strong dependence on accurate phoneme-level alignment and manually annotated melody contours, requirements that are resource-intensive and hinder scalability. To overcome these limitations, we propose a melody-driven SVS framework capable of synthesizing arbitrary lyrics following any reference melody, without relying on phoneme-level alignment. Our method builds on a Diffusion Transformer (DiT) architecture, enhanced with a dedicated melody extraction module that derives melody representations directly from reference audio. To ensure robust melody encoding, we employ a teacher model to guide the optimization of the melody extractor, alongside an implicit alignment mechanism that enforces similarity distribution constraints for improved melodic stability and coherence. Additionally, we refine duration modeling using weakly annotated song data and introduce a Flow-GRPO reinforcement learning strategy with a multi-objective reward function to jointly enhance pronunciation clarity and melodic fidelity. Experiments show that our model achieves superior performance over existing approaches in both objective measures and subjective listening tests, especially in zero-shot and lyric adaptation settings, while maintaining high audio quality without manual annotation. This work offers a practical and scalable solution for advancing data-efficient singing voice synthesis. To support reproducibility, we release our inference code and model checkpoints.

Haomin Zhang, Kristin Qi, Shuxin Yang et al.

Generating high-quality and temporally synchronized audio from video content is essential for video editing and post-production tasks, enabling the creation of semantically aligned audio for silent videos. However, most existing approaches focus on short-form audio generation for video segments under 10 seconds or rely on noisy datasets for long-form video-to-audio zsynthesis. To address these limitations, we introduce LD-LAudio-V1, an extension of state-of-the-art video-to-audio models and it incorporates dual lightweight adapters to enable long-form audio generation. In addition, we release a clean and human-annotated video-to-audio dataset that contains pure sound effects without noise or artifacts. Our method significantly reduces splicing artifacts and temporal inconsistencies while maintaining computational efficiency. Compared to direct fine-tuning with short training videos, LD-LAudio-V1 achieves significant improvements across multiple metrics: $FD_{\text{passt}}$ 450.00 $\rightarrow$ 327.29 (+27.27%), $FD_{\text{panns}}$ 34.88 $\rightarrow$ 22.68 (+34.98%), $FD_{\text{vgg}}$ 3.75 $\rightarrow$ 1.28 (+65.87%), $KL_{\text{panns}}$ 2.49 $\rightarrow$ 2.07 (+16.87%), $KL_{\text{passt}}$ 1.78 $\rightarrow$ 1.53 (+14.04%), $IS_{\text{panns}}$ 4.17 $\rightarrow$ 4.30 (+3.12%), $IB_{\text{score}}$ 0.25 $\rightarrow$ 0.28 (+12.00%), $Energy\Delta10\text{ms}$ 0.3013 $\rightarrow$ 0.1349 (+55.23%), $Energy\Delta10\text{ms(vs.GT)}$ 0.0531 $\rightarrow$ 0.0288 (+45.76%), and $Sem.\,Rel.$ 2.73 $\rightarrow$ 3.28 (+20.15%). Our dataset aims to facilitate further research in long-form video-to-audio generation and is available at https://github.com/deepreasonings/long-form-video2audio.

Chaoyi Wang, Junjie Zheng, Zihao Chen et al.

Movie dubbing has advanced significantly, yet assessing the real-world effectiveness of these models remains challenging. A comprehensive evaluation benchmark is crucial for two key reasons: 1) Existing metrics fail to fully capture the complexities of dialogue, narration, monologue, and actor adaptability in movie dubbing. 2) A practical evaluation system should offer valuable insights to improve movie dubbing quality and advancement in film production. To this end, we introduce Talking Adaptive Dubbing Benchmarks (TA-Dubbing), designed to improve film production by adapting to dialogue, narration, monologue, and actors in movie dubbing. TA-Dubbing offers several key advantages: 1) Comprehensive Dimensions: TA-Dubbing covers a variety of dimensions of movie dubbing, incorporating metric evaluations for both movie understanding and speech generation. 2) Versatile Benchmarking: TA-Dubbing is designed to evaluate state-of-the-art movie dubbing models and advanced multi-modal large language models. 3) Full Open-Sourcing: We fully open-source TA-Dubbing at https://github.com/woka- 0a/DeepDubber- V1 including all video suits, evaluation methods, annotations. We also continuously integrate new movie dubbing models into the TA-Dubbing leaderboard at https://github.com/woka- 0a/DeepDubber-V1 to drive forward the field of movie dubbing.

Jiahui Chen, Weida Wang, Runhua Shi et al.

While diffusion model for audio-driven avatar video generation have achieved notable process in synthesizing long sequences with natural audio-visual synchronization and identity consistency, the generation of music-performance videos with camera motions remains largely unexplored. We present YingVideo-MV, the first cascaded framework for music-driven long-video generation. Our approach integrates audio semantic analysis, an interpretable shot planning module (MV-Director), temporal-aware diffusion Transformer architectures, and long-sequence consistency modeling to enable automatic synthesis of high-quality music performance videos from audio signals. We construct a large-scale Music-in-the-Wild Dataset by collecting web data to support the achievement of diverse, high-quality results. Observing that existing long-video generation methods lack explicit camera motion control, we introduce a camera adapter module that embeds camera poses into latent noise. To enhance continulity between clips during long-sequence inference, we further propose a time-aware dynamic window range strategy that adaptively adjust denoising ranges based on audio embedding. Comprehensive benchmark tests demonstrate that YingVideo-MV achieves outstanding performance in generating coherent and expressive music videos, and enables precise music-motion-camera synchronization. More videos are available in our project page: https://giantailab.github.io/YingVideo-MV/ .

Shuangtao Li, Shuaihao Dong, Kexin Luan et al.

Large language models (LLMs) have demonstrated their remarkable capacity across a variety of tasks. However, reasoning remains a challenge for LLMs. To improve LLMs' reasoning ability, process supervision has proven to be better than outcome supervision. In this work, we study using Monte Carlo Tree Search (MCTS) to generate process supervision data with LLMs themselves for training them. We sample reasoning steps with an LLM and assign each step a score that captures its "relative correctness," and the LLM is then trained by minimizing weighted log-likelihood of generating the reasoning steps. This generate-then-train process is repeated iteratively until convergence.Our experimental results demonstrate that the proposed methods considerably improve the performance of LLMs on two mathematical reasoning datasets. Furthermore, models trained on one dataset also exhibit improved performance on the other, showing the transferability of the enhanced reasoning ability.

Zihao Chen, Haomin Zhang, Xinhan Di et al.

Generating sound effects for product-level videos, where only a small amount of labeled data is available for diverse scenes, requires the production of high-quality sounds in few-shot settings. To tackle the challenge of limited labeled data in real-world scenes, we introduce YingSound, a foundation model designed for video-guided sound generation that supports high-quality audio generation in few-shot settings. Specifically, YingSound consists of two major modules. The first module uses a conditional flow matching transformer to achieve effective semantic alignment in sound generation across audio and visual modalities. This module aims to build a learnable audio-visual aggregator (AVA) that integrates high-resolution visual features with corresponding audio features at multiple stages. The second module is developed with a proposed multi-modal visual-audio chain-of-thought (CoT) approach to generate finer sound effects in few-shot settings. Finally, an industry-standard video-to-audio (V2A) dataset that encompasses various real-world scenarios is presented. We show that YingSound effectively generates high-quality synchronized sounds across diverse conditional inputs through automated evaluations and human studies. Project Page: \url{https://giantailab.github.io/yingsound/}

Huchen Jiang, Yangyang Ma, Chaofan Ding et al.

With current state-of-the-art approaches aimed at enhancing the reasoning capabilities of Large Language Models(LLMs) through iterative preference learning inspired by AlphaZero, we propose to further enhance the step-wise reasoning capabilities through intrinsic self-correction to some extent. Our work leverages step-wise preference learning to enhance self-verification via reinforcement learning. We initially conduct our work through a two-stage training procedure. At the first stage, the self-correction reasoning ability of an LLM is enhanced through its own predictions, relying entirely on self-generated data within the intrinsic self-correction to some extent. At the second stage, the baseline step-wise preference learning is leveraged via the application of the enhanced self-correct policy achieved at the first stage. In the evaluation of arithmetic reasoning tasks, our approach outperforms OpenMath2-Llama3.1-8B, dart-math-mistral-7b-uniform on MATH with increases in accuracy to 71.34%(+4.18%) and 48.06%(+4.94%) and LLama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.1 on GSM8K with increases in accuracy to 86.76%(+2.00%) and 38.06%(+2.28%).

Haomin Zhang, Chang Liu, Junjie Zheng et al.

Currently, high-quality, synchronized audio is synthesized using various multi-modal joint learning frameworks, leveraging video and optional text inputs. In the video-to-audio benchmarks, video-to-audio quality, semantic alignment, and audio-visual synchronization are effectively achieved. However, in real-world scenarios, speech and audio often coexist in videos simultaneously, and the end-to-end generation of synchronous speech and audio given video and text conditions are not well studied. Therefore, we propose an end-to-end multi-modal generation framework that simultaneously produces speech and audio based on video and text conditions. Furthermore, the advantages of video-to-audio (V2A) models for generating speech from videos remain unclear. The proposed framework, DeepAudio, consists of a video-to-audio (V2A) module, a text-to-speech (TTS) module, and a dynamic mixture of modality fusion (MoF) module. In the evaluation, the proposed end-to-end framework achieves state-of-the-art performance on the video-audio benchmark, video-speech benchmark, and text-speech benchmark. In detail, our framework achieves comparable results in the comparison with state-of-the-art models for the video-audio and text-speech benchmarks, and surpassing state-of-the-art models in the video-speech benchmark, with WER 16.57% to 3.15% (+80.99%), SPK-SIM 78.30% to 89.38% (+14.15%), EMO-SIM 66.24% to 75.56% (+14.07%), MCD 8.59 to 7.98 (+7.10%), MCD SL 11.05 to 9.40 (+14.93%) across a variety of dubbing settings.

Junjie Zheng, Zihao Chen, Chaofan Ding et al.

Current movie dubbing technology can generate the desired voice from a given speech prompt, ensuring good synchronization between speech and visuals while accurately conveying the intended emotions. However, in movie dubbing, key aspects such as adapting to different dubbing styles, handling dialogue, narration, and monologue effectively, and understanding subtle details like the age and gender of speakers, have not been well studied. To address this challenge, we propose a framework of multi-modal large language model. First, it utilizes multimodal Chain-of-Thought (CoT) reasoning methods on visual inputs to understand dubbing styles and fine-grained attributes. Second, it generates high-quality dubbing through large speech generation models, guided by multimodal conditions. Additionally, we have developed a movie dubbing dataset with CoT annotations. The evaluation results demonstrate a performance improvement over state-of-the-art methods across multiple datasets. In particular, for the evaluation metrics, the SPK-SIM and EMO-SIM increases from 82.48% to 89.74%, 66.24% to 78.88% for dubbing setting 2.0 on V2C Animation dataset, LSE-D and MCD-SL decreases from 14.79 to 14.63, 5.24 to 4.74 for dubbing setting 2.0 on Grid dataset, SPK-SIM increases from 64.03 to 83.42 and WER decreases from 52.69% to 23.20% for initial reasoning setting on proposed CoT-Movie-Dubbing dataset in the comparison with the state-of-the art models.

Junjie Zheng, Gongyu Chen, Chaofan Ding et al.

In real-world singing voice conversion (SVC) applications, environmental noise and the demand for expressive output pose significant challenges. Conventional methods, however, are typically designed without accounting for real deployment scenarios, as both training and inference usually rely on clean data. This mismatch hinders practical use, given the inevitable presence of diverse noise sources and artifacts from music separation. To tackle these issues, we propose R2-SVC, a robust and expressive SVC framework. First, we introduce simulation-based robustness enhancement through random fundamental frequency ($F_0$) perturbations and music separation artifact simulations (e.g., reverberation, echo), substantially improving performance under noisy conditions. Second, we enrich speaker representation using domain-specific singing data: alongside clean vocals, we incorporate DNSMOS-filtered separated vocals and public singing corpora, enabling the model to preserve speaker timbre while capturing singing style nuances. Third, we integrate the Neural Source-Filter (NSF) model to explicitly represent harmonic and noise components, enhancing the naturalness and controllability of converted singing. R2-SVC achieves state-of-the-art results on multiple SVC benchmarks under both clean and noisy conditions.

Ziqi Chen, Gongyu Chen, Yihua Wang et al.

Dialect speech embodies rich cultural and linguistic diversity, yet building text-to-speech (TTS) systems for dialects remains challenging due to scarce data, inconsistent orthographies, and complex phonetic variation. To address these issues, we present DiaMoE-TTS, a unified IPA-based framework that standardizes phonetic representations and resolves grapheme-to-phoneme ambiguities. Built upon the F5-TTS architecture, the system introduces a dialect-aware Mixture-of-Experts (MoE) to model phonological differences and employs parameter-efficient adaptation with Low-Rank Adaptors (LoRA) and Conditioning Adapters for rapid transfer to new dialects. Unlike approaches dependent on large-scale or proprietary resources, DiaMoE-TTS enables scalable, open-data-driven synthesis. Experiments demonstrate natural and expressive speech generation, achieving zero-shot performance on unseen dialects and specialized domains such as Peking Opera with only a few hours of data.

Junjie Zheng, Zihao Chen, Chaofan Ding et al.

Current movie dubbing technology can produce the desired speech using a reference voice and input video, maintaining perfect synchronization with the visuals while effectively conveying the intended emotions. However, crucial aspects of movie dubbing, including adaptation to various dubbing styles, effective handling of dialogue, narration, and monologues, as well as consideration of subtle details such as speaker age and gender, remain insufficiently explored. To tackle these challenges, we introduce a multi-modal generative framework. First, it utilizes a multi-modal large vision-language model (VLM) to analyze visual inputs, enabling the recognition of dubbing types and fine-grained attributes. Second, it produces high-quality dubbing using large speech generation models, guided by multi-modal inputs. Additionally, a movie dubbing dataset with annotations for dubbing types and subtle details is constructed to enhance movie understanding and improve dubbing quality for the proposed multi-modal framework. Experimental results across multiple benchmark datasets show superior performance compared to state-of-the-art (SOTA) methods. In details, the LSE-D, SPK-SIM, EMO-SIM, and MCD exhibit improvements of up to 1.09%, 8.80%, 19.08%, and 18.74%, respectively.

Haomin Zhang, Sizhe Shan, Haoyu Wang et al.

Creating high-quality sound effects from videos and text prompts requires precise alignment between visual and audio domains, both semantically and temporally, along with step-by-step guidance for professional audio generation. However, current state-of-the-art video-guided audio generation models often fall short of producing high-quality audio for both general and specialized use cases. To address this challenge, we introduce a multi-stage, multi-modal, end-to-end generative framework with Chain-of-Thought-like (CoT-like) guidance learning, termed Chain-of-Perform (CoP). First, we employ a transformer-based network architecture designed to achieve CoP guidance, enabling the generation of both general and professional audio. Second, we implement a multi-stage training framework that follows step-by-step guidance to ensure the generation of high-quality sound effects. Third, we develop a CoP multi-modal dataset, guided by video, to support step-by-step sound effects generation. Evaluation results highlight the advantages of the proposed multi-stage CoP generative framework compared to the state-of-the-art models on a variety of datasets, with FAD 0.79 to 0.74 (+6.33%), CLIP 16.12 to 17.70 (+9.80%) on VGGSound, SI-SDR 1.98dB to 3.35dB (+69.19%), MOS 2.94 to 3.49(+18.71%) on PianoYT-2h, and SI-SDR 2.22dB to 3.21dB (+44.59%), MOS 3.07 to 3.42 (+11.40%) on Piano-10h.

Gongyu Chen, Haomin Zhang, Chaofan Ding et al.

One fascinating aspect of pre-trained Audio-Language Models (ALMs) learning is their impressive zero-shot generalization capability and test-time adaptation (TTA) methods aiming to improve domain performance without annotations. However, previous test time adaptation (TTA) methods for ALMs in zero-shot classification tend to be stuck in incorrect model predictions. In order to further boost the performance, we propose multiple guidance on prompt learning without annotated labels. First, guidance of consistency on both context tokens and domain tokens of ALMs is set. Second, guidance of both consistency across multiple augmented views of each single test sample and contrastive learning across different test samples is set. Third, we propose a corresponding end-end learning framework for the proposed test-time adaptation method without annotated labels. We extensively evaluate our approach on 12 downstream tasks across domains, our proposed adaptation method leads to 4.41% (max 7.50%) average zero-shot performance improvement in comparison with the state-of-the-art models.

Changqun Li, Chaofan Ding, Kexin Luan et al.

Fine-tuning pre-trained large language models in a parameter-efficient manner is widely studied for its effectiveness and efficiency. LoRA is one of the most widely used methods, which assumes that the optimization process is essentially low dimensional. Although LoRA has demonstrated commendable performance, there remains a significant performance gap between LoRA and full fine-tuning when learning new tasks. In this work, we propose Low-Rank Adaptation with Task-Relevant Feature Enhancement(LoRATRF) for enhancing task-relevant features from the perspective of editing neural network representations. To prioritize task-relevant features, a task-aware filter that selectively extracts valuable knowledge from hidden representations for the target or current task is designed. As the experiments on a vareity of datasets including NLU, commonsense reasoning and mathematical reasoning tasks demonstrates, our method reduces 33.71% parameters and achieves better performance on a variety of datasets in comparison with SOTA low-rank methods.