Ziyang Ma

Wenxi Chen, Dongya Jia, Yushen Chen et al.

Recently, diffusion models operating on VAE latents or mel-spectrograms have become the dominant paradigm for zero-shot TTS. Although these compressed representations improve generation efficiency, they inevitably suffer from information loss and non-end-to-end training. Theoretically, directly modeling raw waveforms circumvents these issues; however, this direction remains underexplored and is often deemed difficult due to the extremely long sequence length of audio signals. To overcome this, we propose WavTTS, the first raw waveform generative TTS model that substantially narrows the gap with latent-space generative models. Built upon the flow matching with Diffusion Transformer (DiT), WavTTS directly models speech waveforms via a simple patchification strategy, while integrating multi-scale mel-spectrogram supervision to provide perceptual guidance during training. Furthermore, we investigate the impact of prediction targets and noise scheduling in waveform diffusion, and develop an effective schedule design to improve generation quality. Evaluations on open-source benchmarks demonstrate that WavTTS closely approaches the performance of current state-of-the-art latent generative zero-shot TTS models, while substantially outperforming previous end-to-end speech generation models. Our findings demonstrate the feasibility of scaling diffusion-based TTS directly in the waveform space, opening a new direction for end-to-end speech generation.

Keyu An, Qian Chen, Chong Deng et al.

This report introduces FunAudioLLM, a model family designed to enhance natural voice interactions between humans and large language models (LLMs). At its core are two innovative models: SenseVoice, which handles multilingual speech recognition, emotion recognition, and audio event detection; and CosyVoice, which facilitates natural speech generation with control over multiple languages, timbre, speaking style, and speaker identity. SenseVoice-Small delivers exceptionally low-latency ASR for 5 languages, and SenseVoice-Large supports high-precision ASR for over 50 languages, while CosyVoice excels in multi-lingual voice generation, zero-shot in-context learning, cross-lingual voice cloning, and instruction-following capabilities. The models related to SenseVoice and CosyVoice have been open-sourced on Modelscope and Huggingface, along with the corresponding training, inference, and fine-tuning codes released on GitHub. By integrating these models with LLMs, FunAudioLLM enables applications such as speech-to-speech translation, emotional voice chat, interactive podcasts, and expressive audiobook narration, thereby pushing the boundaries of voice interaction technology. Demos are available at https://fun-audio-llm.github.io, and the code can be accessed at https://github.com/FunAudioLLM.

Yexing Du, Youcheng Pan, Ziyang Ma et al.

Multimodal Large Language Models (MLLMs) have achieved significant success in Speech-to-Text Translation (S2TT) tasks. While most existing research has focused on English-centric translation directions, the exploration of many-to-many translation is still limited by the scarcity of parallel data. To address this, we propose a three-stage curriculum learning strategy that leverages the machine translation capabilities of large language models and adapts them to S2TT tasks, enabling effective learning in low-resource settings. We trained MLLMs with varying parameter sizes (3B, 7B, and 32B) and evaluated the proposed strategy using the FLEURS and CoVoST-2 datasets. Experimental results show that the proposed strategy achieves state-of-the-art average performance in $15\times14$ language pairs, requiring fewer than 10 hours of speech data per language to achieve competitive results. The source code and models are released at https://github.com/yxduir/LLM-SRT.

Zhifei Xie, Zihang Liu, Ze An et al.

Audio is an inherently interactive modality, yet today's Large Audio Language Models (LALMs) are offline, and streaming audio models each handle only a single task such as streaming ASR or voice chatting. It is time to unify them into one online LALM: a model that, through an always-on perceive-decide-respond loop, listens to sound, environment, and instructions in real time and reacts on the fly. We formalize this regime as the Audio Interaction Model, and realize it with Audio-Interaction, a unified streaming model that retains offline task execution while adding online general audio instruction following, from dialogue to full voice chatting, deciding when to respond from the semantics of the stream. To enable this, we propose SoundFlow, a framework that instantiates the perceive-decide-respond loop end to end, from data to training to deployment, through streaming-native data construction, comprehension-aware training, and asynchronous low-latency inference for stable real-time interaction. We further construct StreamAudio-2M, a 2.6M-item streaming corpus spanning 7 fundamental abilities and 28 sub-tasks, and Proactive-Sound-Bench for evaluating proactive audio intervention. Across 8 benchmarks, Audio-Interaction preserves competitive performance on mainstream audio tasks while unlocking capabilities inaccessible to offline LALMs, including real-time ASR, streaming audio instruction following, and proactive help.

Yiwei Guo, Chenpeng Du, Ziyang Ma et al.

Although diffusion models in text-to-speech have become a popular choice due to their strong generative ability, the intrinsic complexity of sampling from diffusion models harms their efficiency. Alternatively, we propose VoiceFlow, an acoustic model that utilizes a rectified flow matching algorithm to achieve high synthesis quality with a limited number of sampling steps. VoiceFlow formulates the process of generating mel-spectrograms into an ordinary differential equation conditional on text inputs, whose vector field is then estimated. The rectified flow technique then effectively straightens its sampling trajectory for efficient synthesis. Subjective and objective evaluations on both single and multi-speaker corpora showed the superior synthesis quality of VoiceFlow compared to the diffusion counterpart. Ablation studies further verified the validity of the rectified flow technique in VoiceFlow.

Zhihao Du, Qian Chen, Shiliang Zhang et al.

Recent years have witnessed a trend that large language model (LLM) based text-to-speech (TTS) emerges into the mainstream due to their high naturalness and zero-shot capacity. In this paradigm, speech signals are discretized into token sequences, which are modeled by an LLM with text as prompts and reconstructed by a token-based vocoder to waveforms. Obviously, speech tokens play a critical role in LLM-based TTS models. Current speech tokens are learned in an unsupervised manner, which lacks explicit semantic information and alignment to the text. In this paper, we propose to represent speech with supervised semantic tokens, which are derived from a multilingual speech recognition model by inserting vector quantization into the encoder. Based on the tokens, we further propose a scalable zero-shot TTS synthesizer, CosyVoice, which consists of an LLM for text-to-token generation and a conditional flow matching model for token-to-speech synthesis. Experimental results show that supervised semantic tokens significantly outperform existing unsupervised tokens in terms of content consistency and speaker similarity for zero-shot voice cloning. Moreover, we find that utilizing large-scale data further improves the synthesis performance, indicating the scalable capacity of CosyVoice. To the best of our knowledge, this is the first attempt to involve supervised speech tokens into TTS models.

Yinghao Ma, Anders Øland, Anton Ragni et al.

In recent years, foundation models (FMs) such as large language models (LLMs) and latent diffusion models (LDMs) have profoundly impacted diverse sectors, including music. This comprehensive review examines state-of-the-art (SOTA) pre-trained models and foundation models in music, spanning from representation learning, generative learning and multimodal learning. We first contextualise the significance of music in various industries and trace the evolution of AI in music. By delineating the modalities targeted by foundation models, we discover many of the music representations are underexplored in FM development. Then, emphasis is placed on the lack of versatility of previous methods on diverse music applications, along with the potential of FMs in music understanding, generation and medical application. By comprehensively exploring the details of the model pre-training paradigm, architectural choices, tokenisation, finetuning methodologies and controllability, we emphasise the important topics that should have been well explored, like instruction tuning and in-context learning, scaling law and emergent ability, as well as long-sequence modelling etc. A dedicated section presents insights into music agents, accompanied by a thorough analysis of datasets and evaluations essential for pre-training and downstream tasks. Finally, by underscoring the vital importance of ethical considerations, we advocate that following research on FM for music should focus more on such issues as interpretability, transparency, human responsibility, and copyright issues. The paper offers insights into future challenges and trends on FMs for music, aiming to shape the trajectory of human-AI collaboration in the music realm.

Ziyang Ma, Guanrou Yang, Wenxi Chen et al.

The recent surge in open-source Multimodal Large Language Models (MLLM) frameworks, such as LLaVA, provides a convenient kickoff for artificial intelligence developers and researchers. However, most of the MLLM frameworks take vision as the main input modality, and provide limited in-depth support for the modality of speech, audio, and music. This situation hinders the development of audio-language models, and forces researchers to spend a lot of effort on code writing and hyperparameter tuning. We present SLAM-LLM, an open-source deep learning framework designed to train customized MLLMs, focused on speech, language, audio, and music processing. SLAM-LLM provides a modular configuration of different encoders, projectors, LLMs, and parameter-efficient fine-tuning plugins. SLAM-LLM also includes detailed training and inference recipes for mainstream tasks, along with high-performance checkpoints like LLM-based Automatic Speech Recognition (ASR), Automated Audio Captioning (AAC), and Music Captioning (MC). Some of these recipes have already reached or are nearing state-of-the-art performance, and some relevant techniques have also been accepted by academic papers. We hope SLAM-LLM will accelerate iteration, development, data engineering, and model training for researchers. We are committed to continually pushing forward audio-based MLLMs through this open-source framework, and call on the community to contribute to the LLM-based speech, audio and music processing.

Zhihao Du, Jiaming Wang, Qian Chen et al.

Generative Pre-trained Transformer (GPT) models have achieved remarkable performance on various natural language processing tasks, and have shown great potential as backbones for audio-and-text large language models (LLMs). Previous mainstream audio-and-text LLMs use discrete audio tokens to represent both input and output audio; however, they suffer from performance degradation on tasks such as automatic speech recognition, speech-to-text translation, and speech enhancement over models using continuous speech features. In this paper, we propose LauraGPT, a novel unified audio-and-text GPT-based LLM for audio recognition, understanding, and generation. LauraGPT is a versatile LLM that can process both audio and text inputs and generate outputs in either modalities. We propose a novel data representation that combines continuous and discrete features for audio: LauraGPT encodes input audio into continuous representations using an audio encoder and generates output audio from discrete codec codes. We propose a one-step codec vocoder to overcome the prediction challenge caused by the multimodal distribution of codec tokens. We fine-tune LauraGPT using supervised multi-task learning. Extensive experiments show that LauraGPT consistently achieves comparable to superior performance compared to strong baselines on a wide range of audio tasks related to content, semantics, paralinguistics, and audio-signal analysis, such as automatic speech recognition, speech-to-text translation, text-to-speech synthesis, speech enhancement, automated audio captioning, speech emotion recognition, and spoken language understanding.

Zheng Liang, Zheshu Song, Ziyang Ma et al.

Recently, end-to-end (E2E) automatic speech recognition (ASR) models have made great strides and exhibit excellent performance in general speech recognition. However, there remain several challenging scenarios that E2E models are not competent in, such as code-switching and named entity recognition (NER). Data augmentation is a common and effective practice for these two scenarios. However, the current data augmentation methods mainly rely on audio splicing and text-to-speech (TTS) models, which might result in discontinuous, unrealistic, and less diversified speech. To mitigate these potential issues, we propose a novel data augmentation method by applying the text-based speech editing model. The augmented speech from speech editing systems is more coherent and diversified, also more akin to real speech. The experimental results on code-switching and NER tasks show that our proposed method can significantly outperform the audio splicing and neural TTS based data augmentation systems.

Ziyang Ma, Yakun Song, Chenpeng Du et al.

Dialogue serves as the most natural manner of human-computer interaction (HCI). Recent advancements in speech language models (SLM) have significantly enhanced speech-based conversational AI. However, these models are limited to turn-based conversation, lacking the ability to interact with humans in real-time spoken scenarios, for example, being interrupted when the generated content is not satisfactory. To address these limitations, we explore full duplex modeling (FDM) in interactive speech language models (iSLM), focusing on enhancing real-time interaction and, more explicitly, exploring the quintessential ability of interruption. We introduce a novel model design, namely listening-while-speaking language model (LSLM), an end-to-end system equipped with both listening and speaking channels. Our LSLM employs a token-based decoder-only TTS for speech generation and a streaming self-supervised learning (SSL) encoder for real-time audio input. LSLM fuses both channels for autoregressive generation and detects turn-taking in real time. Three fusion strategies -- early fusion, middle fusion, and late fusion -- are explored, with middle fusion achieving an optimal balance between speech generation and real-time interaction. Two experimental settings, command-based FDM and voice-based FDM, demonstrate LSLM's robustness to noise and sensitivity to diverse instructions. Our results highlight LSLM's capability to achieve duplex communication with minimal impact on existing systems. This study aims to advance the development of interactive speech dialogue systems, enhancing their applicability in real-world contexts.

Ziyang Ma, Zhisheng Zheng, Guanrou Yang et al.

The excellent generalization ability of self-supervised learning (SSL) for speech foundation models has garnered significant attention. HuBERT is a successful example that utilizes offline clustering to convert speech features into discrete units for a masked language modeling pretext task. However, simply clustering features as targets by k-means does not fully inspire the model's performance. In this work, we present an unsupervised method to improve SSL targets. Two models are proposed, MonoBERT and PolyBERT, which leverage context-independent and context-dependent phoneme-based units for pre-training. Our models outperform other SSL models significantly on the LibriSpeech benchmark without the need for iterative re-clustering and re-training. Furthermore, our models equipped with context-dependent units even outperform target-improvement models that use labeled data during pre-training. How we progressively improve the unit discovery process is demonstrated through experiments.

Ziyang Ma, Wen Wu, Zhisheng Zheng et al.

In this paper, we explored how to boost speech emotion recognition (SER) with the state-of-the-art speech pre-trained model (PTM), data2vec, text generation technique, GPT-4, and speech synthesis technique, Azure TTS. First, we investigated the representation ability of different speech self-supervised pre-trained models, and we found that data2vec has a good representation ability on the SER task. Second, we employed a powerful large language model (LLM), GPT-4, and emotional text-to-speech (TTS) model, Azure TTS, to generate emotionally congruent text and speech. We carefully designed the text prompt and dataset construction, to obtain the synthetic emotional speech data with high quality. Third, we studied different ways of data augmentation to promote the SER task with synthetic speech, including random mixing, adversarial training, transfer learning, and curriculum learning. Experiments and ablation studies on the IEMOCAP dataset demonstrate the effectiveness of our method, compared with other data augmentation methods, and data augmentation with other synthetic data.

Ziyang Ma, Zhisheng Zheng, Changli Tang et al.

In this paper, we provide a new perspective on self-supervised speech models from how the training targets are obtained. We generalize the targets extractor into Offline Targets Extractor (Off-TE) and Online Targets Extractor (On-TE). Based on this, we propose a new multi-tasking learning framework for self-supervised learning, MT4SSL, which stands for Boosting Self-Supervised Speech Representation Learning by Integrating Multiple Targets. MT4SSL uses the K-means algorithm as an Off-TE and a teacher network without gradients as an On-TE, respectively. Our model outperforms previous SSL methods by nontrivial margins on the LibriSpeech benchmark, and is comparable to or even better than the best-performing models with fewer data. Furthermore, we find that using both Off-TE and On-TE results in better convergence in the pre-training phase. With both effectiveness and efficiency, we think doing multi-task learning on self-supervised speech models from our perspective is a promising trend.

Ziyang Ma, Dingyi Zhang, Sichu Liang et al.

Although large language model (LLM) based multi-agent systems (MAS) show their capability to solve complex tasks and achieve higher performance over single agent systems, they lead to huge computational overheads because of heavy communication between agents. Previous research has made efforts to train a sparse multi-agent graph or fine-tune a planner to orchestrate the workflow better. However, such extra training processes introduce computational costs and limit MAS to specific domains, therefore compromising their generalizability. In this paper, we propose CONCAT, a training-free multi-agent collaboration framework based on CONsensus and Confidence-driven Ad hoc Teaming to efficiently organize agent interactions. Specifically, agents are clustered based on their initial answers, and leaders of each cluster are selected based on the agents' confidence. Then, a heuristic function based on the Theory of Mind is designed to predict the collaboration benefits between every two leaders according to their answers and confidence. Finally, an ad hoc multi-agent network is organized after evicting a percentage of communications based on the predicted benefits. Experiments across three LLMs and three benchmarks show that CONCAT achieves up to 2.02x higher efficiency (accuracy/latency ratio) than LLM-Debate and outperforms training-aware methods such as AgentDropout, while reducing average latency by 50.1% on Qwen2.5-14B-Instruct, without any task-specific training.

Guanrou Yang, Ziyang Ma, Zhisheng Zheng et al.

Recent years have witnessed significant advancements in self-supervised learning (SSL) methods for speech-processing tasks. Various speech-based SSL models have been developed and present promising performance on a range of downstream tasks including speech recognition. However, existing speech-based SSL models face a common dilemma in terms of computational cost, which might hinder their potential application and in-depth academic research. To address this issue, we first analyze the computational cost of different modules during HuBERT pre-training and then introduce a stack of efficiency optimizations, which is named Fast-HuBERT in this paper. The proposed Fast-HuBERT can be trained in 1.1 days with 8 V100 GPUs on the Librispeech 960h benchmark, without performance degradation, resulting in a 5.2x speedup, compared to the original implementation. Moreover, we explore two well-studied techniques in the Fast-HuBERT and demonstrate consistent improvements as reported in previous work.

Ziyang Ma, Mengsha Liu, Guian Fang et al.

False information can spread quickly on social media, negatively influencing the citizens' behaviors and responses to social events. To better detect all of the fake news, especially long texts which are harder to find completely, a Long-Text Chinese Rumor detection dataset named LTCR is proposed. The LTCR dataset provides a valuable resource for accurately detecting misinformation, especially in the context of complex fake news related to COVID-19. The dataset consists of 1,729 and 500 pieces of real and fake news, respectively. The average lengths of real and fake news are approximately 230 and 152 characters. We also propose \method, Salience-aware Fake News Detection Model, which achieves the highest accuracy (95.85%), fake news recall (90.91%) and F-score (90.60%) on the dataset. (https://github.com/Enderfga/DoubleCheck)

Xie Chen, Ziyang Ma, Changli Tang et al.

Recent years have witnessed a boom in self-supervised learning (SSL) in various areas including speech processing. Speech based SSL models present promising performance in a range of speech related tasks. However, the training of SSL models is computationally expensive and a common practice is to fine-tune a released SSL model on the specific task. It is essential to use consistent front-end input during pre-training and fine-tuning. This consistency may introduce potential issues when the optimal front-end is not the same as that used in pre-training. In this paper, we propose a simple but effective front-end adapter to address this front-end discrepancy. By minimizing the distance between the outputs of different front-ends, the filterbank feature (Fbank) can be compatible with SSL models which are pre-trained with waveform. The experiment results demonstrate the effectiveness of our proposed front-end adapter on several popular SSL models for the speech recognition task.

Yujin Wang, Changli Tang, Ziyang Ma et al.

Recent years have witnessed great strides in self-supervised learning (SSL) on the speech processing. The SSL model is normally pre-trained on a great variety of unlabelled data and a large model size is preferred to increase the modeling capacity. However, this might limit its potential applications due to the expensive computation and memory costs introduced by the oversize model. Miniaturization for SSL models has become an important research direction of practical value. To this end, we explore the effective distillation of HuBERT-based SSL models for automatic speech recognition (ASR). First, in order to establish a strong baseline, a comprehensive study on different student model structures is conducted. On top of this, as a supplement to the regression loss widely adopted in previous works, a discriminative loss is introduced for HuBERT to enhance the distillation performance, especially in low-resource scenarios. In addition, we design a simple and effective algorithm to distill the front-end input from waveform to Fbank feature, resulting in 17% parameter reduction and doubling inference speed, at marginal performance degradation.

Guanrou Yang, Tian Tan, Qian Chen et al.

Integrating speech understanding and generation is a pivotal step toward building unified speech models. However, the different representations required for these two tasks currently pose significant compatibility challenges. Typically, semantics-oriented features are learned from self-supervised learning (SSL), and acoustic-oriented features from reconstruction. Such fragmented representations hinder the realization of truly unified speech systems. We present WavCube, a compact continuous latent derived from an SSL speech encoder that simultaneously supports speech understanding, reconstruction, and generation. WavCube employs a two-stage training scheme. Stage 1 trains a semantic bottleneck to filter off-manifold redundancy that makes raw SSL features intractable for diffusion. Stage 2 injects fine-grained acoustic details via end-to-end reconstruction, while a semantic anchoring loss ensures the representation remains grounded within its original semantic manifold. Comprehensive experiments show that WavCube closely approaches WavLM performance on SUPERB despite an 8x dimensional compression, attains reconstruction quality on par with existing acoustic representations, delivers state-of-the-art zero-shot TTS performance with markedly faster training convergence, and excels in speech enhancement, separation, and voice conversion tasks on the SUPERB-SG benchmark. Systematic ablations reveal that WavCube's two-stage recipe resolves two intrinsic flaws of SSL features for generative modeling, paving the way for future unified speech systems. Codes and checkpoints are available at https://github.com/yanghaha0908/WavCube.

Yizhou Peng, Ziyang Ma, Changsong Liu et al.

Cascaded Automatic Speech Recognition -- Large Language Model (ASR-LLM) pipelines remain popular for industrial Spoken Dialogue Systems (SDS), primarily because their decoupled design ensures perceptual verifiability. However, cascaded systems suffer from error propagation, as transcription failures inevitably cascade to subsequent components, thereby degrading the final interaction quality. Although ASR confidence scores offer a simple filter for unreliable inputs, this approach is fundamentally limited because it typically fails to detect deletion errors or to distinguish between acoustic (inability to hear clearly) and linguistic (inability to understand) mismatches, both of which require targeted recovery strategies. In this paper, we propose a cause-aware error recovery paradigm that fundamentally rethinks robustness in SDS. Unlike traditional confidence filtering, we introduce a suite of small precision-focused detectors that exploit deep ASR latent representations to disentangle token-level errors into perception, comprehension, and deletion failures. This fine-grained diagnostic intelligence empowers the LLM to orchestrate targeted, multi-turn clarification strategies, effectively transforming ambiguous signals into seamless user interactions. Experimental results validate the precision of our approach, which more than doubles the recall on domain-shift errors (57.96% vs. 23.66%) compared to baselines. Crucially, this diagnostic precision yields up to a 30% reduction in WER and a 17% improvement on the downstream task across diverse accents, distortions, and domains.

Haina Zhu, Yao Xiao, Xiquan Li et al.

We study the fine-grained text-to-audio (T2A) generation task. While recent models can synthesize high-quality audio from text descriptions, they often lack precise control over attributes such as loudness, pitch, and sound events. Unlike prior approaches that retrain models for specific control types, we propose to train ControlNet models on top of pre-trained T2A backbones to achieve controllable generation over loudness, pitch, and event roll. We introduce two designs, T2A-ControlNet and T2A-Adapter, and show that the T2A-Adapter model offers a more efficient structure with strong control ability. With only 38M additional parameters, T2A-Adapter achieves state-of-the-art performance on the AudioSet-Strong in both event-level and segment-level F1 scores. We further extend this framework to audio editing, proposing T2A-Editor for removing and inserting audio events at time locations specified by instructions. Models, code, dataset pipelines, and benchmarks will be released to support future research on controllable audio generation and editing.

Pengchao Feng, Yao Xiao, Ziyang Ma et al.

Recent advances in text-to-speech (TTS) have yielded remarkable improvements in naturalness and intelligibility. Building on these achievements, research has increasingly shifted toward enhancing the expressiveness of generated speech, such as dialectal and emotional TTS. However, cross-style synthesis combining both dialect and emotion remains challenging and largely unexplored, mainly due to the scarcity of dialectal data with emotional labels. To address this, we propose Hierarchical Expressive Vector (HE-Vector), a two-stage method for Emotional Dialectal TTS. In the first stage, we construct different task vectors to model dialectal and emotional styles independently, and then enhance single-style synthesis by adjusting their weights, a method we refer to as Expressive Vector (E-Vector). For the second stage, we hierarchically integrate these vectors to achieve controllable emotionally expressive dialect synthesis without requiring jointly labeled data, corresponding to Hierarchical Expressive Vector (HE-Vector). Experimental results demonstrate that HE-Vectors achieve superior performance in dialect synthesis, and promising results in synthesizing emotionally expressive dialectal speech in a zero-shot setting.

Xinsheng Wang, Mingqi Jiang, Ziyang Ma et al.

Recent advancements in large language models (LLMs) have driven significant progress in zero-shot text-to-speech (TTS) synthesis. However, existing foundation models rely on multi-stage processing or complex architectures for predicting multiple codebooks, limiting efficiency and integration flexibility. To overcome these challenges, we introduce Spark-TTS, a novel system powered by BiCodec, a single-stream speech codec that decomposes speech into two complementary token types: low-bitrate semantic tokens for linguistic content and fixed-length global tokens for speaker attributes. This disentangled representation, combined with the Qwen2.5 LLM and a chain-of-thought (CoT) generation approach, enables both coarse-grained control (e.g., gender, speaking style) and fine-grained adjustments (e.g., precise pitch values, speaking rate). To facilitate research in controllable TTS, we introduce VoxBox, a meticulously curated 100,000-hour dataset with comprehensive attribute annotations. Extensive experiments demonstrate that Spark-TTS not only achieves state-of-the-art zero-shot voice cloning but also generates highly customizable voices that surpass the limitations of reference-based synthesis. Source code, pre-trained models, and audio samples are available at https://github.com/SparkAudio/Spark-TTS.

Ruibin Yuan, Hanfeng Lin, Yi Wang et al.

While Large Language Models (LLMs) demonstrate impressive capabilities in text generation, we find that their ability has yet to be generalized to music, humanity's creative language. We introduce ChatMusician, an open-source LLM that integrates intrinsic musical abilities. It is based on continual pre-training and finetuning LLaMA2 on a text-compatible music representation, ABC notation, and the music is treated as a second language. ChatMusician can understand and generate music with a pure text tokenizer without any external multi-modal neural structures or tokenizers. Interestingly, endowing musical abilities does not harm language abilities, even achieving a slightly higher MMLU score. Our model is capable of composing well-structured, full-length music, conditioned on texts, chords, melodies, motifs, musical forms, etc, surpassing GPT-4 baseline. On our meticulously curated college-level music understanding benchmark, MusicTheoryBench, ChatMusician surpasses LLaMA2 and GPT-3.5 on zero-shot setting by a noticeable margin. Our work reveals that LLMs can be an excellent compressor for music, but there remains significant territory to be conquered. We release our 4B token music-language corpora MusicPile, the collected MusicTheoryBench, code, model and demo in GitHub.

Yakun Song, Zhuo Chen, Xiaofei Wang et al.

The language model (LM) approach based on acoustic and linguistic prompts, such as VALL-E, has achieved remarkable progress in the field of zero-shot audio generation. However, existing methods still have some limitations: 1) repetitions, transpositions, and omissions in the output synthesized speech due to limited alignment constraints between audio and phoneme tokens; 2) challenges of fine-grained control over the synthesized speech with autoregressive (AR) language model; 3) infinite silence generation due to the nature of AR-based decoding, especially under the greedy strategy. To alleviate these issues, we propose ELLA-V, a simple but efficient LM-based zero-shot text-to-speech (TTS) framework, which enables fine-grained control over synthesized audio at the phoneme level. The key to ELLA-V is interleaving sequences of acoustic and phoneme tokens, where phoneme tokens appear ahead of the corresponding acoustic tokens. The experimental findings reveal that our model outperforms VALL-E in terms of accuracy and delivers more stable results using both greedy and sampling-based decoding strategies. The code of ELLA-V will be open-sourced after cleanups. Audio samples are available at https://ereboas.github.io/ELLAV/.

Haina Zhu, Yizhi Zhou, Hangting Chen et al.

Recent years have witnessed the success of foundation models pre-trained with self-supervised learning (SSL) in various music informatics understanding tasks, including music tagging, instrument classification, key detection, and more. In this paper, we propose a self-supervised music representation learning model for music understanding. Distinguished from previous studies adopting random projection or existing neural codec, the proposed model, named MuQ, is trained to predict tokens generated by Mel Residual Vector Quantization (Mel-RVQ). Our Mel-RVQ utilizes residual linear projection structure for Mel spectrum quantization to enhance the stability and efficiency of target extraction and lead to better performance. Experiments in a large variety of downstream tasks demonstrate that MuQ outperforms previous self-supervised music representation models with only 0.9K hours of open-source pre-training data. Scaling up the data to over 160K hours and adopting iterative training consistently improve the model performance. To further validate the strength of our model, we present MuQ-MuLan, a joint music-text embedding model based on contrastive learning, which achieves state-of-the-art performance in the zero-shot music tagging task on the MagnaTagATune dataset. Code and checkpoints are open source in https://github.com/tencent-ailab/MuQ.

Zheng Lian, Haiyang Sun, Licai Sun et al.

Multimodal emotion recognition is an important research topic in artificial intelligence. Over the past few decades, researchers have made remarkable progress by increasing the dataset size and building more effective algorithms. However, due to problems such as complex environments and inaccurate annotations, current systems are hard to meet the demands of practical applications. Therefore, we organize the MER series of competitions to promote the development of this field. Last year, we launched MER2023, focusing on three interesting topics: multi-label learning, noise robustness, and semi-supervised learning. In this year's MER2024, besides expanding the dataset size, we further introduce a new track around open-vocabulary emotion recognition. The main purpose of this track is that existing datasets usually fix the label space and use majority voting to enhance the annotator consistency. However, this process may lead to inaccurate annotations, such as ignoring non-majority or non-candidate labels. In this track, we encourage participants to generate any number of labels in any category, aiming to describe emotional states as accurately as possible. Our baseline code relies on MERTools and is available at: https://github.com/zeroQiaoba/MERTools/tree/master/MER2024.

Jin Xu, Zhifang Guo, Hangrui Hu et al. · pku

We present Qwen3-Omni, a single multimodal model that, for the first time, maintains state-of-the-art performance across text, image, audio, and video without any degradation relative to single-modal counterparts. Qwen3-Omni matches the performance of same-sized single-modal models within the Qwen series and excels particularly on audio tasks. Across 36 audio and audio-visual benchmarks, Qwen3-Omni achieves open-source SOTA on 32 benchmarks and overall SOTA on 22, outperforming strong closed-source models such as Gemini-2.5-Pro, Seed-ASR, and GPT-4o-Transcribe. Qwen3-Omni adopts a Thinker-Talker MoE architecture that unifies perception and generation across text, images, audio, and video, yielding fluent text and natural real-time speech. It supports text interaction in 119 languages, speech understanding in 19 languages, and speech generation in 10 languages. To reduce first-packet latency in streaming synthesis, Talker autoregressively predicts discrete speech codecs using a multi-codebook scheme. Leveraging the representational capacity of these codebooks, we replace computationally intensive block-wise diffusion with a lightweight causal ConvNet, enabling streaming from the first codec frame. In cold-start settings, Qwen3-Omni achieves a theoretical end-to-end first-packet latency of 234 ms. To further strengthen multimodal reasoning, we introduce a Thinking model that explicitly reasons over inputs from any modality. Since the research community currently lacks a general-purpose audio captioning model, we fine-tuned Qwen3-Omni-30B-A3B to obtain Qwen3-Omni-30B-A3B-Captioner, which produces detailed, low-hallucination captions for arbitrary audio inputs. Qwen3-Omni-30B-A3B, Qwen3-Omni-30B-A3B-Thinking, and Qwen3-Omni-30B-A3B-Captioner are publicly released under the Apache 2.0 license.

Xingwei Qu, Yuelin Bai, Yinghao Ma et al.

In this paper, we explore the application of Large Language Models (LLMs) to the pre-training of music. While the prevalent use of MIDI in music modeling is well-established, our findings suggest that LLMs are inherently more compatible with ABC Notation, which aligns more closely with their design and strengths, thereby enhancing the model's performance in musical composition. To address the challenges associated with misaligned measures from different tracks during generation, we propose the development of a Synchronized Multi-Track ABC Notation (SMT-ABC Notation), which aims to preserve coherence across multiple musical tracks. Our contributions include a series of models capable of handling up to 8192 tokens, covering 90% of the symbolic music data in our training set. Furthermore, we explore the implications of the Symbolic Music Scaling Law (SMS Law) on model performance. The results indicate a promising direction for future research in music generation, offering extensive resources for community-led research through our open-source contributions.

Guanrou Yang, Chen Yang, Qian Chen et al.

Human speech goes beyond the mere transfer of information; it is a profound exchange of emotions and a connection between individuals. While Text-to-Speech (TTS) models have made huge progress, they still face challenges in controlling the emotional expression in the generated speech. In this work, we propose EmoVoice, a novel emotion-controllable TTS model that exploits large language models (LLMs) to enable fine-grained freestyle natural language emotion control, and a phoneme boost variant design that makes the model output phoneme tokens and audio tokens in parallel to enhance content consistency, inspired by chain-of-thought (CoT) and chain-of-modality (CoM) techniques. Besides, we introduce EmoVoice-DB, a high-quality 40-hour English emotion dataset featuring expressive speech and fine-grained emotion labels with natural language descriptions. EmoVoice achieves state-of-the-art performance on the English EmoVoice-DB test set using only synthetic training data, and on the Chinese Secap test set using our in-house data. We further investigate the reliability of existing emotion evaluation metrics and their alignment with human perceptual preferences, and explore using SOTA multimodal LLMs GPT-4o-audio and Gemini to assess emotional speech. Dataset, code, checkpoints, and demo samples are available at https://github.com/yanghaha0908/EmoVoice.

Liumeng Xue, Weizhen Bian, Jiahao Pan et al.

Non-verbal vocalizations (NVVs) like laugh, sigh, and sob are essential for human-like speech, yet standardized evaluation remains limited in jointly assessing whether systems can generate the intended NVVs, place them correctly, and keep them salient without harming speech. We present Non-verbal Vocalization Benchmark (NVBench), a bilingual (English/Chinese) benchmark that evaluates speech synthesis with NVVs. NVBench pairs a unified 45-type taxonomy with a curated bilingual dataset and introduces a multi-axis protocol that separates general speech naturalness and quality from NVV-specific controllability, placement, and salience. We benchmark 15 TTS systems using objective metrics, listening tests, and an LLM-based multi-rater evaluation. Results reveal that NVVs controllability often decouples from quality, while low-SNR oral cues and long-duration affective NVVs remain persistent bottlenecks. NVBench enables fair cross-system comparison across diverse control interfaces under a unified, standardized framework.

Suqing Wang, Ziyang Ma, Xinyi Li et al.

Large Language Models (LLMs) have rapidly advanced and are widely adopted across diverse fields. Due to the substantial computational cost and data requirements of training from scratch, many developers choose to fine-tune or modify existing open-source models. While most adhere to open-source licenses, some falsely claim original training despite clear derivation from public models. This raises pressing concerns about intellectual property protection and highlights the need for reliable methods to verify model provenance. In this paper, we propose GhostSpec, a lightweight yet effective method for verifying LLM lineage without access to training data or modification of model behavior. Our approach constructs compact and robust fingerprints by applying singular value decomposition (SVD) to invariant products of internal attention weight matrices, effectively capturing the structural identity of a model. Unlike watermarking or output-based methods, GhostSpec is fully data-free, non-invasive, and computationally efficient. It demonstrates strong robustness to sequential fine-tuning, pruning, block expansion, and even adversarial transformations. Extensive experiments show that GhostSpec can reliably trace the lineage of transformed models with minimal overhead. By offering a practical solution for model verification and reuse tracking, our method contributes to the protection of intellectual property and fosters a transparent, trustworthy ecosystem for large-scale language models.

Ruiqi Yan, Xiquan Li, Wenxi Chen et al.

Recent advances in large language models (LLMs) have driven significant progress in end-to-end spoken dialogue models (SDMs). In contrast to text-based LLMs, the evaluation framework for SDMs should encompass both cognitive dimensions (e.g., logical reasoning, knowledge) and speech-related aspects (e.g., paralinguistic cues, audio quality). However, there is still a lack of comprehensive evaluations for SDMs in speech-to-speech (S2S) scenarios. To address this gap, we propose URO-Bench, an extensive benchmark for SDMs. Notably, URO-Bench is the first S2S benchmark that covers evaluations about multilingualism, multi-round dialogues, and paralinguistics. Our benchmark is divided into two difficulty levels: basic track and pro track, each comprising 20 test sets, evaluating the spoken dialogue model's abilities in Understanding, Reasoning, and Oral conversation. Evaluations on our proposed benchmark reveal that current open-source SDMs perform rather well in daily QA tasks, but lag behind their backbone LLMs in terms of instruction-following ability and also suffer from catastrophic forgetting. Their performance in advanced evaluations of paralinguistic information and audio understanding remains subpar, highlighting the need for further research in this direction. We hope that URO-Bench can facilitate the development of spoken dialogue models by providing a multifaceted evaluation of existing models and helping to track progress in this area.

Kai Li, Can Shen, Yile Liu et al.

Audio Large Language Models (ALLMs) have gained widespread adoption, yet their trustworthiness remains underexplored. Existing evaluation frameworks, designed primarily for text, fail to address unique vulnerabilities introduced by audio's acoustic properties. We identify significant trustworthiness risks in ALLMs arising from non-semantic acoustic cues, including timbre, accent, and background noise, which can manipulate model behavior. We propose AudioTrust, a comprehensive framework for systematic evaluation of ALLM trustworthiness across audio-specific risks. AudioTrust encompasses six key dimensions: fairness, hallucination, safety, privacy, robustness, and authentication. The framework implements 26 distinct sub-tasks using a curated dataset of over 4,420 audio samples from real-world scenarios, including daily conversations, emergency calls, and voice assistant interactions. We conduct comprehensive evaluations across 18 experimental configurations using human-validated automated pipelines. Our evaluation of 14 state-of-the-art open-source and closed-source ALLMs reveals significant limitations when confronted with diverse high-risk audio scenarios, providing insights for secure deployment of audio models. Code and data are available at https://github.com/JusperLee/AudioTrust.

Yakun Song, Jiawei Chen, Xiaobin Zhuang et al.

Neural audio codecs have made significant strides in efficiently mapping raw audio waveforms into discrete token representations, which are foundational for contemporary audio generative models. However, most existing codecs are optimized primarily for reconstruction quality, often at the expense of the downstream modelability of the encoded tokens. Motivated by the need to overcome this bottleneck, we introduce $\textbf{MagiCodec}$, a novel single-layer, streaming Transformer-based audio codec. MagiCodec is designed with a multistage training pipeline that incorporates Gaussian noise injection and latent regularization, explicitly targeting the enhancement of semantic expressiveness in the generated codes while preserving high reconstruction fidelity. We analytically derive the effect of noise injection in the frequency domain, demonstrating its efficacy in attenuating high-frequency components and fostering robust tokenization. Extensive experimental evaluations show that MagiCodec surpasses state-of-the-art codecs in both reconstruction quality and downstream tasks. Notably, the tokens produced by MagiCodec exhibit Zipf-like distributions, as observed in natural languages, thereby improving compatibility with language-model-based generative architectures. The code and pre-trained models are available at https://github.com/Ereboas/MagiCodec.

Pengchao Feng, Ziyang Ma, Wenxi Chen et al.

End-to-end speech-to-speech (S2S) dialogue systems have recently garnered increasing research attention for their lower latency and more natural integration of nonverbal cues such as emotion and speaker identity. However, these systems face key challenges, particularly in incorporating external knowledge, a capability commonly addressed by Retrieval-Augmented Generation (RAG) in text-based large language models (LLMs). The core difficulty lies in the modality gap between input speech and retrieved textual knowledge, which hinders effective integration of information. To address this issue, we propose a novel end-to-end RAG framework that directly retrieves relevant textual knowledge from speech queries. Experimental results demonstrate that our method significantly improves the performance of end-to-end S2S dialogue systems while achieving higher retrieval efficiency. Although the overall performance still lags behind the SOTA cascaded models, our framework offers a promising direction for enhancing knowledge integration in end-to-end S2S systems. Our code and dataset are released.

Xiquan Li, Junxi Liu, Wenxi Chen et al.

Reinforcement Learning (RL) has become an effective paradigm for enhancing Large Language Models (LLMs) and visual generative models. However, its application in text-to-audio (TTA) generation remains largely under-explored. Prior work typically employs offline methods like Direct Preference Optimization (DPO) and leverages Contrastive Language-Audio Pretraining (CLAP) models as reward functions. In this study, we investigate the integration of online Group Relative Policy Optimization (GRPO) into TTA generation. We adapt the algorithm for Flow Matching-based audio models and demonstrate that online RL significantly outperforms its offline counterparts. Furthermore, we incorporate rewards derived from Large Audio Language Models (LALMs), which can provide fine-grained scoring signals that are better aligned with human perception. With only 470M parameters, our final model, \textbf{Resonate}, establishes a new SOTA on TTA-Bench in terms of both audio quality and semantic alignment.

Bohan Li, Wenbin Huang, Yuhang Qiu et al.

Large Audio Language Models (LALMs), which couple acoustic perception with large language models (LLMs) to extract and understand diverse information from audio, have attracted intense interest from both academic and industrial communities. However, existing LALMs are highly sensitive to how instructions are phrased, affecting both (i) instruction-following rates and (ii) task performance. Yet, no existing benchmarks offer a systematic and comprehensive evaluation of this sensitivity. We introduce ISA-Bench, a dynamic benchmark evaluating instruction sensitivity for LALMs along three axes: instruction description, output format, and task composition. We assess recent open-source and proprietary LALMs using ISA-Bench, profiling both compliance and accuracy under controlled instruction variations. Experimental results reveal that even state-of-the-art LALMs suffer significant instruction sensitivity, leading to degraded performance on fundamental audio understanding tasks. To mitigate this issue, we fine-tune Qwen2-Audio on a specifically constructed complex instruction-variant dataset, achieving a marked improvement in instruction-following performance. However, this also induces nontrivial catastrophic forgetting: the model loses some previously mastered task capabilities when exposed to new instruction styles. Our benchmark provides a standardized basis for assessing and improving instruction sensitivity in LALMs, underscoring the need for instruction-robust audio understanding in real-world pipelines.

Wenming Tu, Guanrou Yang, Ruiqi Yan et al.

Spoken dialogue models currently lack the ability for fine-grained speech style control, a critical capability for human-like interaction that is often overlooked in favor of purely functional capabilities like reasoning and question answering. To address this limitation, we introduce UltraVoice, the first large-scale speech dialogue dataset engineered for multiple fine-grained speech style control. Encompassing over 830 hours of speech dialogues, UltraVoice provides instructions across six key speech stylistic dimensions: emotion, speed, volume, accent, language, and composite styles. Fine-tuning leading models such as SLAM-Omni and VocalNet on UltraVoice significantly enhances their fine-grained speech stylistic controllability without degrading core conversational abilities. Specifically, our fine-tuned models achieve improvements of 29.12-42.33% in Mean Opinion Score (MOS) and 14.61-40.09 percentage points in Instruction Following Rate (IFR) on multi-dimensional control tasks designed in the UltraVoice. Moreover, on the URO-Bench benchmark, our fine-tuned models demonstrate substantial gains in core understanding, reasoning, and conversational abilities, with average improvements of +10.84% on the Basic setting and +7.87% on the Pro setting. Furthermore, the dataset's utility extends to training controllable Text-to-Speech (TTS) models, underscoring its high quality and broad applicability for expressive speech synthesis. The complete dataset and model checkpoints are available at: https://github.com/bigai-nlco/UltraVoice.

Ziyang Ma, Ruiyang Xu, Zhenghao Xing et al.

Fine-grained perception of multimodal information is critical for advancing human-AI interaction. With recent progress in audio-visual technologies, Omni Language Models (OLMs), capable of processing audio and video signals in parallel, have emerged as a promising paradigm for achieving richer understanding and reasoning. However, their capacity to capture and describe fine-grained details remains limited explored. In this work, we present a systematic and comprehensive investigation of omni detailed perception from the perspectives of the data pipeline, models, and benchmark. We first identify an inherent "co-growth" between detail and hallucination in current OLMs. To address this, we propose Omni-Detective, an agentic data generation pipeline integrating tool-calling, to autonomously produce highly detailed yet minimally hallucinatory multimodal data. Based on the data generated with Omni-Detective, we train two captioning models: Audio-Captioner for audio-only detailed perception, and Omni-Captioner for audio-visual detailed perception. Under the cascade evaluation protocol, Audio-Captioner achieves the best performance on MMAU and MMAR among all open-source models, surpassing Gemini 2.5 Flash and delivering performance comparable to Gemini 2.5 Pro. On existing detailed captioning benchmarks, Omni-Captioner sets a new state-of-the-art on VDC and achieves the best trade-off between detail and hallucination on the video-SALMONN 2 testset. Given the absence of a dedicated benchmark for omni detailed perception, we design Omni-Cloze, a novel cloze-style evaluation for detailed audio, visual, and audio-visual captioning that ensures stable, efficient, and reliable assessment. Experimental results and analysis demonstrate the effectiveness of Omni-Detective in generating high-quality detailed captions, as well as the superiority of Omni-Cloze in evaluating such detailed captions.

Yihao Wu, Tianrui Wang, Yizhou Peng et al.

While biases in large language models (LLMs), such as stereotypes and cultural tendencies in outputs, have been examined and identified, their presence and characteristics in spoken dialogue models (SDMs) with audio input and output remain largely unexplored. Paralinguistic features, such as age, gender, and accent, can affect model outputs; when compounded by multi-turn conversations, these effects may exacerbate biases, with potential implications for fairness in decision-making and recommendation tasks. In this paper, we systematically evaluate biases in speech LLMs and study the impact of multi-turn dialogues with repeated negative feedback. Bias is measured using Group Unfairness Score (GUS) for decisions and similarity-based normalized statistics rate (SNSR) for recommendations, across both open-source models like Qwen2.5-Omni and GLM-4-Voice, as well as closed-source APIs such as GPT-4o Audio and Gemini-2.5-Flash. Our analysis reveals that closed-source models generally exhibit lower bias, while open-source models are more sensitive to age and gender, and recommendation tasks tend to amplify cross-group disparities. We found that biased decisions may persist in multi-turn conversations. This work provides the first systematic study of biases in end-to-end spoken dialogue models, offering insights towards fair and reliable audio-based interactive systems. To facilitate further research, we release the FairDialogue dataset and evaluation code.

Junyu Wang, Ziyang Ma, Zhengding Luo et al.

Large Audio-Language Models (LALMs) often suffer from audio-textual attention imbalance, prioritizing text over acoustic information, particularly in the multi-modal fusion layers of the Transformer architecture. This bias hinders their ability to fully utilize acoustic cues, causing suboptimal performance on audio reasoning tasks. To mitigate this, we propose \textbf{MATA}, a novel training-free method that dynamically pushes LALMs to pay \textbf{M}ore \textbf{A}ttention \textbf{T}o \textbf{A}udio tokens within the self-attention mechanism. Specifically, MATA intervenes post raw attention scoring, targeting only the last token in intermediate layers without introducing additional parameters or computational overhead. Experiments on the MMAU and MMAR benchmarks confirm MATA's effectiveness, with consistent performance gains. Notably, on MMAR, MATA enables an open-source model to surpass the proprietary Gemini 2.0 Flash for the first time. Our work provides an efficient solution to mitigate attention bias and opens a new research direction for enhancing the audio-processing capabilities of multi-modal models.

Yakun Song, Zhuo Chen, Xiaofei Wang et al.

Neural codec language model (LM) has demonstrated strong capability in zero-shot text-to-speech (TTS) synthesis. However, the codec LM often suffers from limitations in inference speed and stability, due to its auto-regressive nature and implicit alignment between text and audio. In this work, to handle these challenges, we introduce a new variant of neural codec LM, namely TacoLM. Specifically, TacoLM introduces a gated attention mechanism to improve the training and inference efficiency and reduce the model size. Meanwhile, an additional gated cross-attention layer is included for each decoder layer, which improves the efficiency and content accuracy of the synthesized speech. In the evaluation of the Librispeech corpus, the proposed TacoLM achieves a better word error rate, speaker similarity, and mean opinion score, with 90% fewer parameters and 5.2 times speed up, compared with VALL-E. Demo and code is available at https://ereboas.github.io/TacoLM/.

Ziyang Ma, Zhisheng Zheng, Jiaxin Ye et al.

We propose emotion2vec, a universal speech emotion representation model. emotion2vec is pre-trained on open-source unlabeled emotion data through self-supervised online distillation, combining utterance-level loss and frame-level loss during pre-training. emotion2vec outperforms state-of-the-art pre-trained universal models and emotion specialist models by only training linear layers for the speech emotion recognition task on the mainstream IEMOCAP dataset. In addition, emotion2vec shows consistent improvements among 10 different languages of speech emotion recognition datasets. emotion2vec also shows excellent results on other emotion tasks, such as song emotion recognition, emotion prediction in conversation, and sentiment analysis. Comparison experiments, ablation experiments, and visualization comprehensively demonstrate the universal capability of the proposed emotion2vec. To the best of our knowledge, emotion2vec is the first universal representation model in various emotion-related tasks, filling a gap in the field.

Ziyang Ma, Guanrou Yang, Yifan Yang et al.

In this paper, we focus on solving one of the most important tasks in the field of speech processing, i.e., automatic speech recognition (ASR), with speech foundation encoders and large language models (LLM). Recent works have complex designs such as compressing the output temporally for the speech encoder, tackling modal alignment for the projector, and utilizing parameter-efficient fine-tuning for the LLM. We found that delicate designs are not necessary, while an embarrassingly simple composition of off-the-shelf speech encoder, LLM, and the only trainable linear projector is competent for the ASR task. To be more specific, we benchmark and explore various combinations of LLMs and speech encoders, leading to the optimal LLM-based ASR system, which we call SLAM-ASR. The proposed SLAM-ASR provides a clean setup and little task-specific design, where only the linear projector is trained. To the best of our knowledge, SLAM-ASR achieves the best performance on the Librispeech benchmark among LLM-based ASR models and even outperforms the latest LLM-based audio-universal model trained on massive pair data. Finally, we explore the capability emergence of LLM-based ASR in the process of modal alignment. We hope that our study can facilitate the research on extending LLM with cross-modality capacity and shed light on the LLM-based ASR community.

Wenxi Chen, Yuzhe Liang, Ziyang Ma et al.

Audio self-supervised learning (SSL) pre-training, which aims to learn good representations from unlabeled audio, has made remarkable progress. However, the extensive computational demands during pre-training pose a significant barrier to the potential application and optimization of audio SSL models. In this paper, inspired by the success of data2vec 2.0 in image modality and Audio-MAE in audio modality, we introduce Efficient Audio Transformer (EAT) to further improve the effectiveness and efficiency in audio SSL. The proposed EAT adopts the bootstrap self-supervised training paradigm to the audio domain. A novel Utterance-Frame Objective (UFO) is designed to enhance the modeling capability of acoustic events. Furthermore, we reveal that the masking strategy is critical in audio SSL pre-training, and superior audio representations can be obtained with large inverse block masks. Experiment results demonstrate that EAT achieves state-of-the-art (SOTA) performance on a range of audio-related tasks, including AudioSet (AS-2M, AS-20K), ESC-50, and SPC-2, along with a significant pre-training speedup up to ~15x compared to existing audio SSL models.

Ziyang Ma, Zhuo Chen, Yuping Wang et al.

Large Audio-Language Models (LALMs) have demonstrated remarkable performance in tasks involving audio perception and understanding, such as speech recognition and audio captioning. However, their reasoning capabilities - critical for solving complex real-world problems - remain underexplored. In this work, we conduct the first exploration into integrating Chain-of-Thought (CoT) reasoning into LALMs to enhance their reasoning ability across auditory modalities. We evaluate representative CoT methods, analyzing their performance in both information extraction and reasoning tasks across sound, music, and speech domains. Our findings reveal that CoT methods significantly improve performance on easy and medium tasks but encounter challenges with hard tasks, where reasoning chains can confuse the model rather than improve accuracy. Additionally, we identify a positive correlation between reasoning path length and accuracy, demonstrating the potential of scaling inference for advanced instruction-following and reasoning. This study not only highlights the promise of CoT in enhancing LALM reasoning capabilities but also identifies key limitations and provides actionable directions for future research.

Ziyang Ma, Yinghao Ma, Yanqiao Zhu et al.

We introduce MMAR, a new benchmark designed to evaluate the deep reasoning capabilities of Audio-Language Models (ALMs) across massive multi-disciplinary tasks. MMAR comprises 1,000 meticulously curated audio-question-answer triplets, collected from real-world internet videos and refined through iterative error corrections and quality checks to ensure high quality. Unlike existing benchmarks that are limited to specific domains of sound, music, or speech, MMAR extends them to a broad spectrum of real-world audio scenarios, including mixed-modality combinations of sound, music, and speech. Each question in MMAR is hierarchically categorized across four reasoning layers: Signal, Perception, Semantic, and Cultural, with additional sub-categories within each layer to reflect task diversity and complexity. To further foster research in this area, we annotate every question with a Chain-of-Thought (CoT) rationale to promote future advancements in audio reasoning. Each item in the benchmark demands multi-step deep reasoning beyond surface-level understanding. Moreover, a part of the questions requires graduate-level perceptual and domain-specific knowledge, elevating the benchmark's difficulty and depth. We evaluate MMAR using a broad set of models, including Large Audio-Language Models (LALMs), Large Audio Reasoning Models (LARMs), Omni Language Models (OLMs), Large Language Models (LLMs), and Large Reasoning Models (LRMs), with audio caption inputs. The performance of these models on MMAR highlights the benchmark's challenging nature, and our analysis further reveals critical limitations of understanding and reasoning capabilities among current models. We hope MMAR will serve as a catalyst for future advances in this important but little-explored area.

Zhisheng Zheng, Puyuan Peng, Ziyang Ma et al.

Spatial sound reasoning is a fundamental human skill, enabling us to navigate and interpret our surroundings based on sound. In this paper we present BAT, which combines the spatial sound perception ability of a binaural acoustic scene analysis model with the natural language reasoning capabilities of a large language model (LLM) to replicate this innate ability. To address the lack of existing datasets of in-the-wild spatial sounds, we synthesized a binaural audio dataset using AudioSet and SoundSpaces 2.0. Next, we developed SpatialSoundQA, a spatial sound-based question-answering dataset, offering a range of QA tasks that train BAT in various aspects of spatial sound perception and reasoning. The acoustic front end encoder of BAT is a novel spatial audio encoder named Spatial Audio Spectrogram Transformer, or Spatial-AST, which by itself achieves strong performance across sound event detection, spatial localization, and distance estimation. By integrating Spatial-AST with LLaMA-2 7B model, BAT transcends standard Sound Event Localization and Detection (SELD) tasks, enabling the model to reason about the relationships between the sounds in its environment. Our experiments demonstrate BAT's superior performance on both spatial sound perception and reasoning, showcasing the immense potential of LLMs in navigating and interpreting complex spatial audio environments.