Heting Gao

Kaizhi Qian, Yang Zhang, Heting Gao et al. · mit

Self-supervised learning in speech involves training a speech representation network on a large-scale unannotated speech corpus, and then applying the learned representations to downstream tasks. Since the majority of the downstream tasks of SSL learning in speech largely focus on the content information in speech, the most desirable speech representations should be able to disentangle unwanted variations, such as speaker variations, from the content. However, disentangling speakers is very challenging, because removing the speaker information could easily result in a loss of content as well, and the damage of the latter usually far outweighs the benefit of the former. In this paper, we propose a new SSL method that can achieve speaker disentanglement without severe loss of content. Our approach is adapted from the HuBERT framework, and incorporates disentangling mechanisms to regularize both the teacher labels and the learned representations. We evaluate the benefit of speaker disentanglement on a set of content-related downstream tasks, and observe a consistent and notable performance advantage of our speaker-disentangled representations.

Eunseop Yoon, Hee Suk Yoon, Dhananjaya Gowda et al.

Text-to-Text Transfer Transformer (T5) has recently been considered for the Grapheme-to-Phoneme (G2P) transduction. As a follow-up, a tokenizer-free byte-level model based on T5 referred to as ByT5, recently gave promising results on word-level G2P conversion by representing each input character with its corresponding UTF-8 encoding. Although it is generally understood that sentence-level or paragraph-level G2P can improve usability in real-world applications as it is better suited to perform on heteronyms and linking sounds between words, we find that using ByT5 for these scenarios is nontrivial. Since ByT5 operates on the character level, it requires longer decoding steps, which deteriorates the performance due to the exposure bias commonly observed in auto-regressive generation models. This paper shows that the performance of sentence-level and paragraph-level G2P can be improved by mitigating such exposure bias using our proposed loss-based sampling method.

Chaoyou Fu, Haojia Lin, Xiong Wang et al.

Recent Multimodal Large Language Models (MLLMs) have typically focused on integrating visual and textual modalities, with less emphasis placed on the role of speech in enhancing interaction. However, speech plays a crucial role in multimodal dialogue systems, and implementing high-performance in both vision and speech tasks remains a significant challenge due to the fundamental modality differences. In this paper, we propose a carefully designed multi-stage training methodology that progressively trains LLM to understand both visual and speech information, ultimately enabling fluent vision and speech interaction. Our approach not only preserves strong vision-language capacity, but also enables efficient speech-to-speech dialogue capabilities without separate ASR and TTS modules, significantly accelerating multimodal end-to-end response speed. By comparing our method against state-of-the-art counterparts across benchmarks for image, video, and speech tasks, we demonstrate that our model is equipped with both strong visual and speech capabilities, making near real-time vision and speech interaction. Code has been released at https://github.com/VITA-MLLM/VITA.

Jiacheng Xu, Heting Gao, Liufei Xie et al.

Human speech conveys expressiveness beyond linguistic content, including personality, mood, or performance elements, such as a comforting tone or humming a song, which we formalize as role-playing and singing. We present VITA-QinYu, the first expressive end-to-end (E2E) spoken language model (SLM) that goes beyond natural conversation to support both role-playing and singing generation. VITA-QinYu adopts a hybrid speech-text paradigm that extends interleaved text-audio modeling with multi-codebook audio tokens, a design enabling richer paralinguistic representation while preserving a clear separation between modalities to avoid interference. We further develop a comprehensive data generation pipeline to synthesize a total of 15.8K hours of natural conversation, role-playing, and singing data for training. VITA-QinYu demonstrates superior expressiveness, outperforming peer SLMs by 7 percentage points on objective role-playing benchmarks, and surpassing peer models by 0.13 points on a 5-point MOS scale for singing. Simultaneously, it achieves state-of-the-art conversational accuracy and fluency, exceeding prior SLMs by 1.38 and 4.98 percentage points on the C3 and URO benchmarks, respectively. We open-source our code and models and provide an easy-to-use demo with full-stack support for streaming and full-duplex interaction.

Yi Li, Heting Gao, Mingde He et al.

In scoliosis surgery, the limited field of view of the C-arm X-ray machine restricts the surgeons' holistic analysis of spinal structures .This paper presents an end-to-end efficient and robust intraoperative X-ray image stitching method for scoliosis surgery,named SX-Stitch. The method is divided into two stages:segmentation and stitching. In the segmentation stage, We propose a medical image segmentation model named Vision Mamba of Spine-UNet (VMS-UNet), which utilizes the state space Mamba to capture long-distance contextual information while maintaining linear computational complexity, and incorporates the SimAM attention mechanism, significantly improving the segmentation performance.In the stitching stage, we simplify the alignment process between images to the minimization of a registration energy function. The total energy function is then optimized to order unordered images, and a hybrid energy function is introduced to optimize the best seam, effectively eliminating parallax artifacts. On the clinical dataset, Sx-Stitch demonstrates superiority over SOTA schemes both qualitatively and quantitatively.

Zuwei Long, Yunhang Shen, Chaoyou Fu et al.

With the growing requirement for natural human-computer interaction, speech-based systems receive increasing attention as speech is one of the most common forms of daily communication. However, the existing speech models still experience high latency when generating the first audio token during streaming, which poses a significant bottleneck for deployment. To address this issue, we propose VITA-Audio, an end-to-end large speech model with fast audio-text token generation. Specifically, we introduce a lightweight Multiple Cross-modal Token Prediction (MCTP) module that efficiently generates multiple audio tokens within a single model forward pass, which not only accelerates the inference but also significantly reduces the latency for generating the first audio in streaming scenarios. In addition, a four-stage progressive training strategy is explored to achieve model acceleration with minimal loss of speech quality. To our knowledge, VITA-Audio is the first multi-modal large language model capable of generating audio output during the first forward pass, enabling real-time conversational capabilities with minimal latency. VITA-Audio is fully reproducible and is trained on open-source data only. Experimental results demonstrate that our model achieves an inference speedup of 3~5x at the 7B parameter scale, but also significantly outperforms open-source models of similar model size on multiple benchmarks for automatic speech recognition (ASR), text-to-speech (TTS), and spoken question answering (SQA) tasks.

Hang Shao, Heting Gao, Yunhang Shen et al.

Native multimodal large language models (MLLMs) restructure a single large language model (LLM) into a spoken language model (SLM) capable of both speech and text generation. Compared to modular and aligned MLLMs, native MLLMs preserve richer paralinguistic features such as emotion and prosody, and generate speech responses directly within the backbone LLM rather than using a separate speech decoder. This integration also results in lower response latency and smoother interaction. However, native MLLMs suffer from catastrophic forgetting and performance degradation because the available paired speech-text data is insufficient to support the pretraining of MLLMs compared to the vast amount of text data required to pretrain text LLMs. To address this issue, we propose DeepTalk, a framework for adaptive modality expert learning based on a Mixture of Experts (MoE) architecture. DeepTalk first adaptively distinguishes modality experts according to their modality load within the LLM. Each modality expert then undergoes specialized single-modality training, followed by joint multimodal collaborative training. As a result, DeepTalk incurs only a 5.5% performance drop compared to the original LLM, which is significantly lower than the average performance drop of over 20% typically seen in native MLLMs (such as GLM-4-Voice), and is on par with modular MLLMs. Meanwhile, the end-to-end dialogue latency remains within 0.5 seconds, ensuring a seamless and intelligent speech interaction experience. Code and models are released at https://github.com/talkking/DeepTalk.

Heting Gao, Junrui Ni, Kaizhi Qian et al.

Large-scale auto-regressive language models pretrained on massive text have demonstrated their impressive ability to perform new natural language tasks with only a few text examples, without the need for fine-tuning. Recent studies further show that such a few-shot learning ability can be extended to the text-image setting by training an encoder to encode the images into embeddings functioning like the text embeddings of the language model. Interested in exploring the possibility of transferring the few-shot learning ability to the audio-text setting, we propose a novel speech understanding framework, WavPrompt, where we finetune a wav2vec model to generate a sequence of audio embeddings understood by the language model. We show that WavPrompt is a few-shot learner that can perform speech understanding tasks better than a naive text baseline. We conduct detailed ablation studies on different components and hyperparameters to empirically identify the best model configuration. In addition, we conduct a non-speech understanding experiment to show WavPrompt can extract more information than just the transcriptions. Code is available at https://github.com/Hertin/WavPrompt

Junrui Ni, Liming Wang, Heting Gao et al.

An unsupervised text-to-speech synthesis (TTS) system learns to generate speech waveforms corresponding to any written sentence in a language by observing: 1) a collection of untranscribed speech waveforms in that language; 2) a collection of texts written in that language without access to any transcribed speech. Developing such a system can significantly improve the availability of speech technology to languages without a large amount of parallel speech and text data. This paper proposes an unsupervised TTS system based on an alignment module that outputs pseudo-text and another synthesis module that uses pseudo-text for training and real text for inference. Our unsupervised system can achieve comparable performance to the supervised system in seven languages with about 10-20 hours of speech each. A careful study on the effect of text units and vocoders has also been conducted to better understand what factors may affect unsupervised TTS performance. The samples generated by our models can be found at https://cactuswiththoughts.github.io/UnsupTTS-Demo, and our code can be found at https://github.com/lwang114/UnsupTTS.

Heting Gao, Hang Shao, Xiong Wang et al.

The film Her features Samantha, a sophisticated AI audio agent who is capable of understanding both linguistic and paralinguistic information in human speech and delivering real-time responses that are natural, informative and sensitive to emotional subtleties. Moving one step toward more sophisticated audio agent from recent advancement in end-to-end (E2E) speech systems, we propose LUCY, a E2E speech model that (1) senses and responds to user's emotion, (2) deliver responses in a succinct and natural style, and (3) use external tool to answer real-time inquiries. Experiment results show that LUCY is better at emotion control than peer models, generating emotional responses based on linguistic emotional instructions and responding to paralinguistic emotional cues. Lucy is also able to generate responses in a more natural style, as judged by external language models, without sacrificing much performance on general question answering. Finally, LUCY can leverage function calls to answer questions that are out of its knowledge scope.

Lijiang Li, Zuwei Long, Yunhang Shen et al.

While recent multimodal large language models (MLLMs) have made impressive strides, they predominantly employ a conventional autoregressive architecture as their backbone, leaving significant room to explore effective and efficient alternatives in architectural design. Concurrently, recent studies have successfully applied discrete diffusion models to various domains, such as visual understanding and image generation, revealing their considerable potential as a promising backbone for multimodal systems. Drawing inspiration from these pioneering research, we introduce Omni-Diffusion, the first any-to-any multimodal language model built entirely on mask-based discrete diffusion models, which unifies understanding and generation across text, speech, and images. Omni-Diffusion employs a unified mask-based discrete diffusion model to directly capture the joint distribution over discrete multimodal tokens. This approach supports not only bimodal tasks but also more complex scenarios involving multiple modalities. On a diverse set of benchmarks, our method outperforms or performs on par with existing multimodal systems that process two or more modalities, highlighting the significant promise of diffusion models in powering the next generation of multimodal foundation models. Project webpage: https://omni-diffusion.github.io.

Xulin Fan, Heting Gao, Ziyi Chen et al.

Talking head synthesis, also known as speech-to-lip synthesis, reconstructs the facial motions that align with the given audio tracks. The synthesized videos are evaluated on mainly two aspects, lip-speech synchronization and image fidelity. Recent studies demonstrate that GAN-based and diffusion-based models achieve state-of-the-art (SOTA) performance on this task, with diffusion-based models achieving superior image fidelity but experiencing lower synchronization compared to their GAN-based counterparts. To this end, we propose SyncDiff, a simple yet effective approach to improve diffusion-based models using a temporal pose frame with information bottleneck and facial-informative audio features extracted from AVHuBERT, as conditioning input into the diffusion process. We evaluate SyncDiff on two canonical talking head datasets, LRS2 and LRS3 for direct comparison with other SOTA models. Experiments on LRS2/LRS3 datasets show that SyncDiff achieves a synchronization score 27.7%/62.3% relatively higher than previous diffusion-based methods, while preserving their high-fidelity characteristics.

Fengpu Pan, Heting Gao, Jiangtao Wen et al.

Snapshot Compressed Imaging (SCI) offers high-speed, low-bandwidth, and energy-efficient image acquisition, but remains challenged by low-light and low signal-to-noise ratio (SNR) conditions. Moreover, practical hardware constraints in high-resolution sensors limit the use of large frame-sized masks, necessitating smaller, hardware-friendly designs. In this work, we present a novel SCI-based computer vision framework using pseudo-random binary masks of only 8$\times$8 in size for physically feasible implementations. At its core is CompDAE, a Compressive Denoising Autoencoder built on the STFormer architecture, designed to perform downstream tasks--such as edge detection and depth estimation--directly from noisy compressive raw pixel measurements without image reconstruction. CompDAE incorporates a rate-constrained training strategy inspired by BackSlash to promote compact, compressible models. A shared encoder paired with lightweight task-specific decoders enables a unified multi-task platform. Extensive experiments across multiple datasets demonstrate that CompDAE achieves state-of-the-art performance with significantly lower complexity, especially under ultra-low-light conditions where traditional CMOS and SCI pipelines fail.

Junrui Ni, Liming Wang, Yang Zhang et al.

Recent advancements in supervised automatic speech recognition (ASR) have achieved remarkable performance, largely due to the growing availability of large transcribed speech corpora. However, most languages lack sufficient paired speech and text data to effectively train these systems. In this article, we tackle the challenge of developing ASR systems without paired speech and text corpora by proposing the removal of reliance on a phoneme lexicon. We explore a new research direction: word-level unsupervised ASR, and experimentally demonstrate that an unsupervised speech recognizer can emerge from joint speech-to-speech and text-to-text masked token-infilling. Using a curated speech corpus containing a fixed number of English words, our system iteratively refines the word segmentation structure and achieves a word error rate of between 20-23%, depending on the vocabulary size, without parallel transcripts, oracle word boundaries, or a pronunciation lexicon. This innovative model surpasses the performance of previous unsupervised ASR models under the lexicon-free setting.

Yunan Zhang, Xiang Cheng, Heting Gao et al.

Knowledge-graph-based reasoning has drawn a lot of attention due to its interpretability. However, previous methods suffer from the incompleteness of the knowledge graph, namely the interested link or entity that can be missing in the knowledge graph(explicit missing). Also, most previous models assume the distance between the target and source entity is short, which is not true on a real-world KG like Freebase(implicit missing). The sensitivity to the incompleteness of KG and the incapability to capture the long-distance link between entities have limited the performance of these models on large KG. In this paper, we propose a model that leverages the text corpus to cure such limitations, either the explicit or implicit missing links. We model the question answering on KG as a cooperative task between two agents, a knowledge graph reasoning agent and an information extraction agent. Each agent learns its skill to complete its own task, hopping on KG or select knowledge from the corpus, via maximizing the reward for correctly answering the question. The reasoning agent decides how to find an equivalent path for the given entity and relation. The extraction agent provide shortcut for long-distance target entity or provide missing relations for explicit missing links with messages from the reasoning agent. Through such cooperative reward design, our model can augment the incomplete KG strategically while not introduce much unnecessary noise that could enlarge the search space and lower the performance.

Yunan Zhang, Heting Gao, Tarek Abdelzaher

As the ongoing rapid urbanization takes place with an ever-increasing speed, fully modeling urban dynamics becomes more and more challenging, but also a necessity for socioeconomic development. It is challenging because human activities and constructions are ubiquitous; urban landscape and life content change anywhere and anytime. It's crucial due to the fact that only up-to-date urban dynamics can enable governors to optimize their city planning strategy and help individuals organize their daily lives in a more efficient way. Previous geographic topic model based methods attempt to solve this problem but suffer from high computational cost and memory consumption, limiting their scalability to city level applications. Also, strong prior assumptions make such models fail to capture certain patterns by nature. To bridge the gap, we propose Macross, a metapath guided embedding approach to jointly model location, time and text information. Given a dataset of geo-tagged social media posts, we extract and aggregate location and time and construct a heterogeneous information network using the aggregated space and time. Metapath2vec based approach is used to construct vector representations for times, locations and frequent words such that co-occurrence pairs of nodes are closer in latent space. The vector representations will be used to infer related time, locations or keywords for a user query. Experiments done on enormous datasets show our model can generate comparable if not better quality query results compared to state of the art models and outperform some cutting-edge models for activity recovery and classification.

Suhansanu Kumar, Heting Gao, Changyu Wang et al.

Researchers often query online social platforms through their application programming interfaces (API) to find target populations such as people with mental illness~\cite{De-Choudhury2017} and jazz musicians~\cite{heckathorn2001finding}. Entities of such target population satisfy a property that is typically identified using an oracle (human or a pre-trained classifier). When the property of the target entities is not directly queryable via the API, we refer to the property as `hidden' and the population as a hidden population. Finding individuals who belong to these populations on social networks is hard because they are non-queryable, and the sampler has to explore from a combinatorial query space within a finite budget limit. By exploiting the correlation between queryable attributes and the population of interest and by hierarchically ordering the query space, we propose a Decision tree-based Thompson sampler (\texttt{DT-TMP}) that efficiently discovers the right combination of attributes to query. Our proposed sampler outperforms the state-of-the-art samplers in online experiments, for example by 54\% on Twitter. When the number of matching entities to a query is known in offline experiments, \texttt{DT-TMP} performs exceedingly well by a factor of 0.9-1.5$\times$ over the baseline samplers. In the future, we wish to explore the option of finding hidden populations by formulating more complex queries.