Yi He

Yizhou Peng, Yufei Liu, Jicheng Zhang et al.

Internal Language Model Estimation (ILME) based language model (LM) fusion has been shown significantly improved recognition results over conventional shallow fusion in both intra-domain and cross-domain speech recognition tasks. In this paper, we attempt to apply our ILME method to cross-domain code-switching speech recognition (CSSR) work. Specifically, our curiosity comes from several aspects. First, we are curious about how effective the ILME-based LM fusion is for both intra-domain and cross-domain CSSR tasks. We verify this with or without merging two code-switching domains. More importantly, we train an end-to-end (E2E) speech recognition model by means of merging two monolingual data sets and observe the efficacy of the proposed ILME-based LM fusion for CSSR. Experimental results on SEAME that is from Southeast Asian and another Chinese Mainland CS data set demonstrate the effectiveness of the proposed ILME-based LM fusion method.

Heng Lian, John Scovil Atwood, Bojian Hou et al.

This paper investigates a new online learning problem with doubly-streaming data, where the data streams are described by feature spaces that constantly evolve, with new features emerging and old features fading away. The challenges of this problem are two folds: 1) Data samples ceaselessly flowing in may carry shifted patterns over time, requiring learners to update hence adapt on-the-fly. 2) Newly emerging features are described by very few samples, resulting in weak learners that tend to make error predictions. A plausible idea to overcome the challenges is to establish relationship between the pre-and-post evolving feature spaces, so that an online learner can leverage the knowledge learned from the old features to better the learning performance on the new features. Unfortunately, this idea does not scale up to high-dimensional media streams with complex feature interplay, which suffers an tradeoff between onlineness (biasing shallow learners) and expressiveness(requiring deep learners). Motivated by this, we propose a novel OLD^3S paradigm, where a shared latent subspace is discovered to summarize information from the old and new feature spaces, building intermediate feature mapping relationship. A key trait of OLD^3S is to treat the model capacity as a learnable semantics, yields optimal model depth and parameters jointly, in accordance with the complexity and non-linearity of the input data streams in an online fashion. Both theoretical analyses and empirical studies substantiate the viability and effectiveness of our proposal.

Feilong Chen, Di Wu, Jie Yang et al.

Online streaming feature selection (OSFS), which conducts feature selection in an online manner, plays an important role in dealing with high-dimensional data. In many real applications such as intelligent healthcare platform, streaming feature always has some missing data, which raises a crucial challenge in conducting OSFS, i.e., how to establish the uncertain relationship between sparse streaming features and labels. Unfortunately, existing OSFS algorithms never consider such uncertain relationship. To fill this gap, we in this paper propose an online sparse streaming feature selection with uncertainty (OS2FSU) algorithm. OS2FSU consists of two main parts: 1) latent factor analysis is utilized to pre-estimate the missing data in sparse streaming features before con-ducting feature selection, and 2) fuzzy logic and neighborhood rough set are employed to alleviate the uncertainty between estimated streaming features and labels during conducting feature selection. In the experiments, OS2FSU is compared with five state-of-the-art OSFS algorithms on six real datasets. The results demonstrate that OS2FSU outperforms its competitors when missing data are encountered in OSFS.

Di Wu, Peng Zhang, Yi He et al.

High-dimensional and sparse (HiDS) matrices are omnipresent in a variety of big data-related applications. Latent factor analysis (LFA) is a typical representation learning method that extracts useful yet latent knowledge from HiDS matrices via low-rank approximation. Current LFA-based models mainly focus on a single-metric representation, where the representation strategy designed for the approximation Loss function, is fixed and exclusive. However, real-world HiDS matrices are commonly heterogeneous and inclusive and have diverse underlying patterns, such that a single-metric representation is most likely to yield inferior performance. Motivated by this, we in this paper propose a multi-metric latent factor (MMLF) model. Its main idea is two-fold: 1) two vector spaces and three Lp-norms are simultaneously employed to develop six variants of LFA model, each of which resides in a unique metric representation space, and 2) all the variants are ensembled with a tailored, self-adaptive weighting strategy. As such, our proposed MMLF enjoys the merits originated from a set of disparate metric spaces all at once, achieving the comprehensive and unbiased representation of HiDS matrices. Theoretical study guarantees that MMLF attains a performance gain. Extensive experiments on eight real-world HiDS datasets, spanning a wide range of industrial and science domains, verify that our MMLF significantly outperforms ten state-of-the-art, shallow and deep counterparts.

Cheng Liang, Teng Huang, Yi He et al.

User behavior data produced during interaction with massive items in the significant data era are generally heterogeneous and sparse, leaving the recommender system (RS) a large diversity of underlying patterns to excavate. Deep neural network-based models have reached the state-of-the-art benchmark of the RS owing to their fitting capabilities. However, prior works mainly focus on designing an intricate architecture with fixed loss function and regulation. These single-metric models provide limited performance when facing heterogeneous and sparse user behavior data. Motivated by this finding, we propose a multi-metric AutoRec (MMA) based on the representative AutoRec. The idea of the proposed MMA is mainly two-fold: 1) apply different $L_p$-norm on loss function and regularization to form different variant models in different metric spaces, and 2) aggregate these variant models. Thus, the proposed MMA enjoys the multi-metric orientation from a set of dispersed metric spaces, achieving a comprehensive representation of user data. Theoretical studies proved that the proposed MMA could attain performance improvement. The extensive experiment on five real-world datasets proves that MMA can outperform seven other state-of-the-art models in predicting unobserved user behavior data.

Di Wu, Yi He, Xin Luo

A High-dimensional and sparse (HiDS) matrix is frequently encountered in a big data-related application like an e-commerce system or a social network services system. To perform highly accurate representation learning on it is of great significance owing to the great desire of extracting latent knowledge and patterns from it. Latent factor analysis (LFA), which represents an HiDS matrix by learning the low-rank embeddings based on its observed entries only, is one of the most effective and efficient approaches to this issue. However, most existing LFA-based models perform such embeddings on a HiDS matrix directly without exploiting its hidden graph structures, thereby resulting in accuracy loss. To address this issue, this paper proposes a graph-incorporated latent factor analysis (GLFA) model. It adopts two-fold ideas: 1) a graph is constructed for identifying the hidden high-order interaction (HOI) among nodes described by an HiDS matrix, and 2) a recurrent LFA structure is carefully designed with the incorporation of HOI, thereby improving the representa-tion learning ability of a resultant model. Experimental results on three real-world datasets demonstrate that GLFA outperforms six state-of-the-art models in predicting the missing data of an HiDS matrix, which evidently supports its strong representation learning ability to HiDS data.

Xianzhao Chen, Yist Y. Lin, Kang Wang et al.

End-to-end (E2E) systems have shown comparable performance to hybrid systems for automatic speech recognition (ASR). Word timings, as a by-product of ASR, are essential in many applications, especially for subtitling and computer-aided pronunciation training. In this paper, we improve the frame-level classifier for word timings in E2E system by introducing label priors in connectionist temporal classification (CTC) loss, which is adopted from prior works, and combining low-level Mel-scale filter banks with high-level ASR encoder output as input feature. On the internal Chinese corpus, the proposed method achieves 95.68%/94.18% compared to the hybrid system 93.0%/90.22% on the word timing accuracy metrics. It also surpass a previous E2E approach with an absolute increase of 4.80%/8.02% on the metrics on 7 languages. In addition, we further improve word timing accuracy by delaying CTC peaks with frame-wise knowledge distillation, though only experimenting on LibriSpeech.

Yist Y. Lin, Tao Han, Haihua Xu et al.

One of limitations in end-to-end automatic speech recognition (ASR) framework is its performance would be compromised if train-test utterance lengths are mismatched. In this paper, we propose an on-the-fly random utterance concatenation (RUC) based data augmentation method to alleviate train-test utterance length mismatch issue for short-video ASR task. Specifically, we are motivated by observations that our human-transcribed training utterances tend to be much shorter for short-video spontaneous speech (~3 seconds on average), while our test utterance generated from voice activity detection front-end is much longer (~10 seconds on average). Such a mismatch can lead to suboptimal performance. Empirically, it's observed the proposed RUC method significantly improves long utterance recognition without performance drop on short one. Overall, it achieves 5.72% word error rate reduction on average for 15 languages and improved robustness to various utterance length.

Leizhen Zhang, Lusi Li, Di Wu et al.

Its crux lies in the optimization of a tradeoff between accuracy and fairness of resultant models on the selected feature subset. The technical challenge of our setting is twofold: 1) streaming feature inputs, such that an informative feature may become obsolete or redundant for prediction if its information has been covered by other similar features that arrived prior to it, and 2) non-associational feature correlation, such that bias may be leaked from those seemingly admissible, non-protected features. To overcome this, we propose Streaming Feature Selection with Causal Fairness (SFCF) that builds two causal graphs egocentric to prediction label and protected feature, respectively, striving to model the complex correlation structure among streaming features, labels, and protected information. As such, bias can be eradicated from predictive modeling by removing those features being causally correlated with the protected feature yet independent to the labels. We theorize that the originally redundant features for prediction can later become admissible, when the learning accuracy is compromised by the large number of removed features (non-protected but can be used to reconstruct bias information). We benchmark SFCF\ on five datasets widely used in streaming feature research, and the results substantiate its performance superiority over six rival models in terms of efficiency and sparsity of feature selection and equalized odds of the resultant predictive models.

Xiaoyuan Cheng, Wenxuan Yuan, Yiming Yang et al.

The Koopman operator provides a powerful framework for modeling dynamical systems and has attracted growing interest from the machine learning community. However, its infinite-dimensional nature makes identifying suitable finite-dimensional subspaces challenging, especially for deep architectures. We argue that these difficulties come from suboptimal representation learning, where latent variables fail to balance expressivity and simplicity. This tension is closely related to the information bottleneck (IB) dilemma: constructing compressed representations that are both compact and predictive. Rethinking Koopman learning through this lens, we demonstrate that latent mutual information promotes simplicity, yet an overemphasis on simplicity may cause latent space to collapse onto a few dominant modes. In contrast, expressiveness is sustained by the von Neumann entropy, which prevents such collapse and encourages mode diversity. This insight leads us to propose an information-theoretic Lagrangian formulation that explicitly balances this tradeoff. Furthermore, we propose a new algorithm based on the Lagrangian formulation that encourages both simplicity and expressiveness, leading to a stable and interpretable Koopman representation. Beyond quantitative evaluations, we further visualize the learned manifolds under our representations, observing empirical results consistent with our theoretical predictions. Finally, we validate our approach across a diverse range of dynamical systems, demonstrating improved performance over existing Koopman learning methods. The implementation is publicly available at https://github.com/Wenxuan52/InformationKoopman.

Siyi Zhou, Yiquan Zhou, Yi He et al.

Existing autoregressive large-scale text-to-speech (TTS) models have advantages in speech naturalness, but their token-by-token generation mechanism makes it difficult to precisely control the duration of synthesized speech. This becomes a significant limitation in applications requiring strict audio-visual synchronization, such as video dubbing. This paper introduces IndexTTS2, which proposes a novel, general, and autoregressive model-friendly method for speech duration control. The method supports two generation modes: one explicitly specifies the number of generated tokens to precisely control speech duration; the other freely generates speech in an autoregressive manner without specifying the number of tokens, while faithfully reproducing the prosodic features of the input prompt. Furthermore, IndexTTS2 achieves disentanglement between emotional expression and speaker identity, enabling independent control over timbre and emotion. In the zero-shot setting, the model can accurately reconstruct the target timbre (from the timbre prompt) while perfectly reproducing the specified emotional tone (from the style prompt). To enhance speech clarity in highly emotional expressions, we incorporate GPT latent representations and design a novel three-stage training paradigm to improve the stability of the generated speech. Additionally, to lower the barrier for emotional control, we designed a soft instruction mechanism based on text descriptions by fine-tuning Qwen3, effectively guiding the generation of speech with the desired emotional orientation. Finally, experimental results on multiple datasets show that IndexTTS2 outperforms state-of-the-art zero-shot TTS models in terms of word error rate, speaker similarity, and emotional fidelity. Audio samples are available at: https://index-tts.github.io/index-tts2.github.io/

Jiajun Shen, Yufei Jin, Yi He et al.

This paper presents HGEN that pioneers ensemble learning for heterogeneous graphs. We argue that the heterogeneity in node types, nodal features, and local neighborhood topology poses significant challenges for ensemble learning, particularly in accommodating diverse graph learners. Our HGEN framework ensembles multiple learners through a meta-path and transformation-based optimization pipeline to uplift classification accuracy. Specifically, HGEN uses meta-path combined with random dropping to create Allele Graph Neural Networks (GNNs), whereby the base graph learners are trained and aligned for later ensembling. To ensure effective ensemble learning, HGEN presents two key components: 1) a residual-attention mechanism to calibrate allele GNNs of different meta-paths, thereby enforcing node embeddings to focus on more informative graphs to improve base learner accuracy, and 2) a correlation-regularization term to enlarge the disparity among embedding matrices generated from different meta-paths, thereby enriching base learner diversity. We analyze the convergence of HGEN and attest its higher regularization magnitude over simple voting. Experiments on five heterogeneous networks validate that HGEN consistently outperforms its state-of-the-art competitors by substantial margin.

Yufei Liu, Rao Ma, Haihua Xu et al.

An end-to-end (E2E) ASR model implicitly learns a prior Internal Language Model (ILM) from the training transcripts. To fuse an external LM using Bayes posterior theory, the log likelihood produced by the ILM has to be accurately estimated and subtracted. In this paper we propose two novel approaches to estimate the ILM based on Listen-Attend-Spell (LAS) framework. The first method is to replace the context vector of the LAS decoder at every time step with a vector that is learned with training transcripts. Furthermore, we propose another method that uses a lightweight feed-forward network to directly map query vector to context vector in a dynamic sense. Since the context vectors are learned by minimizing the perplexities on training transcripts, and their estimation is independent of encoder output, hence the ILMs are accurately learned for both methods. Experiments show that the ILMs achieve the lowest perplexity, indicating the efficacy of the proposed methods. In addition, they also significantly outperform the shallow fusion method, as well as two previously proposed ILM Estimation (ILME) approaches on several datasets.

Yi He, Jiayuan Shi, Chuan Wang et al.

In this paper we present a new data-driven method for robust skin detection from a single human portrait image. Unlike previous methods, we incorporate human body as a weak semantic guidance into this task, considering acquiring large-scale of human labeled skin data is commonly expensive and time-consuming. To be specific, we propose a dual-task neural network for joint detection of skin and body via a semi-supervised learning strategy. The dual-task network contains a shared encoder but two decoders for skin and body separately. For each decoder, its output also serves as a guidance for its counterpart, making both decoders mutually guided. Extensive experiments were conducted to demonstrate the effectiveness of our network with mutual guidance, and experimental results show our network outperforms the state-of-the-art in skin detection.

Xu Tian, Jun Zhang, Zejun Ma et al.

Frame stacking is broadly applied in end-to-end neural network training like connectionist temporal classification (CTC), and it leads to more accurate models and faster decoding. However, it is not well-suited to conventional neural network based on context-dependent state acoustic model, if the decoder is unchanged. In this paper, we propose a novel frame retaining method which is applied in decoding. The system which combined frame retaining with frame stacking could reduces the time consumption of both training and decoding. Long short-term memory (LSTM) recurrent neural networks (RNNs) using it achieve almost linear training speedup and reduces relative 41\% real time factor (RTF). At the same time, recognition performance is no degradation or improves sightly on Shenma voice search dataset in Mandarin.

Xu Tian, Jun Zhang, Zejun Ma et al.

Recurrent neural networks (RNNs), especially long short-term memory (LSTM) RNNs, are effective network for sequential task like speech recognition. Deeper LSTM models perform well on large vocabulary continuous speech recognition, because of their impressive learning ability. However, it is more difficult to train a deeper network. We introduce a training framework with layer-wise training and exponential moving average methods for deeper LSTM models. It is a competitive framework that LSTM models of more than 7 layers are successfully trained on Shenma voice search data in Mandarin and they outperform the deep LSTM models trained by conventional approach. Moreover, in order for online streaming speech recognition applications, the shallow model with low real time factor is distilled from the very deep model. The recognition accuracy have little loss in the distillation process. Therefore, the model trained with the proposed training framework reduces relative 14\% character error rate, compared to original model which has the similar real-time capability. Furthermore, the novel transfer learning strategy with segmental Minimum Bayes-Risk is also introduced in the framework. The strategy makes it possible that training with only a small part of dataset could outperform full dataset training from the beginning.

Xu Tian, Jun Zhang, Zejun Ma et al.

As training data rapid growth, large-scale parallel training with multi-GPUs cluster is widely applied in the neural network model learning currently.We present a new approach that applies exponential moving average method in large-scale parallel training of neural network model. It is a non-interference strategy that the exponential moving average model is not broadcasted to distributed workers to update their local models after model synchronization in the training process, and it is implemented as the final model of the training system. Fully-connected feed-forward neural networks (DNNs) and deep unidirectional Long short-term memory (LSTM) recurrent neural networks (RNNs) are successfully trained with proposed method for large vocabulary continuous speech recognition on Shenma voice search data in Mandarin. The character error rate (CER) of Mandarin speech recognition further degrades than state-of-the-art approaches of parallel training.