Rui Wang

Yichong Leng, Xu Tan, Wenjie Liu et al. · microsoft-research

Error correction in automatic speech recognition (ASR) aims to correct those incorrect words in sentences generated by ASR models. Since recent ASR models usually have low word error rate (WER), to avoid affecting originally correct tokens, error correction models should only modify incorrect words, and therefore detecting incorrect words is important for error correction. Previous works on error correction either implicitly detect error words through target-source attention or CTC (connectionist temporal classification) loss, or explicitly locate specific deletion/substitution/insertion errors. However, implicit error detection does not provide clear signal about which tokens are incorrect and explicit error detection suffers from low detection accuracy. In this paper, we propose SoftCorrect with a soft error detection mechanism to avoid the limitations of both explicit and implicit error detection. Specifically, we first detect whether a token is correct or not through a probability produced by a dedicatedly designed language model, and then design a constrained CTC loss that only duplicates the detected incorrect tokens to let the decoder focus on the correction of error tokens. Compared with implicit error detection with CTC loss, SoftCorrect provides explicit signal about which words are incorrect and thus does not need to duplicate every token but only incorrect tokens; compared with explicit error detection, SoftCorrect does not detect specific deletion/substitution/insertion errors but just leaves it to CTC loss. Experiments on AISHELL-1 and Aidatatang datasets show that SoftCorrect achieves 26.1% and 9.4% CER reduction respectively, outperforming previous works by a large margin, while still enjoying fast speed of parallel generation.

Shanfeng Huang, Zezhong Zhang, Shuai Wang et al.

Federated edge learning (FEEL) is envisioned as a promising paradigm to achieve privacy-preserving distributed learning. However, it consumes excessive learning time due to the existence of straggler devices. In this paper, a novel topology-optimized federated edge learning (TOFEL) scheme is proposed to tackle the heterogeneity issue in federated learning and to improve the communication-and-computation efficiency. Specifically, a problem of jointly optimizing the aggregation topology and computing speed is formulated to minimize the weighted summation of energy consumption and latency. To solve the mixed-integer nonlinear problem, we propose a novel solution method of penalty-based successive convex approximation, which converges to a stationary point of the primal problem under mild conditions. To facilitate real-time decision making, an imitation-learning based method is developed, where deep neural networks (DNNs) are trained offline to mimic the penalty-based method, and the trained imitation DNNs are deployed at the edge devices for online inference. Thereby, an efficient imitate-learning based approach is seamlessly integrated into the TOFEL framework. Simulation results demonstrate that the proposed TOFEL scheme accelerates the federated learning process, and achieves a higher energy efficiency. Moreover, we apply the scheme to 3D object detection with multi-vehicle point cloud datasets in the CARLA simulator. The results confirm the superior learning performance of the TOFEL scheme over conventional designs with the same resource and deadline constraints.

Junyi Ao, Rui Wang, Long Zhou et al.

Motivated by the success of T5 (Text-To-Text Transfer Transformer) in pre-trained natural language processing models, we propose a unified-modal SpeechT5 framework that explores the encoder-decoder pre-training for self-supervised speech/text representation learning. The SpeechT5 framework consists of a shared encoder-decoder network and six modal-specific (speech/text) pre/post-nets. After preprocessing the input speech/text through the pre-nets, the shared encoder-decoder network models the sequence-to-sequence transformation, and then the post-nets generate the output in the speech/text modality based on the output of the decoder. Leveraging large-scale unlabeled speech and text data, we pre-train SpeechT5 to learn a unified-modal representation, hoping to improve the modeling capability for both speech and text. To align the textual and speech information into this unified semantic space, we propose a cross-modal vector quantization approach that randomly mixes up speech/text states with latent units as the interface between encoder and decoder. Extensive evaluations show the superiority of the proposed SpeechT5 framework on a wide variety of spoken language processing tasks, including automatic speech recognition, speech synthesis, speech translation, voice conversion, speech enhancement, and speaker identification. We release our code and model at https://github.com/microsoft/SpeechT5.

Rui Wang, Junyi Ao, Long Zhou et al.

Initially developed for natural language processing (NLP), Transformer model is now widely used for speech processing tasks such as speaker recognition, due to its powerful sequence modeling capabilities. However, conventional self-attention mechanisms are originally designed for modeling textual sequence without considering the characteristics of speech and speaker modeling. Besides, different Transformer variants for speaker recognition have not been well studied. In this work, we propose a novel multi-view self-attention mechanism and present an empirical study of different Transformer variants with or without the proposed attention mechanism for speaker recognition. Specifically, to balance the capabilities of capturing global dependencies and modeling the locality, we propose a multi-view self-attention mechanism for speaker Transformer, in which different attention heads can attend to different ranges of the receptive field. Furthermore, we introduce and compare five Transformer variants with different network architectures, embedding locations, and pooling methods to learn speaker embeddings. Experimental results on the VoxCeleb1 and VoxCeleb2 datasets show that the proposed multi-view self-attention mechanism achieves improvement in the performance of speaker recognition, and the proposed speaker Transformer network attains excellent results compared with state-of-the-art models.

Shuai Wang, Dachuan Li, Rui Wang et al.

Wireless federated learning (FL) is an emerging machine learning paradigm that trains a global parametric model from distributed datasets via wireless communications. This paper proposes a unit-modulus wireless FL (UMWFL) framework, which simultaneously uploads local model parameters and computes global model parameters via optimized phase shifting. The proposed framework avoids sophisticated baseband signal processing, leading to both low communication delays and implementation costs. A training loss bound is derived and a penalty alternating minimization (PAM) algorithm is proposed to minimize the nonconvex nonsmooth loss bound. Experimental results in the Car Learning to Act (CARLA) platform show that the proposed UMWFL framework with PAM algorithm achieves smaller training losses and testing errors than those of the benchmark scheme.

Guoliang Li, Shuai Wang, Jie Li et al.

Characterizing the sensing and communication performance tradeoff in integrated sensing and communication (ISAC) systems is challenging in the applications of learning-based human motion recognition. This is because of the large experimental datasets and the black-box nature of deep neural networks. This paper presents SDP3, a Simulation-Driven Performance Predictor and oPtimizer, which consists of SDP3 data simulator, SDP3 performance predictor and SDP3 performance optimizer. Specifically, the SDP3 data simulator generates vivid wireless sensing datasets in a virtual environment, the SDP3 performance predictor predicts the sensing performance based on the function regression method, and the SDP3 performance optimizer investigates the sensing and communication performance tradeoff analytically. It is shown that the simulated sensing dataset matches the experimental dataset very well in the motion recognition accuracy. By leveraging SDP3, it is found that the achievable region of recognition accuracy and communication throughput consists of a communication saturation zone, a sensing saturation zone, and a communication-sensing adversarial zone, of which the desired balanced performance for ISAC systems lies in the third one.

Tong Zhang, Shuai Wang, Guoliang Li et al.

Realizing edge intelligence consists of sensing, communication, training, and inference stages. Conventionally, the sensing and communication stages are executed sequentially, which results in excessive amount of dataset generation and uploading time. This paper proposes to accelerate edge intelligence via integrated sensing and communication (ISAC). As such, the sensing and communication stages are merged so as to make the best use of the wireless signals for the dual purpose of dataset generation and uploading. However, ISAC also introduces additional interference between sensing and communication functionalities. To address this challenge, this paper proposes a classification error minimization formulation to design the ISAC beamforming and time allocation. The globally optimal solution is derived via the rank-1 guaranteed semidefinite relaxation, and performance analysis is performed to quantify the ISAC gain over that of conventional edge intelligence. Simulation results are provided to verify the effectiveness of the proposed ISAC-assisted edge intelligence system. Interestingly, we find that ISAC is always beneficial, when the duration of generating a sample is more than the duration of uploading a sample. Otherwise, the ISAC gain can vanish or even be negative. Nevertheless, we still derive a sufficient condition, under which a positive ISAC gain is feasible.

Guoliang Li, Shuai Wang, Jie Li et al.

Human motion recognition (HMR) based on wireless sensing is a low-cost technique for scene understanding. Current HMR systems adopt support vector machines (SVMs) and convolutional neural networks (CNNs) to classify radar signals. However, whether a deeper learning model could improve the system performance is currently not known. On the other hand, training a machine learning model requires a large dataset, but data gathering from experiment is cost-expensive and time-consuming. Although wireless channel models can be adopted for dataset generation, current channel models are mostly designed for communication rather than sensing. To address the above problems, this paper proposes a deep spectrogram network (DSN) by leveraging the residual mapping technique to enhance the HMR performance. Furthermore, a primitive based autoregressive hybrid (PBAH) channel model is developed, which facilitates efficient training and testing dataset generation for HMR in a virtual environment. Experimental results demonstrate that the proposed PBAH channel model matches the actual experimental data very well and the proposed DSN achieves significantly smaller recognition error than that of CNN.

Zezhong Zhang, Guangxu Zhu, Rui Wang et al.

Recently years, the attempts on distilling mobile data into useful knowledge has been led to the deployment of machine learning algorithms at the network edge. Principal component analysis (PCA) is a classic technique for extracting the linear structure of a dataset, which is useful for feature extraction and data compression. In this work, we propose the deployment of distributed PCA over a multi-access channel based on the algorithm of stochastic gradient descent to learn the dominant feature space of a distributed dataset at multiple devices. Over-the-air aggregation is adopted to reduce the multi-access latency, giving the name over-the-air PCA. The novelty of this design lies in exploiting channel noise to accelerate the descent in the region around each saddle point encountered by gradient descent, thereby increasing the convergence speed of over-the-air PCA. The idea is materialized by proposing a power-control scheme which detects the type of descent region and controlling the level of channel noise accordingly. The scheme is proved to achieve a faster convergence rate than in the case without power control.

Shuai Wang, Yuncong Hong, Rui Wang et al.

Edge federated learning (FL) is an emerging paradigm that trains a global parametric model from distributed datasets based on wireless communications. This paper proposes a unit-modulus over-the-air computation (UMAirComp) framework to facilitate efficient edge federated learning, which simultaneously uploads local model parameters and updates global model parameters via analog beamforming. The proposed framework avoids sophisticated baseband signal processing, leading to low communication delays and implementation costs. Training loss bounds of UMAirComp FL systems are derived and two low-complexity large-scale optimization algorithms, termed penalty alternating minimization (PAM) and accelerated gradient projection (AGP), are proposed to minimize the nonconvex nonsmooth loss bound. Simulation results show that the proposed UMAirComp framework with PAM algorithm achieves a smaller mean square error of model parameters' estimation, training loss, and test error compared with other benchmark schemes. Moreover, the proposed UMAirComp framework with AGP algorithm achieves satisfactory performance while reduces the computational complexity by orders of magnitude compared with existing optimization algorithms. Finally, we demonstrate the implementation of UMAirComp in a vehicle-to-everything autonomous driving simulation platform. It is found that autonomous driving tasks are more sensitive to model parameter errors than other tasks since the neural networks for autonomous driving contain sparser model parameters.

Shanfeng Huang, Shuai Wang, Rui Wang et al.

The ever-growing popularity and rapid improving of artificial intelligence (AI) have raised rethinking on the evolution of wireless networks. Mobile edge computing (MEC) provides a natural platform for AI applications since it is with rich computation resources to train machine learning (ML) models, as well as low-latency access to the data generated by mobile and internet of things (IoT) devices. In this paper, we present an infrastructure to perform ML tasks at an MEC server with the assistance of a reconfigurable intelligent surface (RIS). In contrast to conventional communication systems where the principal criterions are to maximize the throughput, we aim at maximizing the learning performance. Specifically, we minimize the maximum learning error of all participating users by jointly optimizing transmit power of mobile users, beamforming vectors of the base station (BS), and the phase-shift matrix of the RIS. An alternating optimization (AO)-based framework is proposed to optimize the three terms iteratively, where a successive convex approximation (SCA)-based algorithm is developed to solve the power allocation problem, closed-form expressions of the beamforming vectors are derived, and an alternating direction method of multipliers (ADMM)-based algorithm is designed together with an error level searching (ELS) framework to effectively solve the challenging nonconvex optimization problem of the phase-shift matrix. Simulation results demonstrate significant gains of deploying an RIS and validate the advantages of our proposed algorithms over various benchmarks. Lastly, a unified communication-training-inference platform is developed based on the CARLA platform and the SECOND network, and a use case (3D object detection in autonomous driving) for the proposed scheme is demonstrated on the developed platform.

Tong Zhang, Rui Wang

In this paper, we study the secure degrees-of-freedom (SDoF) characterization for the multiple-input multiple-output (MIMO) X channel with confidential messages and delayed CSIT. In particular, we propose a transmission scheme, which can be regarded as a generalization of the state-of-the-art scheme without security and with delayed CSIT. The key of this generalization is performing the security analysis, by which we derive the optimal duration of the artificial noise transmission phase. As a result, we derive the sum-SDoF lower bound. Furthermore, we reveal that if the number of receive antennas, denoted by N, is fixed, the minimum number of transmit antennas achieving the maximum of the lower bound is $\frac{7+\sqrt{33}}{8}N$.

Liangkai Zhou, Yuncong Hong, Shuai Wang et al.

Edge intelligence is an emerging network architecture that integrates sensing, communication, computing components, and supports various machine learning applications, where a fundamental communication question is: how to allocate the limited wireless resources (such as time, energy) to the simultaneous model training of heterogeneous learning tasks? Existing methods ignore two important facts: 1) different models have heterogeneous demands on training data; 2) there is a mismatch between the simulated environment and the real-world environment. As a result, they could lead to low learning performance in practice. This paper proposes the learning centric wireless resource allocation (LCWRA) scheme that maximizes the worst learning performance of multiple tasks. Analysis shows that the optimal transmission time has an inverse power relationship with respect to the generalization error. Finally, both simulation and experimental results are provided to verify the performance of the proposed LCWRA scheme and its robustness in real implementation.

Shuai Wang, Rui Wang, Qi Hao et al.

While machine-type communication (MTC) devices generate massive data, they often cannot process this data due to limited energy and computation power. To this end, edge intelligence has been proposed, which collects distributed data and performs machine learning at the edge. However, this paradigm needs to maximize the learning performance instead of the communication throughput, for which the celebrated water-filling and max-min fairness algorithms become inefficient since they allocate resources merely according to the quality of wireless channels. This paper proposes a learning centric power allocation (LCPA) method, which allocates radio resources based on an empirical classification error model. To get insights into LCPA, an asymptotic optimal solution is derived. The solution shows that the transmit powers are inversely proportional to the channel gain, and scale exponentially with the learning parameters. Experimental results show that the proposed LCPA algorithm significantly outperforms other power allocation algorithms.