Search

This work proposes a frame-wise online/streaming end-to-end neural diarization (EEND) method, which detects speaker activities in a frame-in-frame-out fashion. The proposed model mainly consists of a causal embedding encoder and an online attractor decoder. Speakers are modeled in the self-attention-based decoder along both the time and speaker dimensions, and frame-wise speaker attractors are automatically generated and updated for new speakers and existing speakers, respectively. Retention mechanism is employed and especially adapted for long-form diarization with a linear temporal complexity. A multi-step progressive training strategy is proposed for gradually learning from easy tasks to hard tasks in terms of the number of speakers and audio length. Finally, the proposed model (referred to as long-form streaming EEND, LS-EEND) is able to perform streaming diarization for a high (up to 8) and flexible number speakers and very long (say one hour) audio recordings. Experiments on various simulated and real-world datasets show that: 1) when not using oracle speech activity information, the proposed model achieves new state-of-the-art online diarization error rate on all datasets, including CALLHOME (12.11%), DIHARD II (27.58%), DIHARD III (19.61%), and AMI (20.76%); 2) Due to the frame-in-frame-out processing fashion and the linear temporal complexity, the proposed model achieves several times lower real-time-factor than comparison online diarization models.

This paper works on streaming automatic speech recognition (ASR). Mamba, a recently proposed state space model, has demonstrated the ability to match or surpass Transformers in various tasks while benefiting from a linear complexity advantage. We explore the efficiency of Mamba encoder for streaming ASR and propose an associated lookahead mechanism for leveraging controllable future information. Additionally, a streaming-style unimodal aggregation (UMA) method is implemented, which automatically detects token activity and streamingly triggers token output, and meanwhile aggregates feature frames for better learning token representation. Based on UMA, an early termination (ET) method is proposed to further reduce recognition latency. Experiments conducted on two Mandarin Chinese datasets demonstrate that the proposed model achieves competitive ASR performance in terms of both recognition accuracy and latency., Comment: Submitted to ICASSP 2025

The training of deep learning-based multichannel speech enhancement and source localization systems relies heavily on the simulation of room impulse response and multichannel diffuse noise, due to the lack of large-scale real-recorded datasets. However, the acoustic mismatch between simulated and real-world data could degrade the model performance when applying in real-world scenarios. To bridge this simulation-to-real gap, this paper presents a new relatively large-scale Real-recorded and annotated Microphone Array speech&Noise (RealMAN) dataset. The proposed dataset is valuable in two aspects: 1) benchmarking speech enhancement and localization algorithms in real scenarios; 2) offering a substantial amount of real-world training data for potentially improving the performance of real-world applications. Specifically, a 32-channel array with high-fidelity microphones is used for recording. A loudspeaker is used for playing source speech signals (about 35 hours of Mandarin speech). A total of 83.7 hours of speech signals (about 48.3 hours for static speaker and 35.4 hours for moving speaker) are recorded in 32 different scenes, and 144.5 hours of background noise are recorded in 31 different scenes. Both speech and noise recording scenes cover various common indoor, outdoor, semi-outdoor and transportation environments, which enables the training of general-purpose speech enhancement and source localization networks. To obtain the task-specific annotations, speaker location is annotated with an omni-directional fisheye camera by automatically detecting the loudspeaker. The direct-path signal is set as the target clean speech for speech enhancement, which is obtained by filtering the source speech signal with an estimated direct-path propagation filter., Comment: accepted by NIPS 2024

In end-to-end multi-channel speech enhancement, the traditional approach of designating one microphone signal as the reference for processing may not always yield optimal results. The limitation is particularly in scenarios with large distributed microphone arrays with varying speaker-to-microphone distances or compact, highly directional microphone arrays where speaker or microphone positions change over time. Current mask-based methods often fix the reference channel during training, which makes it not possible to adaptively select the reference channel for optimal performance. To address this problem, we introduce an adaptive approach for selecting the optimal reference channel. Our method leverages a multi-channel masking-based scheme, where multiple masked signals are combined to generate a single-channel output signal. This enhanced signal is then used for loss calculation, while the reference clean speech is adjusted based on the highest scale-invariant signal-to-distortion ratio (SI-SDR). The experimental results on the Spear challenge simulated dataset D4 demonstrate the superiority of our proposed method over the conventional approach of using a fixed reference channel with single-channel masking., Comment: Accepted by EUSIPCO 2024

The increasing difficulty in accurately detecting forged images generated by AIGC(Artificial Intelligence Generative Content) poses many risks, necessitating the development of effective methods to identify and further locate forged areas. In this paper, to facilitate research efforts, we construct a DA-HFNet forged image dataset guided by text or image-assisted GAN and Diffusion model. Our goal is to utilize a hierarchical progressive network to capture forged artifacts at different scales for detection and localization. Specifically, it relies on a dual-attention mechanism to adaptively fuse multi-modal image features in depth, followed by a multi-branch interaction network to thoroughly interact image features at different scales and improve detector performance by leveraging dependencies between layers. Additionally, we extract more sensitive noise fingerprints to obtain more prominent forged artifact features in the forged areas. Extensive experiments validate the effectiveness of our approach, demonstrating significant performance improvements compared to state-of-the-art methods for forged image detection and localization.The code and dataset will be released in the future.

Extracting direct-path spatial feature is crucial for sound source localization in adverse acoustic environments. This paper proposes the IPDnet, a neural network that estimates direct-path inter-channel phase difference (DP-IPD) of sound sources from microphone array signals. The estimated DP-IPD can be easily translated to source location based on the known microphone array geometry. First, a full-band and narrow-band fusion network is proposed for DP-IPD estimation, in which alternating narrow-band and full-band layers are responsible for estimating the rough DP-IPD information in one frequency band and capturing the frequency correlations of DP-IPD, respectively. Second, a new multi-track DP-IPD learning target is proposed for the localization of flexible number of sound sources. Third, the IPDnet is extend to handling variable microphone arrays, once trained which is able to process arbitrary microphone arrays with different number of channels and array topology. Experiments of multiple-moving-speaker localization are conducted on both simulated and real-world data, which show that the proposed full-band and narrow-band fusion network and the proposed multi-track DP-IPD learning target together achieves excellent sound source localization performance. Moreover, the proposed variable-array model generalizes well to unseen microphone arrays.

If two conducting or insulating inclusions are closely located, the gradient of the solution may become arbitrarily large as the distance between inclusions tends to zero, resulting in high concentration of stress in between two inclusions. This happens if the bonding of the inclusions and the matrix is perfect, meaning that the potential and flux are continuous across the interface. In this paper, we consider the case when the bonding is imperfect. We consider the case when there are two circular inclusions of the same radii with the imperfect bonding interfaces and prove that the gradient of the solution is bounded regardless of the distance between inclusions if the bonding parameter is finite. This result is of particular importance since the imperfect bonding interface condition is an approximation of the membrane structure of biological inclusions such as biological cells., Comment: 25 pages, 1 figure

In this work, we extend our previously proposed offline SpatialNet for long-term streaming multichannel speech enhancement in both static and moving speaker scenarios. SpatialNet exploits spatial information, such as the spatial/steering direction of speech, for discriminating between target speech and interferences, and achieved outstanding performance. The core of SpatialNet is a narrow-band self-attention module used for learning the temporal dynamic of spatial vectors. Towards long-term streaming speech enhancement, we propose to replace the offline self-attention network with online networks that have linear inference complexity w.r.t signal length and meanwhile maintain the capability of learning long-term information. Three variants are developed based on (i) masked self-attention, (ii) Retention, a self-attention variant with linear inference complexity, and (iii) Mamba, a structured-state-space-based RNN-like network. Moreover, we investigate the length extrapolation ability of different networks, namely test on signals that are much longer than training signals, and propose a short-signal training plus long-signal fine-tuning strategy, which largely improves the length extrapolation ability of the networks within limited training time. Overall, the proposed online SpatialNet achieves outstanding speech enhancement performance for long audio streams, and for both static and moving speakers. The proposed method is open-sourced in https://github.com/Audio-WestlakeU/NBSS., Comment: Accepted by IEEE Signal Processing Letters

In this work, we propose Mel-FullSubNet, a single-channel Mel-spectrogram denoising and dereverberation network for improving both speech quality and automatic speech recognition (ASR) performance. Mel-FullSubNet takes as input the noisy and reverberant Mel-spectrogram and predicts the corresponding clean Mel-spectrogram. The enhanced Mel-spectrogram can be either transformed to speech waveform with a neural vocoder or directly used for ASR. Mel-FullSubNet encapsulates interleaved full-band and sub-band networks, for learning the full-band spectral pattern of signals and the sub-band/narrow-band properties of signals, respectively. Compared to linear-frequency domain or time-domain speech enhancement, the major advantage of Mel-spectrogram enhancement is that Mel-frequency presents speech in a more compact way and thus is easier to learn, which will benefit both speech quality and ASR. Experimental results demonstrate a significant improvement in both speech quality and ASR performance achieved by the proposed model.

Having shown early promise, free-space optical communications (FSO) face formidable challenges in the age of information explosion. The ever-growing demand for greater channel communication capacity is one of the challenges. The inter-channel crosstalk, which severely degrades the quality of transmitted information, creates another roadblock in the way of efficient FSO implementation. Here we advance theoretically and realize experimentally a potentially high-capacity FSO protocol that enables high-fidelity transfer of an image, or set of images through a complex environment. In our protocol, we complement random light structuring at the transmitter with a deep learning image classification platform at the receiver. Multiplexing novel, independent, mutually orthogonal degrees of freedom available to structured random light can potentially significantly boost the channel communication capacity of our protocol without introducing any deleterious crosstalk. Specifically, we show how one can multiplex the degrees of freedom associated with the source coherence radius and a spatial position of a beamlet within an array of structured random beams to greatly enhance the capacity of our communication link. The superb resilience of structured random light to environmental noise, as well as extreme efficiency of deep learning networks at classifying images guarantees high-fidelity image transfer within the framework of our protocol., Comment: 18 pages,13 figures

When two stiff inclusions are closely located, the gradient of the solution to the Lam\'{e} system, in other words the stress, may become arbitrarily large as the distance between two inclusions tends to zero. To compute the gradient of the solution in the narrow region, extremely fine meshes are required. It is a challenging problem to numerically compute the stress near the narrow region between two inclusions of general shapes as their distance goes to zero. A recent study [15] has shown that the major singularity of the gradient can be extracted in an explicit way for two general shaped inclusions. Thus the complexity of the computation can be greatly reduced by removing the singular term and it suffices to compute the residual term only using regular meshes. The goal of this paper is to numerically compute the stress concentration in a fast and efficient way. In this paper, we compute the value of the stress concentration factor, which is the normalized magnitude of the stress concentration, for general shaped domain as the distance between two inclusions tends to zero. We also compute the solution for two closely located inclusions of general shapes and show the convergence of the solution. Only regular meshes are used in our numerical computation and the results clearly show that the characterization of the singular term method can be efficiently used for computation of the stress concentration between two closely located inclusions of general shapes., Comment: 18 pages, 13 figures

Supervised learning methods have shown effectiveness in estimating spatial acoustic parameters such as time difference of arrival, direct-to-reverberant ratio and reverberation time. However, they still suffer from the simulation-to-reality generalization problem due to the mismatch between simulated and real-world acoustic characteristics and the deficiency of annotated real-world data. To this end, this work proposes a self-supervised method that takes full advantage of unlabeled data for spatial acoustic parameter estimation. First, a new pretext task, i.e. cross-channel signal reconstruction (CCSR), is designed to learn a universal spatial acoustic representation from unlabeled multi-channel microphone signals. We mask partial signals of one channel and ask the model to reconstruct them, which makes it possible to learn spatial acoustic information from unmasked signals and extract source information from the other microphone channel. An encoder-decoder structure is used to disentangle the two kinds of information. By fine-tuning the pre-trained spatial encoder with a small annotated dataset, this encoder can be used to estimate spatial acoustic parameters. Second, a novel multi-channel audio Conformer (MC-Conformer) is adopted as the encoder model architecture, which is suitable for both the pretext and downstream tasks. It is carefully designed to be able to capture the local and global characteristics of spatial acoustics exhibited in the time-frequency domain. Experimental results of five acoustic parameter estimation tasks on both simulated and real-world data show the effectiveness of the proposed method. To the best of our knowledge, this is the first self-supervised learning method in the field of spatial acoustic representation learning and multi-channel audio signal processing.

The nearby radio galaxy M87 offers a unique opportunity to explore the connections between the central supermassive black hole and relativistic jets. Previous studies of the inner region of M87 revealed a wide opening angle for the jet originating near the black hole. The Event Horizon Telescope resolved the central radio source and found an asymmetric ring structure consistent with expectations from General Relativity. With a baseline of 17 years of observations, there was a shift in the jet's transverse position, possibly arising from an eight to ten-year quasi-periodicity. However, the origin of this sideways shift remains unclear. Here we report an analysis of radio observations over 22 years that suggests a period of about 11 years in the position angle variation of the jet. We infer that we are seeing a spinning black hole that induces the Lense-Thirring precession of a misaligned accretion disk. Similar jet precession may commonly occur in other active galactic nuclei but has been challenging to detect owing to the small magnitude and long period of the variation., Comment: 41 pages, 7 figures, 7 tables

This work proposes a frame-wise online/streaming end-to-end neural diarization (FS-EEND) method in a frame-in-frame-out fashion. To frame-wisely detect a flexible number of speakers and extract/update their corresponding attractors, we propose to leverage a causal speaker embedding encoder and an online non-autoregressive self-attention-based attractor decoder. A look-ahead mechanism is adopted to allow leveraging some future frames for effectively detecting new speakers in real time and adaptively updating speaker attractors. The proposed method processes the audio stream frame by frame, and has a low inference latency caused by the look-ahead frames. Experiments show that, compared with the recently proposed block-wise online methods, our method FS-EEND achieves state-of-the-art diarization results, with a low inference latency and computational cost.

In indoor scenes, reverberation is a crucial factor in degrading the perceived quality and intelligibility of speech. In this work, we propose a generative dereverberation method. Our approach is based on a probabilistic model utilizing a recurrent variational auto-encoder (RVAE) network and the convolutive transfer function (CTF) approximation. Different from most previous approaches, the output of our RVAE serves as the prior of the clean speech. And our target is the maximum a posteriori (MAP) estimation of clean speech, which is achieved iteratively through the expectation maximization (EM) algorithm. The proposed method integrates the capabilities of network-based speech prior modelling and CTF-based observation modelling. Experiments on single-channel speech dereverberation show that the proposed generative method noticeably outperforms the advanced discriminative networks., Comment: Submitted to ICASSP2024

Sound event detection (SED) often suffers from the data deficiency problem. The recent baseline system in the DCASE2023 challenge task 4 leverages the large pretrained self-supervised learning (SelfSL) models to mitigate such restriction, where the pretrained models help to produce more discriminative features for SED. However, the pretrained models are regarded as a frozen feature extractor in the challenge baseline system and most of the challenge submissions, and fine-tuning of the pretrained models has been rarely studied. In this work, we study the fine-tuning method of the pretrained models for SED. We first introduce ATST-Frame, our newly proposed SelfSL model, to the SED system. ATST-Frame was especially designed for learning frame-level representations of audio signals and obtained state-of-the-art (SOTA) performances on a series of downstream tasks. We then propose a fine-tuning method for ATST-Frame using both (in-domain) unlabelled and labelled SED data. Our experiments show that, the proposed method overcomes the overfitting problem when fine-tuning the large pretrained network, and our SED system obtains new SOTA results of 0.587/0.812 PSDS1/PSDS2 scores on the DCASE challenge task 4 dataset., Comment: 5 pages, 3 figures, camera-ready version for ICASSP 2024

This paper works on non-autoregressive automatic speech recognition. A unimodal aggregation (UMA) is proposed to segment and integrate the feature frames that belong to the same text token, and thus to learn better feature representations for text tokens. The frame-wise features and weights are both derived from an encoder. Then, the feature frames with unimodal weights are integrated and further processed by a decoder. Connectionist temporal classification (CTC) loss is applied for training. Compared to the regular CTC, the proposed method learns better feature representations and shortens the sequence length, resulting in lower recognition error and computational complexity. Experiments on three Mandarin datasets show that UMA demonstrates superior or comparable performance to other advanced non-autoregressive methods, such as self-conditioned CTC. Moreover, by integrating self-conditioned CTC into the proposed framework, the performance can be further noticeably improved., Comment: Accepted by ICASSP 2024

In recent years, high-frequency trading has emerged as a crucial strategy in stock trading. This study aims to develop an advanced high-frequency trading algorithm and compare the performance of three different mathematical models: the combination of the cross-entropy loss function and the quasi-Newton algorithm, the FCNN model, and the vector machine. The proposed algorithm employs neural network predictions to generate trading signals and execute buy and sell operations based on specific conditions. By harnessing the power of neural networks, the algorithm enhances the accuracy and reliability of the trading strategy. To assess the effectiveness of the algorithm, the study evaluates the performance of the three mathematical models. The combination of the cross-entropy loss function and the quasi-Newton algorithm is a widely utilized logistic regression approach. The FCNN model, on the other hand, is a deep learning algorithm that can extract and classify features from stock data. Meanwhile, the vector machine is a supervised learning algorithm recognized for achieving improved classification results by mapping data into high-dimensional spaces. By comparing the performance of these three models, the study aims to determine the most effective approach for high-frequency trading. This research makes a valuable contribution by introducing a novel methodology for high-frequency trading, thereby providing investors with a more accurate and reliable stock trading strategy.

This work proposes a neural network to extensively exploit spatial information for multichannel joint speech separation, denoising and dereverberation, named SpatialNet. In the short-time Fourier transform (STFT) domain, the proposed network performs end-to-end speech enhancement. It is mainly composed of interleaved narrow-band and cross-band blocks to respectively exploit narrow-band and cross-band spatial information. The narrow-band blocks process frequencies independently, and use self-attention mechanism and temporal convolutional layers to respectively perform spatial-feature-based speaker clustering and temporal smoothing/filtering. The cross-band blocks process frames independently, and use full-band linear layer and frequency convolutional layers to respectively learn the correlation between all frequencies and adjacent frequencies. Experiments are conducted on various simulated and real datasets, and the results show that 1) the proposed network achieves the state-of-the-art performance on almost all tasks; 2) the proposed network suffers little from the spectral generalization problem; and 3) the proposed network is indeed performing speaker clustering (demonstrated by attention maps).

Self-supervised learning (SSL) has emerged as a popular approach for learning audio representations. One goal of audio self-supervised pre-training is to transfer knowledge to downstream audio tasks, generally including clip-level and frame-level tasks. While frame-level tasks are important for fine-grained acoustic scene/event understanding, prior studies primarily evaluate on clip-level downstream tasks. In order to tackle both clip-level and frame-level tasks, this paper proposes Audio Teacher-Student Transformer (ATST), with a clip-level version (named ATST-Clip) and a frame-level version (named ATST-Frame), responsible for learning clip-level and frame-level representations, respectively. Both methods use a Transformer encoder and a teacher-student training scheme. We have carefully designed the view creation strategy for ATST-Clip and ATST-Frame. Specifically, ATST-Clip uses segment-wise data augmentations, and ATST-Frame integrates frame-wise data augmentations and masking. Experimental results show that our ATST-Frame model obtains state-of-the-art (SOTA) performances on most of the clip-level and frame-level downstream tasks. Especially, it outperforms other models by a large margin on the frame-level sound event detection task. In addition, the performance can be further improved by combining the two models through knowledge distillation. Our code is available online., Comment: Submitted to IEEE TASLP. arXiv admin note: text overlap with arXiv:2204.12076

Extracting direct-path spatial features is critical for sound source localization in adverse acoustic environments. This paper proposes a full-band and narrow-band fusion network for estimating direct-path inter-channel phase difference (DP-IPD) from microphone signals. The alternating full-band and narrow-band layers are responsible for learning the full-band correlation and narrow-band extraction of DP-IPD, respectively. Experiments show that the proposed network noticeably outperforms other advanced methods on both simulated and real-world data.

This paper considers the two-dimensional narrow escape problem in a domain which is composed of a relatively big head and several thin necks. The narrow escape problem is to compute the mean first passage time(MFPT) of a Brownian particle traveling from inside the head to the end of the necks. The original model for MFPT is to solve a mixed Dirichlet-Neumann boundary value problem for the Poisson equation in the composite domain, and is computationally challenging. In this paper, we compute the MFPT by solving an equivalent Neumann-Robin type boundary value problem. By solving the new model, we obtain the high order asymptotic expansion of the MFPT. We also conduct numerical experiments to show the accuracy of the high order expansion. As far as we know, this is the first result on high order asymptotic solution for NEP in a general shaped domain with several absorbing neck windows. This work is motivated by \cite{Li}, where the Neumann-Robin model was proposed to solve the NEP in a domain with a single absorbing neck., Comment: 23 pages, 7 figures

In this paper we present our contribution to the TSAR-2022 Shared Task on Lexical Simplification of the EMNLP 2022 Workshop on Text Simplification, Accessibility, and Readability. Our approach builds on and extends the unsupervised lexical simplification system with pretrained encoders (LSBert) system in the following ways: For the subtask of simplification candidate selection, it utilizes a RoBERTa transformer language model and expands the size of the generated candidate list. For subsequent substitution ranking, it introduces a new feature weighting scheme and adopts a candidate filtering method based on textual entailment to maximize semantic similarity between the target word and its simplification. Our best-performing system improves LSBert by 5.9% accuracy and achieves second place out of 33 ranked solutions., Comment: accepted at EMNLP2022

State-of-the-art text simplification (TS) systems adopt end-to-end neural network models to directly generate the simplified version of the input text, and usually function as a blackbox. Moreover, TS is usually treated as an all-purpose generic task under the assumption of homogeneity, where the same simplification is suitable for all. In recent years, however, there has been increasing recognition of the need to adapt the simplification techniques to the specific needs of different target groups. In this work, we aim to advance current research on explainable and controllable TS in two ways: First, building on recently proposed work to increase the transparency of TS systems, we use a large set of (psycho-)linguistic features in combination with pre-trained language models to improve explainable complexity prediction. Second, based on the results of this preliminary task, we extend a state-of-the-art Seq2Seq TS model, ACCESS, to enable explicit control of ten attributes. The results of experiments show (1) that our approach improves the performance of state-of-the-art models for predicting explainable complexity and (2) that explicitly conditioning the Seq2Seq model on ten attributes leads to a significant improvement in performance in both within-domain and out-of-domain settings., Comment: accepted at EMNLP2022

FullSubNet is our recently proposed real-time single-channel speech enhancement network that achieves outstanding performance on the Deep Noise Suppression (DNS) Challenge dataset. A number of variants of FullSubNet have been proposed, but they all focus on the structure design towards better performance and are rarely concerned with computational efficiency. For many speech enhancement applications, a key feature is that system runs on a real-time, latency-sensitive, battery-powered platform, which strictly limits the algorithm latency and computational complexity. In this work, we propose a new architecture named Fast FullSubNet dedicated to accelerating the computation of FullSubNet. Specifically, Fast FullSubNet processes sub-band speech spectra in the mel-frequency domain by using cascaded linear-to-mel full-band, sub-band, and mel-to-linear full-band models such that frequencies involved in the sub-band computation are vastly reduced. After that, a down-sampling operation is proposed for the sub-band input sequence to further reduce the computational complexity along the time axis. Experimental results show that, compared to FullSubNet, Fast FullSubNet has only 13\% computational complexity and 16\% processing time, and achieves comparable or even better performance. Code and audio samples are available at https://github.com/Audio-WestlakeU/FullSubNet.

The East Asian VLBI Network (EAVN) is an international VLBI facility in East Asia and is operated under mutual collaboration between East Asian countries, as well as part of Southeast Asian and European countries. EAVN currently consists of 16 radio telescopes and three correlators located in China, Japan, and Korea, and is operated mainly at three frequency bands, 6.7, 22, and 43 GHz with the longest baseline length of 5078 km, resulting in the highest angular resolution of 0.28 milliarcseconds at 43 GHz. One of distinct capabilities of EAVN is multi-frequency simultaneous data reception at nine telescopes, which enable us to employ the frequency phase transfer technique to obtain better sensitivity at higher observing frequencies. EAVN started its open-use program in the second half of 2018, providing a total observing time of more than 1100 hours in a year. EAVN fills geographical gap in global VLBI array, resulting in enabling us to conduct contiguous high-resolution VLBI observations. EAVN has produced various scientific accomplishments especially in observations toward active galactic nuclei, evolved stars, and star-forming regions. These activities motivate us to initiate launch of the 'Global VLBI Alliance' to provide an opportunity of VLBI observation with the longest baselines on the earth., Comment: 27 pages, appeared in Galaxies special issue 'Challenges in Understanding Black Hole Powered Jets with VLBI' as an invited review

This work proposes a multichannel narrow-band speech separation network. In the short-time Fourier transform (STFT) domain, the proposed network processes each frequency independently, and all frequencies use a shared network. For each frequency, the network performs end-to-end speech separation, namely taking as input the STFT coefficients of microphone signals, and predicting the separated STFT coefficients of multiple speakers. The proposed network learns to cluster the frame-wise spatial/steering vectors that belong to different speakers. It is mainly composed of three components. First, a self-attention network. Clustering of spatial vectors shares a similar principle with the self-attention mechanism in the sense of computing the similarity of vectors and then aggregating similar vectors. Second, a convolutional feed-forward network. The convolutional layers are employed for signal smoothing and reverberation processing. Third, a novel hidden-layer normalization method, i.e. group batch normalization (GBN), is especially designed for the proposed narrow-band network to maintain the distribution of hidden units over frequencies. Overall, the proposed network is named NBC2, as it is a revised version of our previous NBC (narrow-band conformer) network. Experiments show that 1) the proposed network outperforms other state-of-the-art methods by a large margin, 2) the proposed GBN improves the signal-to-distortion ratio by 3 dB, relative to other normalization methods, such as batch/layer/group normalization, 3) the proposed narrow-band network is spectrum-agnostic, as it does not learn spectral patterns, and 4) the proposed network is indeed performing frame clustering (demonstrated by the attention maps)., Comment: submitted to TASLP

In multichannel speech enhancement, both spectral and spatial information are vital for discriminating between speech and noise. How to fully exploit these two types of information and their temporal dynamics remains an interesting research problem. As a solution to this problem, this paper proposes a multi-cue fusion network named McNet, which cascades four modules to respectively exploit the full-band spatial, narrow-band spatial, sub-band spectral, and full-band spectral information. Experiments show that each module in the proposed network has its unique contribution and, as a whole, notably outperforms other state-of-the-art methods., Comment: submitted to icassp 2023