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 will be open-sourced in https://github.com/Audio-WestlakeU/NBSS.
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.
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.
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.
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.
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.
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 processes 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).
In recent years, self-supervised learning (SSL) has emerged as a popular approach for learning audio representations. The ultimate goal of audio self-supervised pre-training is to transfer knowledge to downstream audio tasks, generally including clip-level and frame-level tasks. Clip-level tasks classify the scene or sound of an entire audio clip, e.g. audio tagging, instrument recognition, etc. While frame-level tasks detect event-level timestamps from an audio clip, e.g. sound event detection, speaker diarization, etc. Prior studies primarily evaluate on clip-level downstream tasks. Frame-level tasks are important for fine-grained acoustic scene/event understanding, and are generally more challenging than clip-level tasks. In order to tackle both clip-level and frame-level tasks, this paper proposes two self-supervised audio representation learning methods: ATST-Clip and ATST-Frame, responsible for learning clip-level and frame-level representations, respectively. ATST stands for Audio Teacher-Student Transformer, which means both methods use a transformer encoder and a teacher-student training scheme.Experimental results show that our ATST-Frame model obtains state-of-the-art (SOTA) performance 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.