arXiv subjects

Dan Su

Publications and source records attributed to Dan Su.

105 records · Page 6Linked to original sources

arXiv2020

Enhancing End-to-End Multi-channel Speech Separation via Spatial Feature Learning

Hand-crafted spatial features (e.g., inter-channel phase difference, IPD) play a fundamental role in recent deep learning based multi-channel speech separation (MCSS) methods. However, these manually designed spatial features are hard to incorporate into the end-to-end optimized MCSS framework. In this work, we propose an integrated architecture for learning spatial features directly from the multi-channel speech waveforms within an end-to-end speech separation framework. In this architecture, time-domain filters spanning signal channels are trained to perform adaptive spatial filtering. These filters are implemented by a 2d convolution (conv2d) layer and their parameters are optimized using a speech separation objective function in a purely data-driven fashion. Furthermore, inspired by the IPD formulation, we design a conv2d kernel to compute the inter-channel convolution differences (ICDs), which are expected to provide the spatial cues that help to distinguish the directional sources. Evaluation results on simulated multi-channel reverberant WSJ0 2-mix dataset demonstrate that our proposed ICD based MCSS model improves the overall signal-to-distortion ratio by 10.4% over the IPD based MCSS model.

eess.AS↗

arXiv2020

Mixup-breakdown: a consistency training method for improving generalization of speech separation models

Deep-learning based speech separation models confront poor generalization problem that even the state-of-the-art models could abruptly fail when evaluating them in mismatch conditions. To address this problem, we propose an easy-to-implement yet effective consistency based semi-supervised learning (SSL) approach, namely Mixup-Breakdown training (MBT). It learns a teacher model to "breakdown" unlabeled inputs, and the estimated separations are interpolated to produce more useful pseudo "mixup" input-output pairs, on which the consistency regularization could apply for learning a student model. In our experiment, we evaluate MBT under various conditions with ascending degrees of mismatch, including unseen interfering speech, noise, and music, and compare MBT's generalization capability against state-of-the-art supervised learning and SSL approaches. The result indicates that MBT significantly outperforms several strong baselines with up to 13.77% relative SI-SNRi improvement. Moreover, MBT only adds negligible computational overhead to standard training schemes.

eess.AS↗

arXiv2020

A Random Gossip BMUF Process for Neural Language Modeling

Neural network language model (NNLM) is an essential component of industrial ASR systems. One important challenge of training an NNLM is to leverage between scaling the learning process and handling big data. Conventional approaches such as block momentum provides a blockwise model update filtering (BMUF) process and achieves almost linear speedups with no performance degradation for speech recognition. However, it needs to calculate the model average from all computing nodes (e.g., GPUs) and when the number of computing nodes is large, the learning suffers from the severe communication latency. As a consequence, BMUF is not suitable under restricted network conditions. In this paper, we present a decentralized BMUF process, in which the model is split into different components, each of which is updated by communicating to some randomly chosen neighbor nodes with the same component, followed by a BMUF-like process. We apply this method to several LSTM language modeling tasks. Experimental results show that our approach achieves consistently better performance than conventional BMUF. In particular, we obtain a lower perplexity than the single-GPU baseline on the wiki-text-103 benchmark using 4 GPUs. In addition, no performance degradation is observed when scaling to 8 and 16 GPUs.

cs.CL↗

arXiv2019

Maximizing Mutual Information for Tacotron

End-to-end speech synthesis methods already achieve close-to-human quality performance. However compared to HMM-based and NN-based frame-to-frame regression methods, they are prone to some synthesis errors, such as missing or repeating words and incomplete synthesis. We attribute the comparatively high utterance error rate to the local information preference of conditional autoregressive models, and the ill-posed training objective of the model, which describes mostly the training status of the autoregressive module, but rarely that of the condition module. Inspired by InfoGAN, we propose to maximize the mutual information between the text condition and the predicted acoustic features to strengthen the dependency between them for CAR speech synthesis model, which would alleviate the local information preference issue and reduce the utterance error rate. The training objective of maximizing mutual information can be considered as a metric of the dependency between the autoregressive module and the condition module. Experiment results show that our method can reduce the utterance error rate.

eess.AS↗

arXiv2019

Overlapped speech recognition from a jointly learned multi-channel neural speech extraction and representation

We propose an end-to-end joint optimization framework of a multi-channel neural speech extraction and deep acoustic model without mel-filterbank (FBANK) extraction for overlapped speech recognition. First, based on a multi-channel convolutional TasNet with STFT kernel, we unify the multi-channel target speech enhancement front-end network and a convolutional, long short-term memory and fully connected deep neural network (CLDNN) based acoustic model (AM) with the FBANK extraction layer to build a hybrid neural network, which is thus jointly updated only by the recognition loss. The proposed framework achieves 28% word error rate reduction (WERR) over a separately optimized system on AISHELL-1 and shows consistent robustness to signal to interference ratio (SIR) and angle difference between overlapping speakers. Next, a further exploration shows that the speech recognition is improved with a simplified structure by replacing the FBANK extraction layer in the joint model with a learnable feature projection. Finally, we also perform the objective measurement of speech quality on the reconstructed waveform from the enhancement network in the joint model.

eess.AS↗

arXiv2019

DFSMN-SAN with Persistent Memory Model for Automatic Speech Recognition

Self-attention networks (SAN) have been introduced into automatic speech recognition (ASR) and achieved state-of-the-art performance owing to its superior ability in capturing long term dependency. One of the key ingredients is the self-attention mechanism which can be effectively performed on the whole utterance level. In this paper, we try to investigate whether even more information beyond the whole utterance level can be exploited and beneficial. We propose to apply self-attention layer with augmented memory to ASR. Specifically, we first propose a variant model architecture which combines deep feed-forward sequential memory network (DFSMN) with self-attention layers to form a better baseline model compared with a purely self-attention network. Then, we propose and compare two kinds of additional memory structures added into self-attention layers. Experiments on large-scale LVCSR tasks show that on four individual test sets, the DFSMN-SAN architecture outperforms vanilla SAN encoder by 5% relatively in character error rate (CER). More importantly, the additional memory structure provides further 5% to 11% relative improvement in CER.

eess.AS↗

arXiv2019

DurIAN: Duration Informed Attention Network For Multimodal Synthesis

In this paper, we present a generic and robust multimodal synthesis system that produces highly natural speech and facial expression simultaneously. The key component of this system is the Duration Informed Attention Network (DurIAN), an autoregressive model in which the alignments between the input text and the output acoustic features are inferred from a duration model. This is different from the end-to-end attention mechanism used, and accounts for various unavoidable artifacts, in existing end-to-end speech synthesis systems such as Tacotron. Furthermore, DurIAN can be used to generate high quality facial expression which can be synchronized with generated speech with/without parallel speech and face data. To improve the efficiency of speech generation, we also propose a multi-band parallel generation strategy on top of the WaveRNN model. The proposed Multi-band WaveRNN effectively reduces the total computational complexity from 9.8 to 5.5 GFLOPS, and is able to generate audio that is 6 times faster than real time on a single CPU core. We show that DurIAN could generate highly natural speech that is on par with current state of the art end-to-end systems, while at the same time avoid word skipping/repeating errors in those systems. Finally, a simple yet effective approach for fine-grained control of expressiveness of speech and facial expression is introduced.

cs.CL↗

arXiv2019

Phrase-Level Class based Language Model for Mandarin Smart Speaker Query Recognition

The success of speech assistants requires precise recognition of a number of entities on particular contexts. A common solution is to train a class-based n-gram language model and then expand the classes into specific words or phrases. However, when the class has a huge list, e.g., more than 20 million songs, a fully expansion will cause memory explosion. Worse still, the list items in the class need to be updated frequently, which requires a dynamic model updating technique. In this work, we propose to train pruned language models for the word classes to replace the slots in the root n-gram. We further propose to use a novel technique, named Difference Language Model (DLM), to correct the bias from the pruned language models. Once the decoding graph is built, we only need to recalculate the DLM when the entities in word classes are updated. Results show that the proposed method consistently and significantly outperforms the conventional approaches on all datasets, esp. for large lists, which the conventional approaches cannot handle.

cs.CL↗

arXiv2019

Teach an all-rounder with experts in different domains

In many automatic speech recognition (ASR) tasks, an ideal model has to be applicable over multiple domains. In this paper, we propose to teach an all-rounder with experts in different domains. Concretely, we build a multi-domain acoustic model by applying the teacher-student training framework. First, for each domain, a teacher model (domain-dependent model) is trained by fine-tuning a multi-condition model with domain-specific subset. Then all these teacher models are used to teach one single student model simultaneously. We perform experiments on two predefined domain setups. One is domains with different speaking styles, the other is nearfield, far-field and far-field with noise. Moreover, two types of models are examined: deep feedforward sequential memory network (DFSMN) and long short term memory (LSTM). Experimental results show that the model trained with this framework outperforms not only multi-condition model but also domain-dependent model. Specially, our training method provides up to 10.4% relative character error rate improvement over baseline model (multi-condition model).

eess.AS↗

arXiv2019

End-to-End Multi-Channel Speech Separation

The end-to-end approach for single-channel speech separation has been studied recently and shown promising results. This paper extended the previous approach and proposed a new end-to-end model for multi-channel speech separation. The primary contributions of this work include 1) an integrated waveform-in waveform-out separation system in a single neural network architecture. 2) We reformulate the traditional short time Fourier transform (STFT) and inter-channel phase difference (IPD) as a function of time-domain convolution with a special kernel. 3) We further relaxed those fixed kernels to be learnable, so that the entire architecture becomes purely data-driven and can be trained from end-to-end. We demonstrate on the WSJ0 far-field speech separation task that, with the benefit of learnable spatial features, our proposed end-to-end multi-channel model significantly improved the performance of previous end-to-end single-channel method and traditional multi-channel methods.

cs.SD↗

arXiv2019

Learning discriminative features in sequence training without requiring framewise labelled data

In this work, we try to answer two questions: Can deeply learned features with discriminative power benefit an ASR system's robustness to acoustic variability? And how to learn them without requiring framewise labelled sequence training data? As existing methods usually require knowing where the labels occur in the input sequence, they have so far been limited to many real-world sequence learning tasks. We propose a novel method which simultaneously models both the sequence discriminative training and the feature discriminative learning within a single network architecture, so that it can learn discriminative deep features in sequence training that obviates the need for presegmented training data. Our experiment in a realistic industrial ASR task shows that, without requiring any specific fine-tuning or additional complexity, our proposed models have consistently outperformed state-of-the-art models and significantly reduced Word Error Rate (WER) under all test conditions, and especially with highest improvements under unseen noise conditions, by relative 12.94%, 8.66% and 5.80%, showing our proposed models can generalize better to acoustic variability.

cs.LG↗

arXiv2018

Deep Extractor Network for Target Speaker Recovery From Single Channel Speech Mixtures

Speaker-aware source separation methods are promising workarounds for major difficulties such as arbitrary source permutation and unknown number of sources. However, it remains challenging to achieve satisfying performance provided a very short available target speaker utterance (anchor). Here we present a novel "deep extractor network" which creates an extractor point for the target speaker in a canonical high dimensional embedding space, and pulls together the time-frequency bins corresponding to the target speaker. The proposed model is different from prior works in that the canonical embedding space encodes knowledges of both the anchor and the mixture during an end-to-end training phase: First, embeddings for the anchor and mixture speech are separately constructed in a primary embedding space, and then combined as an input to feed-forward layers to transform to a canonical embedding space which we discover more stable than the primary one. Experimental results show that given a very short utterance, the proposed model can efficiently recover high quality target speech from a mixture, which outperforms various baseline models, with 5.2% and 6.6% relative improvements in SDR and PESQ respectively compared with a baseline oracle deep attracor model. Meanwhile, we show it can be generalized well to more than one interfering speaker.

cs.SD↗

arXiv2018

Discovering Communities of Malapps on Android-based Mobile Cyber-physical Systems

Android-based devices like smartphones have become ideal mobile cyber-physical systems (MCPS) due to their powerful processors and variety of sensors. In recent years, an explosively and continuously growing number of malicious applications (malapps) have posed a great threat to Android-based MCPS as well as users' privacy. The effective detection of malapps is an emerging yet crucial task. How to establish relationships among malapps, discover their potential communities, and explore their evolution process has become a challenging issue in effective detection of malapps. To deal with this issue, in this work, we are motivated to propose an automated community detection method for Android malapps by building a relation graph based on their static features. First, we construct a large feature set to profile the behaviors of malapps. Second, we propose an E-N algorithm by combining epsilon graph and k-nearest neighbor (k-NN) graph for graph construction. It solves the problem of an incomplete graph led by epsilon method and the problem of noise generated by k-NN graph. Finally, a community detection method, Infomap, is employed to explore the underlying structures of the relation graph, and obtain the communities of malapps. We evaluate our community detection method with 3996 malapp samples. Extensive experimental results show that our method outperforms the traditional clustering methods and achieves the best performance with rand statistic of 94.93% and accuracy of 79.53%.

cs.CR↗

arXiv2017

RGB-D Salient Object Detection Based on Discriminative Cross-modal Transfer Learning

In this work, we propose to utilize Convolutional Neural Networks to boost the performance of depth-induced salient object detection by capturing the high-level representative features for depth modality. We formulate the depth-induced saliency detection as a CNN-based cross-modal transfer problem to bridge the gap between the "data-hungry" nature of CNNs and the unavailability of sufficient labeled training data in depth modality. In the proposed approach, we leverage the auxiliary data from the source modality effectively by training the RGB saliency detection network to obtain the task-specific pre-understanding layers for the target modality. Meanwhile, we exploit the depth-specific information by pre-training a modality classification network that encourages modal-specific representations during the optimizing course. Thus, it could make the feature representations of the RGB and depth modalities as discriminative as possible. These two modules are pre-trained independently and then stitched to initialize and optimize the eventual depth-induced saliency detection model. Experiments demonstrate the effectiveness of the proposed novel pre-training strategy as well as the significant and consistent improvements of the proposed approach over other state-of-the-art methods.

cs.CV↗

arXiv2011

On the best constants of Hardy inequality in $\mathbb{R}^{n-k}\times (\mathbb{R}_{+})^{k}$ and related improvements

We compute the explicit sharp constants of Hardy inequalities in the cone $\mathbb{R}_{k_+}^{n}:=\mathbb{R}^{n-k}\times (\mathbb{R}_{+})^{k}=\{(x_{1},...,x_{n})|x_{n-k+1}>0,...,x_{n}>0\}$ with $1\leq k\leq n$. Furthermore, the spherical harmonic decomposition is given for a function $u\in C^{\infty}_{0}(\mathbb{R}_{k_+}^{n})$. Using this decomposition and following the idea of Tertikas and Zographopoulos, we obtain the Filippas-Tertikas improvement of the Hardy inequality.

math.FA↗