No abstract

Deep dilated temporal convolutional networks (TCN) have been proved to be very effective in sequence modeling. In this paper we propose several improvements of TCN for end-to-end approach to monaural speech separation, which consists of 1) multi-scale dynamic weighted gated dilated convolutional pyramids network (FurcaPy), 2) gated TCN with intra-parallel convolutional components (FurcaPa), 3) weight-shared multi-scale gated TCN (FurcaSh), 4) dilated TCN with gated difference-convolutional component (FurcaSu), that all these networks take the mixed utterance of two speakers and maps it to two separated utterances, where each utterance contains only one speaker's voice. For the objective, we propose to train the network by directly optimizing utterance level signal-to-distortion ratio (SDR) in a permutation invariant training (PIT) style. Our experiments on the the public WSJ0-2mix data corpus results in 18.4dB SDR improvement, which shows our proposed networks can leads to performance improvement on the speaker separation task.

No abstract

J-vector and joint Bayesian have been proved to be very effective in text dependent speaker verification with shortduration speech. However current state-of-the-art framework often consider training the J-vector extractor and the joint Bayesian classifier separately. Such an approach will result in information loss for j-vector learning and also fail to exploit an end-to-end framework. In this paper we present a integrated approach to text dependent speaker verification, which consists of a siamese deep neural network that takes two variable length speech segments and maps them to the likelihood score and speaker/phrase labels, where the likelihood score as a loss guide is computed by a variant joint Bayesian model. The likelihood loss guide can constrain the j-vector extractor for improving the verification performance. Since the strengths of j-vector and joint Bayesian analysis appear complementary the joint learning significantly outperforms traditional separate training scheme. Our experiments on the the public RSR2015 part I data corpus demonstrate that this new training scheme can produce more discriminative j-vectors and leading to performance improvement on the speaker verification task.

In this paper, we present a method called HODGEPODGE 1 for largescale detection of sound events using weakly labeled, synthetic, and unlabeled data proposed in the Detection and Classification of Acoustic Scenes and Events (DCASE) 2019 challenge Task 4: Sound event detection in domestic environments. To perform this task, we adopted the convolutional recurrent neural networks (CRNN) as our backbone network. In order to deal with a small amount of tagged data and a large amounts of unlabeled in-domain data, we aim to focus primarily on how to apply semi-supervise learning methods efficiently to make full use of limited data. Three semi-supervised learning principles have been used in our system, including: 1) Consistency regularization applies data augmentation; 2) MixUp regularizer requiring that the predictions for a interpolation of two inputs is close to the interpolation of the prediction for each individual input; 3) MixUp regularization applies to interpolation between data augmentations. We also tried an ensemble of various models, which are trained by using different semisupervised learning principles. Our proposed approach significantly improved the performance of the baseline, achieving the event-based f-measure of 42.0% compared to 25.8% event-based f-measure of the baseline in the provided official evaluation dataset. Our submissions ranked third among 18 teams in the task 4. 1 HODGEPODGE has two layers of meanings. The first layer is the variety of training data involved in the method, including weakly labeled, synthetic, and unlabeled data. The second layer refers to several semi-supervised principles involved in our method.