Wei Xia

Szu-Jui Chen, Wei Xia, John H. L. Hansen

In this study, we propose to investigate triplet loss for the purpose of an alternative feature representation for ASR. We consider a general non-semantic speech representation, which is trained with a self-supervised criteria based on triplet loss called TRILL, for acoustic modeling to represent the acoustic characteristics of each audio. This strategy is then applied to the CHiME-4 corpus and CRSS-UTDallas Fearless Steps Corpus, with emphasis on the 100-hour challenge corpus which consists of 5 selected NASA Apollo-11 channels. An analysis of the extracted embeddings provides the foundation needed to characterize training utterances into distinct groups based on acoustic distinguishing properties. Moreover, we also demonstrate that triplet-loss based embedding performs better than i-Vector in acoustic modeling, confirming that the triplet loss is more effective than a speaker feature. With additional techniques such as pronunciation and silence probability modeling, plus multi-style training, we achieve a +5.42% and +3.18% relative WER improvement for the development and evaluation sets of the Fearless Steps Corpus. To explore generalization, we further test the same technique on the 1 channel track of CHiME-4 and observe a +11.90% relative WER improvement for real test data.

Wei Xia, Chunlei Zhang, Chao Weng et al.

In this study, we investigate self-supervised representation learning for speaker verification (SV). First, we examine a simple contrastive learning approach (SimCLR) with a momentum contrastive (MoCo) learning framework, where the MoCo speaker embedding system utilizes a queue to maintain a large set of negative examples. We show that better speaker embeddings can be learned by momentum contrastive learning. Next, alternative augmentation strategies are explored to normalize extrinsic speaker variabilities of two random segments from the same speech utterance. Specifically, augmentation in the waveform largely improves the speaker representations for SV tasks. The proposed MoCo speaker embedding is further improved when a prototypical memory bank is introduced, which encourages the speaker embeddings to be closer to their assigned prototypes with an intermediate clustering step. In addition, we generalize the self-supervised framework to a semi-supervised scenario where only a small portion of the data is labeled. Comprehensive experiments on the Voxceleb dataset demonstrate that our proposed self-supervised approach achieves competitive performance compared with existing techniques, and can approach fully supervised results with partially labeled data.

Mufan Sang, Wei Xia, John H. L. Hansen

Despite speaker verification has achieved significant performance improvement with the development of deep neural networks, domain mismatch is still a challenging problem in this field. In this study, we propose a novel framework to disentangle speaker-related and domain-specific features and apply domain adaptation on the speaker-related feature space solely. Instead of performing domain adaptation directly on the feature space where domain information is not removed, using disentanglement can efficiently boost adaptation performance. To be specific, our model's input speech from the source and target domains is first encoded into different latent feature spaces. The adversarial domain adaptation is conducted on the shared speaker-related feature space to encourage the property of domain-invariance. Further, we minimize the mutual information between speaker-related and domain-specific features for both domains to enforce the disentanglement. Experimental results on the VOiCES dataset demonstrate that our proposed framework can effectively generate more speaker-discriminative and domain-invariant speaker representations with a relative 20.3% reduction of EER compared to the original ResNet-based system.

Mufan Sang, Wei Xia, John H. L. Hansen

In forensic applications, it is very common that only small naturalistic datasets consisting of short utterances in complex or unknown acoustic environments are available. In this study, we propose a pipeline solution to improve speaker verification on a small actual forensic field dataset. By leveraging large-scale out-of-domain datasets, a knowledge distillation based objective function is proposed for teacher-student learning, which is applied for short utterance forensic speaker verification. The objective function collectively considers speaker classification loss, Kullback-Leibler divergence, and similarity of embeddings. In order to advance the trained deep speaker embedding network to be robust for a small target dataset, we introduce a novel strategy to fine-tune the pre-trained student model towards a forensic target domain by utilizing the model as a finetuning start point and a reference in regularization. The proposed approaches are evaluated on the 1st48-UTD forensic corpus, a newly established naturalistic dataset of actual homicide investigations consisting of short utterances recorded in uncontrolled conditions. We show that the proposed objective function can efficiently improve the performance of teacher-student learning on short utterances and that our fine-tuning strategy outperforms the commonly used weight decay method by providing an explicit inductive bias towards the pre-trained model.

Zhenyu Wang, Wei Xia, John H. L. Hansen

Forensic audio analysis for speaker verification offers unique challenges due to location/scenario uncertainty and diversity mismatch between reference and naturalistic field recordings. The lack of real naturalistic forensic audio corpora with ground-truth speaker identity represents a major challenge in this field. It is also difficult to directly employ small-scale domain-specific data to train complex neural network architectures due to domain mismatch and loss in performance. Alternatively, cross-domain speaker verification for multiple acoustic environments is a challenging task which could advance research in audio forensics. In this study, we introduce a CRSS-Forensics audio dataset collected in multiple acoustic environments. We pre-train a CNN-based network using the VoxCeleb data, followed by an approach which fine-tunes part of the high-level network layers with clean speech from CRSS-Forensics. Based on this fine-tuned model, we align domain-specific distributions in the embedding space with the discrepancy loss and maximum mean discrepancy (MMD). This maintains effective performance on the clean set, while simultaneously generalizes the model to other acoustic domains. From the results, we demonstrate that diverse acoustic environments affect the speaker verification performance, and that our proposed approach of cross-domain adaptation can significantly improve the results in this scenario.

Wei Xia, John H. L. Hansen

In this study, we propose the global context guided channel and time-frequency transformations to model the long-range, non-local time-frequency dependencies and channel variances in speaker representations. We use the global context information to enhance important channels and recalibrate salient time-frequency locations by computing the similarity between the global context and local features. The proposed modules, together with a popular ResNet based model, are evaluated on the VoxCeleb1 dataset, which is a large scale speaker verification corpus collected in the wild. This lightweight block can be easily incorporated into a CNN model with little additional computational costs and effectively improves the speaker verification performance compared to the baseline ResNet-LDE model and the Squeeze&Excitation block by a large margin. Detailed ablation studies are also performed to analyze various factors that may impact the performance of the proposed modules. We find that by employing the proposed L2-tf-GTFC transformation block, the Equal Error Rate decreases from 4.56% to 3.07%, a relative 32.68% reduction, and a relative 27.28% improvement in terms of the DCF score. The results indicate that our proposed global context guided transformation modules can efficiently improve the learned speaker representations by achieving time-frequency and channel-wise feature recalibration.

Vinay Kothapally, Wei Xia, Shahram Ghorbani et al.

The reliability of using fully convolutional networks (FCNs) has been successfully demonstrated by recent studies in many speech applications. One of the most popular variants of these FCNs is the `U-Net', which is an encoder-decoder network with skip connections. In this study, we propose `SkipConvNet' where we replace each skip connection with multiple convolutional modules to provide decoder with intuitive feature maps rather than encoder's output to improve the learning capacity of the network. We also propose the use of optimal smoothing of power spectral density (PSD) as a pre-processing step, which helps to further enhance the efficiency of the network. To evaluate our proposed system, we use the REVERB challenge corpus to assess the performance of various enhancement approaches under the same conditions. We focus solely on monitoring improvements in speech quality and their contribution to improving the efficiency of back-end speech systems, such as speech recognition and speaker verification, trained on only clean speech. Experimental findings show that the proposed system consistently outperforms other approaches.

Wei Xia, Jing Huang, John H. L. Hansen

Speaker verification systems often degrade significantly when there is a language mismatch between training and testing data. Being able to improve cross-lingual speaker verification system using unlabeled data can greatly increase the robustness of the system and reduce human labeling costs. In this study, we introduce an unsupervised Adversarial Discriminative Domain Adaptation (ADDA) method to effectively learn an asymmetric mapping that adapts the target domain encoder to the source domain, where the target domain and source domain are speech data from different languages. ADDA, together with a popular Domain Adversarial Training (DAT) approach, are evaluated on a cross-lingual speaker verification task: the training data is in English from NIST SRE04-08, Mixer 6 and Switchboard, and the test data is in Chinese from AISHELL-I. We show that with the ADDA adaptation, Equal Error Rate (EER) of the x-vector system decreases from 9.331\% to 7.645\%, relatively 18.07\% reduction of EER, and 6.32\% reduction from DAT as well. Further data analysis of ADDA adapted speaker embedding shows that the learned speaker embeddings can perform well on speaker classification for the target domain data, and are less dependent with respect to the shift in language.

Wei Xia, Kazuhito Koishida

In this study, we introduce a convolutional time-frequency-channel "Squeeze and Excitation" (tfc-SE) module to explicitly model inter-dependencies between the time-frequency domain and multiple channels. The tfc-SE module consists of two parts: tf-SE block and c-SE block which are designed to provide attention on time-frequency and channel domain, respectively, for adaptively recalibrating the input feature map. The proposed tfc-SE module, together with a popular Convolutional Recurrent Neural Network (CRNN) model, are evaluated on a multi-channel sound event detection task with overlapping audio sources: the training and test data are synthesized TUT Sound Events 2018 datasets, recorded with microphone arrays. We show that the tfc-SE module can be incorporated into the CRNN model at a small additional computational cost and bring significant improvements on sound event detection accuracy. We also perform detailed ablation studies by analyzing various factors that may influence the performance of the SE blocks. We show that with the best tfc-SE block, error rate (ER) decreases from 0.2538 to 0.2026, relative 20.17\% reduction of ER, and 5.72\% improvement of F1 score. The results indicate that the learned acoustic embeddings with the tfc-SE module efficiently strengthen time-frequency and channel-wise feature representations to improve the discriminative performance.