The ability to generalize to mismatches between training and testing conditions and resist interference from other speakers is crucial for the performance of speaker verification. In this paper, we propose two novel approaches to improve the generalization ability to deal with the mismatched recorded scenarios and languages in test conditions and to reduce the influence of interference from other speakers on the similarity measurement of two speaker embeddings. First, parent embedding learning (PEL) is used for model training, which exploits the generalization ability of the shared structure to improve the representation of speaker embeddings. Second, partial adaptive score normalization (PAS-Norm) is used to reduce the influence of interference from other speakers on embedding-based similarity measures. In the experiments, the speaker embedding models are trained using the VoxCeleb2 dataset, and the performance is evaluated on four other datasets under different conditions, including VoxCeleb1, Librispeech, SITW, and CN-Celeb datasets. In the experiments on VoxCeleb1, evaluation results considering a large number of verification speakers and identity restrictions show that the proposed PEL-based system reduces the EER by 6.0% and 4.9% in these two cases, respectively, compared to the state-of-the-art (SOTA) system. Furthermore, in the experiments evaluating speaker verification in mismatch conditions on SITW and CN-Celeb, the proposed PEL-based system also outperforms the SOTA system. In the language mismatched conditions, the EER is reduced by 8.3%. For the evaluation of the influence of interference from other speakers, the EER is significantly reduced by 24.4% when PAS-Norm is used instead of the baseline AS-Norm score normalization method.

Speech content is closely related to the stability of speaker embeddings in speaker verification tasks. In this paper, we propose a novel architecture based on self-constraint learning (SCL) and reconstruction task (RT) to remove the influence of phonetic information on speaker embedding generation. First, SCL is used to reduce the divergence of frame-level features, which can avoid ambiguity between the resulting embeddings of the two utterances being compared. Second, RT is used to further remove phonetic information in frame-level layers, focusing on speaker-discriminative feature transformation. In our experiments, the speaker embedding models were trained on the VoxCeleb2 dataset and evaluated on the VoxCeleb1, Librispeech, SITW and VoxMovies datasets. Experimental results on VoxCeleb1 show that the proposed DROP-TDNN system reduced the EER by 7.5%, compared to the state-of-the-art ECAPA-TDNN system. Furthermore, the proposed DROP-TDNN system also outperformed the ECAPA-TDNN system in the experiments on SITW, Librispeech and VoxMovies under cross-dataset conditions. In the experiments on SITW, the proposed system reduced the EER by 3.4% compared to the ECAPA-TDNN system. In the experiments on Librispeech, the proposed system demonstrated the advantage of removing phonetic information under the clean speech condition, with a significant reduction of 25.5% in EER compared to the ECAPA-TDNN system. In the experiments on VoxMovies, the proposed system reduced the EER by up to 7.9% compared to the ECAPA-TDNN system under different pronunciation and background conditions.

In the clinical diagnosis of mood disorder, a large proportion of patients with bipolar disorder (BD) are misdiagnosed as having unipolar depression (UD). Generally, long-term tracking is required for patients with BD to conduct an appropriate diagnosis by using traditional diagnosis tools. A one-time diagnosis system for facilitating diagnosis procedures is thus highly desirable. Accordingly, in this study, the facial expressions of patients with BD, patients with UD, and healthy controls elicited by emotional video clips were used for conducting mood disorder classification; the classification was performed by exploring the temporal fluctuation characteristics among the three groups. First, macroscopic facial expressions characterized by action units (AUs) were applied for describing the temporal transformation of muscles. Modulation spectrum analysis was applied to extract short-term intensity variations in the AUs. An interval-based multilayer perceptron (MLP) neural network was then used to classify mood disorder on the basis of the detected AU intensities. Moreover, motion vectors (MVs) were employed to describe subtle changes in facial expressions in the microscopic view. Eight basic orientations of MV change were considered for representing microfluctuation. Wavelet decomposition was then applied to extract entropy and energy features in different frequency bands. A long short-term memory model was finally used to model long-term variations for conducting mood disorder classification. A decision-level fusion approach was conducted on the combined results of macroscopic and microscopic facial expressions. For evaluating the described methods, the facial expressions elicited from the 36 subjects (12 from each of the BD, UD, and control groups) were used in 12-fold cross-validation experiments. Approaches for macroscopic and microscopic expressions achieved classification accuracies of 63.9 and 66.7 percent, respectively, and the accuracy of the fusion approach reached 72.2 percent. The results indicate that macroscopic and microscopic view descriptors are complementary to each other and helpful for conducting mood disorder classification.

This paper aims to improve speaker embedding representation based on x-vector for extracting more detailed information for speaker verification. We propose a statistics pooling time delay neural network (TDNN), in which the TDNN structure integrates statistics pooling for each layer, to consider the variation of temporal context in frame-level transformation. The proposed feature vector, named as statsvector, are compared with the baseline x-vector features on the VoxCeleb dataset and the Speakers in the Wild (SITW) dataset for speaker verification. The experimental results showed that the proposed stats-vector with score fusion achieved the best performance on VoxCeleb1 dataset. Furthermore, considering the interference from other speakers in the recordings, we found that the proposed statsvector efficiently reduced the interference and improved the speaker verification performance on the SITW dataset.

In this study, deep embedding of acoustic and articulatory features are combined for speaker identification. First, a convolutional neural network (CNN)-based universal background model (UBM) is constructed to generate acoustic feature (AC) embedding. In addition, as the articulatory features (AFs) represent some important phonological properties during speech production, a multilayer perceptron (MLP)-based AF embedding extraction model is also constructed for AF embedding extraction. The extracted AC and AF embeddings are concatenated as a combined feature vector for speaker identification using a fully-connected neural network. This proposed system was evaluated by three corpora consisting of King-ASR, LibriSpeech and SITW, and the experiments were conducted according to the properties of the datasets. We adopted all three corpora to evaluate the effect of AF embedding, and the results showed that combining AF embedding into the input feature vector improved the performance of speaker identification. The LibriSpeech corpus was used to evaluate the effect of the number of enrolled speakers. The proposed system achieved an EER of 7.80% outperforming the method based on x-vector with PLDA (8.25%). And we further evaluated the effect of signal mismatch using the SITW corpus. The proposed system achieved an EER of 25.19%, which outperformed the other baseline methods.

Speech emotion recognition is becoming increasingly important for many applications. In real-life communication, non-verbal sounds within an utterance also play an important role for people to recognize emotion. In current studies, only few emotion recognition systems considered nonverbal sounds, such as laughter, cries or other emotion interjection, which naturally exists in our daily conversation. In this work, both verbal and nonverbal sounds within an utterance were thus considered for emotion recognition of real-life conversations. Firstly, an SVM-based verbal/nonverbal sound detector was developed. A Prosodic Phrase (PPh) auto-tagger was further employed to extract the verbal/nonverbal segments. For each segment, the emotion and sound features were respectively extracted based on convolutional neural networks (CNNs) and then concatenated to form a CNN-based generic feature vector. Finally, a sequence of CNN-based feature vectors for an entire dialog turn was fed to an attentive long short-term memory (LSTM)-based sequence-to-sequence model to output an emotional sequence as recognition result. Experimental results on the recognition of seven emotional states in the NNIME (The NTHU-NTUA Chinese interactive multimodal emotion corpus) showed that the proposed method achieved a detection accuracy of 52.00% outperforming the traditional methods.