Automated speaker identification (SID) is a crucial step for the personalization of a wide range of speech-enabled services. Typical SID systems use a symmetric enrollment-verification framework with a single model to derive embeddings both offline for voice profiles extracted from enrollment utterances, and online from runtime utterances. Due to the distinct circumstances of enrollment and runtime, such as different computation and latency constraints, several applications would benefit from an asymmetric enrollment-verification framework that uses different models for enrollment and runtime embedding generation. To support this asymmetric SID where each of the two models can be updated independently, we propose using a lightweight neural network to map the embeddings from the two independent models to a shared speaker embedding space. Our results show that this approach significantly outperforms cosine scoring in a shared speaker logit space for models that were trained with a contrastive loss on large datasets with many speaker identities. This proposed Neural Embedding Speaker Space Alignment (NESSA) combined with an asymmetric update of only one of the models delivers at least 60% of the performance gain achieved by updating both models in the standard symmetric SID approach.

This paper contains a post-challenge performance analysis on cross-lingual speaker verification of the IDLab submission to the VoxCeleb Speaker Recognition Challenge 2021 (VoxSRC-21). We show that current speaker embedding extractors consistently underestimate speaker similarity in within-speaker cross-lingual trials. Consequently, the typical training and scoring protocols do not put enough emphasis on the compensation of intra-speaker language variability. We propose two techniques to increase cross-lingual speaker verification robustness. First, we enhance our previously proposed Large-Margin Fine-Tuning (LM-FT) training stage with a mini-batch sampling strategy which increases the amount of intra-speaker cross-lingual samples within the mini-batch. Second, we incorporate language information in the logistic regression calibration stage. We integrate quality metrics based on soft and hard decisions of a VoxLingua107 language identification model. The proposed techniques result in a 11.7% relative improvement over the baseline model on the VoxSRC-21 test set and contributed to our third place finish in the corresponding challenge.

In this paper we propose and analyse a large margin fine-tuning strategy and a quality-aware score calibration in text-independent speaker verification. Large margin fine-tuning is a secondary training stage for DNN based speaker verification systems trained with margin-based loss functions. It enables the network to create more robust speaker embeddings by enabling the use of longer training utterances in combination with a more aggressive margin penalty. Score calibration is a common practice in speaker verification systems to map output scores to well-calibrated log-likelihood-ratios, which can be converted to interpretable probabilities. By including quality features in the calibration system, the decision thresholds of the evaluation metrics become quality-dependent and more consistent across varying trial conditions. Applying both enhancements on the ECAPA-TDNN architecture leads to state-of-the-art results on all publicly available VoxCeleb1 test sets and contributed to our winning submissions in the supervised verification tracks of the Vox-Celeb Speaker Recognition Challenge 2020.

Learning robust speaker embeddings is a crucial step in speaker diarization. Deep neural networks can accurately capture speaker discriminative characteristics and popular deep embeddings such as x-vectors are nowadays a fundamental component of modern diarization systems. Recently, some improvements over the standard TDNN architecture used for x-vectors have been proposed. The ECAPA-TDNN model, for instance, has shown impressive performance in the speaker verification domain, thanks to a carefully designed neural model. In this work, we extend, for the first time, the use of the ECAPA-TDNN model to speaker diarization. Moreover, we improved its robustness with a powerful augmentation scheme that concatenates several contaminated versions of the same signal within the same training batch. The ECAPA-TDNN model turned out to provide robust speaker embeddings under both close-talking and distant-talking conditions. Our results on the popular AMI meeting corpus show that our system significantly outperforms recently proposed approaches.

This paper describes the IDLab submission for the text-independent task of the Short-duration Speaker Verification Challenge 2021 (SdSVC-21). This speaker verification competition focuses on short duration test recordings and cross-lingual trials, along with the constraint of limited availability of in-domain DeepMine Farsi training data. Currently, both Time Delay Neural Networks (TDNNs) and ResNets achieve state-of-the-art results in speaker verification. These architectures are structurally very different and the construction of hybrid networks looks a promising way forward. We introduce a 2D convolutional stem in a strong ECAPA-TDNN baseline to transfer some of the strong characteristics of a ResNet based model to this hybrid CNN-TDNN architecture. Similarly, we incorporate absolute frequency positional encodings in an SE-ResNet34 architecture. These learnable feature map biases along the frequency axis offer this architecture a straightforward way to exploit frequency positional information. We also propose a frequency-wise variant of Squeeze-Excitation (SE) which better preserves frequency-specific information when rescaling the feature maps. Both modified architectures significantly outperform their corresponding baseline on the SdSVC-21 evaluation data and the original VoxCeleb1 test set. A four system fusion containing the two improved architectures achieved a third place in the final SdSVC-21 Task 2 ranking.

Current speaker verification techniques rely on a neural network to extract speaker representations. The successful x-vector architecture is a Time Delay Neural Network (TDNN) that applies statistics pooling to project variable-length utterances into fixed-length speaker characterizing embeddings. In this paper, we propose multiple enhancements to this architecture based on recent trends in the related fields of face verification and computer vision. Firstly, the initial frame layers can be restructured into 1-dimensional Res2Net modules with impactful skip connections. Similarly to SE-ResNet, we introduce Squeeze-and-Excitation blocks in these modules to explicitly model channel interdependencies. The SE block expands the temporal context of the frame layer by rescaling the channels according to global properties of the recording. Secondly, neural networks are known to learn hierarchical features, with each layer operating on a different level of complexity. To leverage this complementary information, we aggregate and propagate features of different hierarchical levels. Finally, we improve the statistics pooling module with channel-dependent frame attention. This enables the network to focus on different subsets of frames during each of the channel's statistics estimation. The proposed ECAPA-TDNN architecture significantly outperforms state-of-the-art TDNN based systems on the VoxCeleb test sets and the 2019 VoxCeleb Speaker Recognition Challenge.

In this paper we describe the top-scoring IDLab submission for the text-independent task of the Short-duration Speaker Verification (SdSV) Challenge 2020. The main difficulty of the challenge exists in the large degree of varying phonetic overlap between the potentially cross-lingual trials, along with the limited availability of in-domain DeepMine Farsi training data. We introduce domain-balanced hard prototype mining to fine-tune the state-of-the-art ECAPA-TDNN x-vector based speaker embedding extractor. The sample mining technique efficiently exploits speaker distances between the speaker prototypes of the popular AAM-softmax loss function to construct challenging training batches that are balanced on the domain-level. To enhance the scoring of cross-lingual trials, we propose a language-dependent s-norm score normalization. The imposter cohort only contains data from the Farsi target-domain which simulates the enrollment data always being Farsi. In case a Gaussian-Backend language model detects the test speaker embedding to contain English, a cross-language compensation offset determined on the AAM-softmax speaker prototypes is subtracted from the maximum expected imposter mean score. A fusion of five systems with minor topological tweaks resulted in a final MinDCF and EER of 0.065 and 1.45% respectively on the SdSVC evaluation set.

Conventional speech emotion recognition based on the extraction of high level descriptors emerging from low level descriptors seldom delivers promising results in cross-corpus experiments.Therefore it might not perform well in real-life applications.Factor analysis, proven in the fields of language identification and speaker verification, could clear a path towards more robust emotion recognition.This paper proposes an iVector-based approach operating on acoustic MFCC features with a separate modeling of the speaker and emotion variabilities respectively.The speech analysis extracts two fixedlength low-dimensional feature vectors corresponding to the two mentioned sources of variation.To model the speakerrelated nuisance variability speaker factors are extracted using an eigenvoice matrix.After compensating for this speaker variability in the supervector space, the emotion factors (one per targeted emotion) are extracted using an emotion variability matrix.The emotion factors are then fed to a basic emotion classifier.Leave-one-speaker-out cross-validation on the Berlin Database of Emotional Speech EMO-DB (German) and IEMO-CAP (English) datasets lead to results that are competitive with the current state-of-the-art.Cross-lingual experiments demonstrate the excellent robustness of the method: the classification accuracies degrade less than 15% relative when emotion models are trained on one corpus and tested on the other.

Speaker diarization includes two steps: speaker segmentation and speaker clustering. Speaker segmentation searches for speaker boundaries, whereas speaker clustering aims at grouping speech segments of the same speaker. In this work, the segmentation is improved by replacing the Bayesian Information Criterion (BIC) with a new iVector-based approach. Unlike BIC-based methods which trigger on any acoustic dissimilarities, the proposed method suppresses phonetic variations and accentuates speaker differences. More specifically our method generates boundaries based on the distance between two speaker factor vectors that are extracted on a frame-by frame basis. The extraction relies on an eigenvoice matrix so that large differences between speaker factor vectors indicate a different speaker. A Mahalanobis-based distance measure, in which the covariance matrix compensates for the remaining and detrimental phonetic variability, is shown to generate accurate boundaries. The detected segments are clustered by a state-of-the-art iVector Probabilistic Linear Discriminant Analysis system. Experiments on the COST278 multilingual broadcast news database show relative reductions of 50% in boundary detection errors. The speaker error rate is reduced by 8% relative.

This paper presents a technique to adapt an acoustically based language classifier to the background conditions and speaker accents. This adaptation improves language classification on a broad spectrum of TV broadcasts. The core of the system consists of an iVector-based setup in which language and channel variabilities are modeled separately. The subsequent language classifier (the backend) operates on the language factors, i.e. those features in the extracted iVectors that explain the observed language variability. The proposed technique adapts the language variability model to the background conditions and to the speaker accents present in the audio. The effect of the adaptation is evaluated on a 28 hours corpus composed of documentaries and monolingual as well as multilingual broadcast news shows. Consistent improvements in the automatic identification of Flemish (Belgian Dutch), English and French are demonstrated for all broadcast types.