Ha-Jin Yu

Hyun-seo Shin, Jungwoo Heo, Ju-ho Kim et al.

Audio deepfake detection (ADD) is the task of detecting spoofing attacks generated by text-to-speech or voice conversion systems. Spoofing evidence, which helps to distinguish between spoofed and bona-fide utterances, might exist either locally or globally in the input features. To capture these, the Conformer, which consists of Transformers and CNN, possesses a suitable structure. However, since the Conformer was designed for sequence-to-sequence tasks, its direct application to ADD tasks may be sub-optimal. To tackle this limitation, we propose HM-Conformer by adopting two components: (1) Hierarchical pooling method progressively reducing the sequence length to eliminate duplicated information (2) Multi-level classification token aggregation method utilizing classification tokens to gather information from different blocks. Owing to these components, HM-Conformer can efficiently detect spoofing evidence by processing various sequence lengths and aggregating them. In experimental results on the ASVspoof 2021 Deepfake dataset, HM-Conformer achieved a 15.71% EER, showing competitive performance compared to recent systems.

KiHyun Nam, Jungwoo Heo, Siu Bae et al.

As audio-first agents become increasingly common in physical AI, conversational robots, and screenless wearables, audio large language models (audio-LLMs) must integrate speaker-specific understanding to support user authorization, personalization, and context-aware interaction. This requires modeling who is speaking, how the voice sounds, and how recording conditions affect speaker cues. Conventional speaker verification systems provide strong scalar scores but little linguistic evidence, while current audio-LLMs and speaker-aware language models have limited ability to organize speaker information beyond binary labels or descriptive profiles. We present SpeakerLLM, a speaker-specialized audio-LLM framework that unifies single-utterance speaker profiling, recording-condition understanding, utterance-pair speaker comparison, and evidence-organized verification reasoning within a natural-language interface. We construct verification-reasoning targets and a decision-composition policy that separate profile-level evidence from the final same-or-different decision and organize recording condition, profile evidence, and the decision into a structured trace. At its core, SpeakerLLM uses a hierarchical speaker tokenizer designed to capture multiple granularities of speaker evidence. Utterance-level speaker embeddings summarize identity and profile-level cues, whereas frame-level speaker features preserve fine-grained acoustic descriptors. Experiments show that SpeakerLLM-Base improves speaker-profile and recording-condition understanding over general audio-LLMs, while SpeakerLLM-VR preserves strong generated-verdict accuracy and produces decision traces grounded in the supervised verification reasoning schema. We will release the metadata-enriched supervision dataset and target-construction code for reproducibility.

KiHyun Nam, Jungwoo Heo, Jee-weon Jung et al.

A primary challenge when deploying speaker recognition systems in real-world applications is performance degradation caused by environmental mismatch. We propose a diffusion-based method that takes speaker embeddings extracted from a pre-trained speaker recognition model and generates refined embeddings. For training, our approach progressively adds Gaussian noise to both clean and noisy speaker embeddings extracted from clean and noisy speech, respectively, via forward process of a diffusion model, and then reconstructs them to clean embeddings in the reverse process. While inferencing, all embeddings are regenerated via diffusion process. Our method needs neither speaker label nor any modification to the existing speaker recognition pipeline. Experiments on evaluation sets simulating environment mismatch scenarios show that our method can improve recognition accuracy by up to 19.6% over baseline models while retaining performance on conventional scenarios. We publish our code here https://github.com/kaistmm/seed-pytorch

Young Joo Han, Ha-Jin Yu

Modeling and synthesizing real sRGB noise is crucial for various low-level vision tasks, such as building datasets for training image denoising systems. The distribution of real sRGB noise is highly complex and affected by a multitude of factors, making its accurate modeling extremely challenging. Therefore, recent studies have proposed methods that employ data-driven generative models, such as Generative Adversarial Networks (GAN) and Normalizing Flows. These studies achieve more accurate modeling of sRGB noise compared to traditional noise modeling methods. However, there are performance limitations due to the inherent characteristics of each generative model. To address this issue, we propose NM-FlowGAN, a hybrid approach that exploits the strengths of both GAN and Normalizing Flows. We combine pixel-wise noise modeling networks based on Normalizing Flows and spatial correlation modeling networks based on GAN. Specifically, the pixel-wise noise modeling network leverages the high training stability of Normalizing Flows to capture noise characteristics that are affected by a multitude of factors, and the spatial correlation networks efficiently model pixel-to-pixel relationships. In particular, unlike recent methods that rely on paired noisy images, our method synthesizes noise using clean images and factors that affect noise characteristics, such as easily obtainable parameters like camera type and ISO settings, making it applicable to various fields where obtaining noisy-clean image pairs is not feasible. In our experiments, our NM-FlowGAN outperforms other baselines in the sRGB noise synthesis task. Moreover, the denoising neural network trained with synthesized image pairs from our model shows superior performance compared to other baselines. Our code is available at: \url{https://github.com/YoungJooHan/NM-FlowGAN}.

Seung-bin Kim, Chan-yeong Lim, Jungwoo Heo et al.

In speaker verification systems, the utilization of short utterances presents a persistent challenge, leading to performance degradation primarily due to insufficient phonetic information to characterize the speakers. To overcome this obstacle, we propose a novel structure, MR-RawNet, designed to enhance the robustness of speaker verification systems against variable duration utterances using raw waveforms. The MR-RawNet extracts time-frequency representations from raw waveforms via a multi-resolution feature extractor that optimally adjusts both temporal and spectral resolutions simultaneously. Furthermore, we apply a multi-resolution attention block that focuses on diverse and extensive temporal contexts, ensuring robustness against changes in utterance length. The experimental results, conducted on VoxCeleb1 dataset, demonstrate that the MR-RawNet exhibits superior performance in handling utterances of variable duration compared to other raw waveform-based systems.

Young-Joo Han, Ha-Jin Yu

Recently, numerous studies have been conducted on supervised learning-based image denoising methods. However, these methods rely on large-scale noisy-clean image pairs, which are difficult to obtain in practice. Denoising methods with self-supervised training that can be trained with only noisy images have been proposed to address the limitation. These methods are based on the convolutional neural network (CNN) and have shown promising performance. However, CNN-based methods do not consider using nonlocal self-similarities essential in the traditional method, which can cause performance limitations. This paper presents self-similarity attention (SS-Attention), a novel self-attention module that can capture nonlocal self-similarities to solve the problem. We focus on designing a lightweight self-attention module in a pixel-wise manner, which is nearly impossible to implement using the classic self-attention module due to the quadratically increasing complexity with spatial resolution. Furthermore, we integrate SS-Attention into the blind-spot network called self-similarity-based blind-spot network (SS-BSN). We conduct the experiments on real-world image denoising tasks. The proposed method quantitatively and qualitatively outperforms state-of-the-art methods in self-supervised denoising on the Smartphone Image Denoising Dataset (SIDD) and Darmstadt Noise Dataset (DND) benchmark datasets.

Jee-weon Jung, Hemlata Tak, Hye-jin Shim et al.

ASV (automatic speaker verification) systems are intrinsically required to reject both non-target (e.g., voice uttered by different speaker) and spoofed (e.g., synthesised or converted) inputs. However, there is little consideration for how ASV systems themselves should be adapted when they are expected to encounter spoofing attacks, nor when they operate in tandem with CMs (spoofing countermeasures), much less how both systems should be jointly optimised. The goal of the first SASV (spoofing-aware speaker verification) challenge, a special sesscion in ISCA INTERSPEECH 2022, is to promote development of integrated systems that can perform ASV and CM simultaneously.

Hye-jin Shim, Jungwoo Heo, Jae-han Park et al.

The objective of this paper is to combine multiple frame-level features into a single utterance-level representation considering pairwise relationship. For this purpose, we propose a novel graph attentive feature aggregation module by interpreting each frame-level feature as a node of a graph. The inter-relationship between all possible pairs of features, typically exploited indirectly, can be directly modeled using a graph. The module comprises a graph attention layer and a graph pooling layer followed by a readout operation. The graph attention layer first models the non-Euclidean data manifold between different nodes. Then, the graph pooling layer discards less informative nodes considering the significance of the nodes. Finally, the readout operation combines the remaining nodes into a single representation. We employ two recent systems, SE-ResNet and RawNet2, with different input features and architectures and demonstrate that the proposed feature aggregation module consistently shows a relative improvement over 10%, compared to the baseline.

Jee-weon Jung, Hee-Soo Heo, Hemlata Tak et al.

Artefacts that differentiate spoofed from bona-fide utterances can reside in spectral or temporal domains. Their reliable detection usually depends upon computationally demanding ensemble systems where each subsystem is tuned to some specific artefacts. We seek to develop an efficient, single system that can detect a broad range of different spoofing attacks without score-level ensembles. We propose a novel heterogeneous stacking graph attention layer which models artefacts spanning heterogeneous temporal and spectral domains with a heterogeneous attention mechanism and a stack node. With a new max graph operation that involves a competitive mechanism and an extended readout scheme, our approach, named AASIST, outperforms the current state-of-the-art by 20% relative. Even a lightweight variant, AASIST-L, with only 85K parameters, outperforms all competing systems.

Hye-jin Shim, Jee-weon Jung, Ju-ho Kim et al.

Various attention mechanisms are being widely applied to acoustic scene classification. However, we empirically found that the attention mechanism can excessively discard potentially valuable information, despite improving performance. We propose the attentive max feature map that combines two effective techniques, attention and a max feature map, to further elaborate the attention mechanism and mitigate the above-mentioned phenomenon. We also explore various joint training methods, including multi-task learning, that allocate additional abstract labels for each audio recording. Our proposed system demonstrates state-of-the-art performance for single systems on Subtask A of the DCASE 2020 challenge by applying the two proposed techniques using relatively fewer parameters. Furthermore, adopting the proposed attentive max feature map, our team placed fourth in the recent DCASE 2021 challenge.

Ju-ho Kim, Hye-jin Shim, Jee-weon Jung et al.

Most speaker verification tasks are studied as an open-set evaluation scenario considering the real-world condition. Thus, the generalization power to unseen speakers is of paramount important to the performance of the speaker verification system. We propose to apply \textit {Mean Teacher}, a temporal averaging model, to extract speaker embeddings with small intra-class variance and large inter-class variance. The mean teacher network refers to the temporal averaging of deep neural network parameters; it can produces more accurate and stable representations than using weights after the training finished. By learning the reliable intermediate representation of the mean teacher network, we expect that the proposed method can explore more discriminatory embedding spaces and improve the generalization performance of the speaker verification system. Experimental results on the VoxCeleb1 test set demonstrate that the proposed method relatively improves performance by 11.61\%, compared to a baseline system.

Jee-weon Jung, Hee-Soo Heo, Ha-Jin Yu et al.

This work presents a novel back-end framework for speaker verification using graph attention networks. Segment-wise speaker embeddings extracted from multiple crops within an utterance are interpreted as node representations of a graph. The proposed framework inputs segment-wise speaker embeddings from an enrollment and a test utterance and directly outputs a similarity score. We first construct a graph using segment-wise speaker embeddings and then input these to graph attention networks. After a few graph attention layers with residual connections, each node is projected into a one-dimensional space using affine transform, followed by a readout operation resulting in a scalar similarity score. To enable successful adaptation for speaker verification, we propose techniques such as separating trainable weights for attention map calculations between segment-wise speaker embeddings from different utterances. The effectiveness of the proposed framework is validated using three different speaker embedding extractors trained with different architectures and objective functions. Experimental results demonstrate consistent improvement over various baseline back-end classifiers, with an average equal error rate improvement of 20% over the cosine similarity back-end without test time augmentation.

Hye-jin Shim, Jee-weon Jung, Ju-ho Kim et al.

Frequently misclassified pairs of classes that share many common acoustic properties exist in acoustic scene classification (ASC). To distinguish such pairs of classes, trivial details scattered throughout the data could be vital clues. However, these details are less noticeable and are easily removed using conventional non-linear activations (e.g. ReLU). Furthermore, making design choices to emphasize trivial details can easily lead to overfitting if the system is not sufficiently generalized. In this study, based on the analysis of the ASC task's characteristics, we investigate various methods to capture discriminative information and simultaneously mitigate the overfitting problem. We adopt a max feature map method to replace conventional non-linear activations in a deep neural network, and therefore, we apply an element-wise comparison between different filters of a convolution layer's output. Two data augment methods and two deep architecture modules are further explored to reduce overfitting and sustain the system's discriminative power. Various experiments are conducted using the detection and classification of acoustic scenes and events 2020 task1-a dataset to validate the proposed methods. Our results show that the proposed system consistently outperforms the baseline, where the single best performing system has an accuracy of 70.4% compared to 65.1% of the baseline.

Hye-jin Shim, Jee-weon Jung, Ju-ho Kim et al.

A number of studies have successfully developed speaker verification or presentation attack detection systems. However, studies integrating the two tasks remain in the preliminary stages. In this paper, we propose two approaches for building an integrated system of speaker verification and presentation attack detection: an end-to-end monolithic approach and a back-end modular approach. The first approach simultaneously trains speaker identification, presentation attack detection, and the integrated system using multi-task learning using a common feature. However, through experiments, we hypothesize that the information required for performing speaker verification and presentation attack detection might differ because speaker verification systems try to remove device-specific information from speaker embeddings, while presentation attack detection systems exploit such information. Therefore, we propose a back-end modular approach using a separate deep neural network (DNN) for speaker verification and presentation attack detection. This approach has thee input components: two speaker embeddings (for enrollment and test each) and prediction of presentation attacks. Experiments are conducted using the ASVspoof 2017-v2 dataset, which includes official trials on the integration of speaker verification and presentation attack detection. The proposed back-end approach demonstrates a relative improvement of 21.77% in terms of the equal error rate for integrated trials compared to a conventional speaker verification system.

Seung-bin Kim, Jee-weon Jung, Hye-jin Shim et al.

Most studies on speaker verification systems focus on long-duration utterances, which are composed of sufficient phonetic information. However, the performances of these systems are known to degrade when short-duration utterances are inputted due to the lack of phonetic information as compared to the long utterances. In this paper, we propose a method that compensates for the performance degradation of speaker verification for short utterances, referred to as "segment aggregation". The proposed method adopts an ensemble-based design to improve the stability and accuracy of speaker verification systems. The proposed method segments an input utterance into several short utterances and then aggregates the segment embeddings extracted from the segmented inputs to compose a speaker embedding. Then, this method simultaneously trains the segment embeddings and the aggregated speaker embedding. In addition, we also modified the teacher-student learning method for the proposed method. Experimental results on different input duration using the VoxCeleb1 test set demonstrate that the proposed technique improves speaker verification performance by about 45.37% relatively compared to the baseline system with 1-second test utterance condition.

Jee-weon Jung, Seung-bin Kim, Hye-jin Shim et al.

Recent advances in deep learning have facilitated the design of speaker verification systems that directly input raw waveforms. For example, RawNet extracts speaker embeddings from raw waveforms, which simplifies the process pipeline and demonstrates competitive performance. In this study, we improve RawNet by scaling feature maps using various methods. The proposed mechanism utilizes a scale vector that adopts a sigmoid non-linear function. It refers to a vector with dimensionality equal to the number of filters in a given feature map. Using a scale vector, we propose to scale the feature map multiplicatively, additively, or both. In addition, we investigate replacing the first convolution layer with the sinc-convolution layer of SincNet. Experiments performed on the VoxCeleb1 evaluation dataset demonstrate the effectiveness of the proposed methods, and the best performing system reduces the equal error rate by half compared to the original RawNet. Expanded evaluation results obtained using the VoxCeleb1-E and VoxCeleb-H protocols marginally outperform existing state-of-the-art systems.

Jee-weon Jung, Hye-jin Shim, Hee-Soo Heo et al.

In this study, we analyze the role of various categories of subsidiary information in conducting replay attack spoofing detection: `Room Size', `Reverberation', `Speaker-to-ASV distance, `Attacker-to-Speaker distance', and `Replay Device Quality'. As a means of analyzing subsidiary information, we use two frameworks to either subtract or include a category of subsidiary information to the code extracted from a deep neural network. For subtraction, we utilize an adversarial process framework which makes the code orthogonal to the basis vectors of the subsidiary information. For addition, we utilize the multi-task learning framework to include subsidiary information to the code. All experiments are conducted using the ASVspoof 2019 physical access scenario with the provided meta data. Through the analysis of the result of the two approaches, we conclude that various categories of subsidiary information does not reside enough in the code when the deep neural network is trained for binary classification. Explicitly including various categories of subsidiary information through the multi-task learning framework can help improve performance in closed set condition.

Hye-jin Shim, Hee-Soo Heo, Jee-weon Jung et al.

Constructing a dataset for replay spoofing detection requires a physical process of playing an utterance and re-recording it, presenting a challenge to the collection of large-scale datasets. In this study, we propose a self-supervised framework for pretraining acoustic configurations using datasets published for other tasks, such as speaker verification. Here, acoustic configurations refer to the environmental factors generated during the process of voice recording but not the voice itself, including microphone types, place and ambient noise levels. Specifically, we select pairs of segments from utterances and train deep neural networks to determine whether the acoustic configurations of the two segments are identical. We validate the effectiveness of the proposed method based on the ASVspoof 2019 physical access dataset utilizing two well-performing systems. The experimental results demonstrate that the proposed method outperforms the baseline approach by 30%.

Hee-Soo Heo, Jee-weon Jung, Hye-jin Shim et al.

An adversarial process between two deep neural networks is a promising approach to train a robust model. In this paper, we propose an adversarial process using cosine similarity, whereas conventional adversarial processes are based on inverted categorical cross entropy (CCE). When used for training an identification model, the adversarial process induces the competition of two discriminative models; one for a primary task such as speaker identification or image recognition, the other one for a subsidiary task such as channel identification or domain identification. In particular, the adversarial process degrades the performance of the subsidiary model by eliminating the subsidiary information in the input which, in assumption, may degrade the performance of the primary model. The conventional adversarial processes maximize the CCE of the subsidiary model to degrade the performance. We have studied a framework for training robust discriminative models by eliminating channel or domain information (subsidiary information) by applying such an adversarial process. However, we found through experiments that using the process of maximizing the CCE does not guarantee the performance degradation of the subsidiary model. In the proposed adversarial process using cosine similarity, on the contrary, the performance of the subsidiary model can be degraded more efficiently by searching feature space orthogonal to the subsidiary model. The experiments on speaker identification and image recognition show that we found features that make the outputs of the subsidiary models independent of the input, and the performances of the primary models are improved.

Hee-Soo Heo, Jee-weon Jung, Hye-jin Shim et al.

Acoustic scene classification identifies an input segment into one of the pre-defined classes using spectral information. The spectral information of acoustic scenes may not be mutually exclusive due to common acoustic properties across different classes, such as babble noises included in both airports and shopping malls. However, conventional training procedure based on one-hot labels does not consider the similarities between different acoustic scenes. We exploit teacher-student learning with the purpose to derive soft-labels that consider common acoustic properties among different acoustic scenes. In teacher-student learning, the teacher network produces soft-labels, based on which the student network is trained. We investigate various methods to extract soft-labels that better represent similarities across different scenes. Such attempts include extracting soft-labels from multiple audio segments that are defined as an identical acoustic scene. Experimental results demonstrate the potential of our approach, showing a classification accuracy of 77.36 % on the DCASE 2018 task 1 validation set.

Jee-weon Jung, Hye-jin Shim, Hee-Soo Heo et al.

In this study, we concentrate on replacing the process of extracting hand-crafted acoustic feature with end-to-end DNN using complementary high-resolution spectrograms. As a result of advance in audio devices, typical characteristics of a replayed speech based on conventional knowledge alter or diminish in unknown replay configurations. Thus, it has become increasingly difficult to detect spoofed speech with a conventional knowledge-based approach. To detect unrevealed characteristics that reside in a replayed speech, we directly input spectrograms into an end-to-end DNN without knowledge-based intervention. Explorations dealt in this study that differentiates from existing spectrogram-based systems are twofold: complementary information and high-resolution. Spectrograms with different information are explored, and it is shown that additional information such as the phase information can be complementary. High-resolution spectrograms are employed with the assumption that the difference between a bona-fide and a replayed speech exists in the details. Additionally, to verify whether other features are complementary to spectrograms, we also examine raw waveform and an i-vector based system. Experiments conducted on the ASVspoof 2019 physical access challenge show promising results, where t-DCF and equal error rates are 0.0570 and 2.45 % for the evaluation set, respectively.

Jee-weon Jung, Hee-Soo Heo, Ju-ho Kim et al.

Recently, direct modeling of raw waveforms using deep neural networks has been widely studied for a number of tasks in audio domains. In speaker verification, however, utilization of raw waveforms is in its preliminary phase, requiring further investigation. In this study, we explore end-to-end deep neural networks that input raw waveforms to improve various aspects: front-end speaker embedding extraction including model architecture, pre-training scheme, additional objective functions, and back-end classification. Adjustment of model architecture using a pre-training scheme can extract speaker embeddings, giving a significant improvement in performance. Additional objective functions simplify the process of extracting speaker embeddings by merging conventional two-phase processes: extracting utterance-level features such as i-vectors or x-vectors and the feature enhancement phase, e.g., linear discriminant analysis. Effective back-end classification models that suit the proposed speaker embedding are also explored. We propose an end-to-end system that comprises two deep neural networks, one front-end for utterance-level speaker embedding extraction and the other for back-end classification. Experiments conducted on the VoxCeleb1 dataset demonstrate that the proposed model achieves state-of-the-art performance among systems without data augmentation. The proposed system is also comparable to the state-of-the-art x-vector system that adopts data augmentation.

Hee-Soo Heo, Jee-weon Jung, IL-Ho Yang et al.

In recent years, speaker verification has primarily performed using deep neural networks that are trained to output embeddings from input features such as spectrograms or Mel-filterbank energies. Studies that design various loss functions, including metric learning have been widely explored. In this study, we propose two end-to-end loss functions for speaker verification using the concept of speaker bases, which are trainable parameters. One loss function is designed to further increase the inter-speaker variation, and the other is designed to conduct the identical concept with hard negative mining. Each speaker basis is designed to represent the corresponding speaker in the process of training deep neural networks. In contrast to the conventional loss functions that can consider only a limited number of speakers included in a mini-batch, the proposed loss functions can consider all the speakers in the training set regardless of the mini-batch composition. In particular, the proposed loss functions enable hard negative mining and calculations of between-speaker variations with consideration of all speakers. Through experiments on VoxCeleb1 and VoxCeleb2 datasets, we confirmed that the proposed loss functions could supplement conventional softmax and center loss functions.

Jee-weon Jung, Hee-soo Heo, Hye-jin Shim et al.

The short duration of an input utterance is one of the most critical threats that degrade the performance of speaker verification systems. This study aimed to develop an integrated text-independent speaker verification system that inputs utterances with short duration of 2 seconds or less. We propose an approach using a teacher-student learning framework for this goal, applied to short utterance compensation for the first time in our knowledge. The core concept of the proposed system is to conduct the compensation throughout the network that extracts the speaker embedding, mainly in phonetic-level, rather than compensating via a separate system after extracting the speaker embedding. In the proposed architecture, phonetic-level features where each feature represents a segment of 130 ms are extracted using convolutional layers. A layer of gated recurrent units extracts an utterance-level feature using phonetic-level features. The proposed approach also adopts a new objective function for teacher-student learning that considers both Kullback-Leibler divergence of output layers and cosine distance of speaker embeddings layers. Experiments were conducted using deep neural networks that take raw waveforms as input, and output speaker embeddings on VoxCeleb1 dataset. The proposed model could compensate approximately 65 \% of the performance degradation due to the shortened duration.

Hye-Jin Shim, Jee-weon Jung, Hee-Soo Heo et al.

In this paper, we propose a replay attack spoofing detection system for automatic speaker verification using multitask learning of noise classes. We define the noise that is caused by the replay attack as replay noise. We explore the effectiveness of training a deep neural network simultaneously for replay attack spoofing detection and replay noise classification. The multi-task learning includes classifying the noise of playback devices, recording environments, and recording devices as well as the spoofing detection. Each of the three types of the noise classes also includes a genuine class. The experiment results on the ASVspoof2017 datasets demonstrate that the performance of our proposed system is improved by 30% relatively on the evaluation set.