Audio deepfake model attribution aims to mitigate the misuse of synthetic speech by identifying the source model responsible for generating a given audio sample, enabling accountability and informing vendors. The task is challenging, but self-supervised learning (SSL)-derived acoustic features have demonstrated state-of-the-art attribution capabilities, yet the underlying factors driving their success and the limits of their discriminative power remain unclear. In this paper, we systematically investigate how SSL-derived features capture architectural signatures in audio deepfakes. By controlling multiple dimensions of the audio generation process we reveal how subtle perturbations in model checkpoints, text prompts, vocoders, or speaker identity influence attribution. Our results provide new insights into the robustness, biases, and limitations of SSL-based deepfake attribution, highlighting both its strengths and vulnerabilities in realistic scenarios.

The growing prevalence of real-world deepfakes presents a critical challenge for existing detection systems, which are often evaluated on datasets collected just for scientific purposes. To address this gap, we introduce a novel dataset of real-world audio deepfakes. Our analysis reveals that these real-world examples pose significant challenges, even for the most performant detection models. Rather than increasing model complexity or exhaustively search for a better alternative, in this work we focus on a data-centric paradigm, employing strategies like dataset curation, pruning, and augmentation to improve model robustness and generalization. Through these methods, we achieve a 55% relative reduction in EER on the In-the-Wild dataset, reaching an absolute EER of 1.7%, and a 63% reduction on our newly proposed real-world deepfakes dataset, AI4T. These results highlight the transformative potential of data-centric approaches in enhancing deepfake detection for real-world applications. Code and data available at: https://github.com/davidcombei/AI4T.

Deepfakes - manipulated or forged audio and video media - pose significant security risks to individuals, organizations, and society at large. To address these challenges, machine learning-based classifiers are commonly employed to detect deepfake content. In this paper, we assess the robustness of such classifiers through a systematic penetration testing methodology, which we introduce as DeePen. Our approach operates without prior knowledge of or access to the target deepfake detection models. Instead, it leverages a set of carefully selected signal processing modifications - referred to as attacks - to evaluate model vulnerabilities. Using DeePen, we analyze both real-world production systems and publicly available academic model checkpoints, demonstrating that all tested systems exhibit weaknesses and can be reliably deceived by simple manipulations such as time-stretching or echo addition. Furthermore, our findings reveal that while some attacks can be mitigated by retraining detection systems with knowledge of the specific attack, others remain persistently effective.

We show how replay attacks undermine audio deepfake detection: By playing and re-recording deepfake audio through various speakers and microphones, we make spoofed samples appear authentic to the detection model. To study this phenomenon in more detail, we introduce ReplayDF, a dataset of recordings derived from M-AILABS and MLAAD, featuring 109 speaker-microphone combinations across six languages and four TTS models. It includes diverse acoustic conditions, some highly challenging for detection. Our analysis of six open-source detection models across five datasets reveals significant vulnerability, with the top-performing W2V2-AASIST model's Equal Error Rate (EER) surging from 4.7% to 18.2%. Even with adaptive Room Impulse Response (RIR) retraining, performance remains compromised with an 11.0% EER. We release ReplayDF for non-commercial research use.

Deepfake detection has gained significant attention across audio, text, and image modalities, with high accuracy in distinguishing real from fake. However, identifying the exact source-such as the system or model behind a deep-fake-remains a less studied problem. In this paper, we take a significant step forward in audio deepfake model attribution or source tracing by proposing a training-free, green AI approach based entirely on k-Nearest Neighbors (kNN). Leveraging a pre-trained self-supervised learning (SSL) model, we show that grouping samples from the same generator is straightforwardwe obtain an 0.93 F1-score across five deepfake datasets. The method also demonstrates strong out-of-domain (OOD) detection, effectively identifying samples from unseen models at an F1-score of 0.84. We further analyse these results in a multi-dimensional approach and provide additional insights. All code and data protocols used in this work are available in our open repository:https://github.com/adrianastan/tada/.

Open-source text-to-speech (TTS) frameworks have emerged as highly adaptable platforms for developing speech synthesis systems across a wide range of languages. However, their applicability is not uniform – particularly when the target language is under-resourced or when computational resources are constrained. In this study, we systematically assess the feasibility of building novel TTS models using four widely adopted open-source architectures: FastPitch, VITS, Grad-TTS, and Matcha-TTS. Our evaluation spans multiple dimensions, including qualitative aspects such as ease of installation, dataset preparation, and hardware requirements, as well as quantitative assessments of synthesis quality for Romanian. We employ both objective metrics and subjective listening tests to evaluate intelligibility, speaker similarity, and naturalness of the generated speech. The results reveal significant challenges in tool chain setup, data preprocessing, and computational efficiency, which can hinder adoption in low-resource contexts. By grounding the analysis in reproducible protocols and accessible evaluation criteria, thiswork aims to inform best practices and promote more inclusive, language-diverse TTS development. All information needed to reproduce this study (i.e. code and data) will be made available after the final review.

Audio deepfake detection has become a pivotal task over the last couple of years, as many recent speech synthesis and voice cloning systems generate highly realistic speech samples, thus enabling their use in malicious activities. In this paper we address the issue of audio deepfake detection as it was set in the ASVspoof5 challenge. First, we benchmark ten types of pretrained representations and show that the self-supervised representations stemming from the wav2vec2 and wavLM families perform best. Of the two, wavLM is better when restricting the pretraining data to LibriSpeech, as required by the challenge rules. To further improve performance, we finetune the wavLM model for the deepfake detection task. We extend the ASVspoof5 dataset with samples from other deepfake detection datasets and apply data augmentation. Our final challenge submission consists of a late fusion combination of four models and achieves an equal error rate of 6.56% and 17.08% on the two evaluation sets.

This paper presents a novel deep neural network-based architecture tailored for Speech Emotion Recognition (SER). The architecture capitalises on dense interconnections among multiple layers of bidirectional dilated convolutions. A linear kernel dynamically fuses the outputs of these layers to yield the final emotion class prediction. This innovative architecture is denoted as TBDM-Net: Temporally-Aware Bi-directional Dense Multi-Scale Network. We conduct a comprehensive performance evaluation of TBDM-Net, including an ablation study, across six widely-acknowledged SER datasets for unimodal speech emotion recognition. Additionally, we explore the influence of gender-informed emotion prediction by appending either golden or predicted gender labels to the architecture's inputs or predictions. The implementation of TBDM-Net is accessible at: https://github.com/adrianastan/tbdm-net.

Generalisation—the ability of a model to perform well on unseen data—is crucial for building reliable deepfake detectors. However, recent studies have shown that the current audio deepfake models fall short of this desideratum. In this work we investigate the potential of pretrained self-supervised representations in building general and calibrated audio deepfake detection models. We show that large frozen representations coupled with a simple logistic regression classifier are extremely effective in achieving strong generalisation capabilities: compared to the RawNet2 model, this approach reduces the equal error rate from 30.9% to 8.8% on a benchmark of eight deepfake datasets, while learning less than 2k parameters. Moreover, the proposed method produces considerably more reliable predictions compared to previous approaches making it more suitable for realistic use.

This paper focuses on adapting the functionalities of the FastPitch model to the Romanian language; extending the set of speakers from one to eighteen; synthesising speech using an anonymous identity; and replicating the identities of new, unseen speakers. During this work, the effects of various configurations and training strategies were tested and discussed, along with their advantages and weaknesses. Finally, we settled on a new configuration, built on top of the FastPitch architecture, capable of producing natural speech synthesis, for both known (identities from the training dataset) and unknown (identities learnt through short reference samples) speakers. The anonymous speaker can be used for text-to-speech synthesis, if one wants to cancel out the identity information while keeping the semantic content whole and clear. At last, we discussed possible limitations of our work, which will form the basis for future investigations and advancements.

In this paper we introduce a first attempt on understanding how a non-autoregressive factorised multi-speaker speech synthesis architecture exploits the information present in different speaker embedding sets.We analyse if jointly learning the representations, and initialising them from pretrained models determine any quality improvements for target speaker identities.In a separate analysis, we investigate how the different sets of embeddings impact the network's core speech abstraction (i.e.zero conditioned) in terms of speaker identity and representation learning.We show that, regardless of the used set of embeddings and learning strategy, the network can handle various speaker identities equally well, with barely noticeable variations in speech output quality, and that speaker leakage within the core structure of the synthesis system is inevitable in the standard training procedures adopted thus far.

Generalisation -- the ability of a model to perform well on unseen data -- is crucial for building reliable deepfake detectors. However, recent studies have shown that the current audio deepfake models fall short of this desideratum. In this work we investigate the potential of pretrained self-supervised representations in building general and calibrated audio deepfake detection models. We show that large frozen representations coupled with a simple logistic regression classifier are extremely effective in achieving strong generalisation capabilities: compared to the RawNet2 model, this approach reduces the equal error rate from 30.9% to 8.8% on a benchmark of eight deepfake datasets, while learning less than 2k parameters. Moreover, the proposed method produces considerably more reliable predictions compared to previous approaches making it more suitable for realistic use.

In this article, we introduce an extended, freely available resource for the Romanian language, named RoLEX. The dataset was developed mainly for speech processing applications, yet its applicability extends beyond this domain. RoLEX includes over 330,000 curated entries with information regarding lemma, morphosyntactic description, syllabification, lexical stress and phonemic transcription. The process of selecting the list of word entries and semi-automatically annotating the complete lexical information associated with each of the entries is thoroughly described. The dataset's inherent knowledge is then evaluated in a task of concurrent prediction of syllabification, lexical stress marking and phonemic transcription. The evaluation looked into several dataset design factors, such as the minimum viable number of entries for correct prediction, the optimisation of the minimum number of required entries through expert selection and the augmentation of the input with morphosyntactic information, as well as the influence of each task in the overall accuracy. The best results were obtained when the orthographic form of the entries was augmented with the complete morphosyntactic tags. A word error rate of 3.08% and a character error rate of 1.08% were obtained this way. We show that using a carefully selected subset of entries for training can result in a similar performance to the performance obtained by a larger set of randomly selected entries (twice as many). In terms of prediction complexity, the lexical stress marking posed most problems and accounts for around 60% of the errors in the predicted sequence.

Large speech models-derived features have recently shown increased performance over signal-based features across multiple downstream tasks, even when the networks are not fine-tuned towards the target task. In this paper we show the results of an analysis of several signal- and neural models-derived features for speech emotion recognition. We use pretrained models and explore their inherent potential abstractions of emotions. Simple classification methods are used so as to not interfere or add knowledge to the task. We show that, even without finetuning, some of these large neural speech models’ representations can enclose information that enables performances close to, and even beyond state-of-the-art results across six standard speech emotion recognition datasets.

The field of text-to-speech (TTS) technology has been rapidly advancing in recent years

The task of converting text input into video content is becoming an important topic for synthetic media generation. Several methods have been proposed with some of them reaching close-to-natural performances in constrained tasks. In this paper, we tackle a subissue of the text-to-video generation problem, by converting the text into lip landmarks. However, we do this using a modular, controllable system architecture and evaluate each of its individual components. Our system, entitled FlexLip, is split into two separate modules: text-to-speech and speech-to-lip, both having underlying controllable deep neural network architectures. This modularity enables the easy replacement of each of its components, while also ensuring the fast adaptation to new speaker identities by disentangling or projecting the input features. We show that by using as little as 20 min of data for the audio generation component, and as little as 5 min for the speech-to-lip component, the objective measures of the generated lip landmarks are comparable with those obtained when using a larger set of training samples. We also introduce a series of objective evaluation measures over the complete flow of our system by taking into consideration several aspects of the data and system configuration. These aspects pertain to the quality and amount of training data, the use of pretrained models, and the data contained therein, as well as the identity of the target speaker; with regard to the latter, we show that we can perform zero-shot lip adaptation to an unseen identity by simply updating the shape of the lips in our model.

This article analyses the ways in which the rural – understood in its double nature, of traditional mythological space, and of degraded periphery of the urban – was both instrumentalized by the cultural and political discourses within Romanian postcommunism, and revealed as a symptomatic field in which the harsh economic realities of the post-1989 implementation of neoliberal capitalism in Eastern Europe became evident. The first part of the article describes how misplaced public policies during 1980s’ communism favoured the underdevelopment of the rural. Later, the process of transition encouraged both external migration, which depopulated Romanian villages, and the internal migration of the urban middle-class to an economically more secure rural area. The second part of the article analyses – from a materialist viewpoint – corresponding topics developed in Romanian contemporary literature and films (authored by Adrian Șchiop, Lavinia Braniște, Bogdan Mirică, and Radu Muntean). These works envision the rural/periurban as an asymmetrical geography, where violence toward the vulnerable is enacted, where the loss of community values leaves black markets flourish, and the capitalist dismantling of social ties becomes striking.

The educational process was put to a lot of pressure during the last couple of pandemic years. The teaching related activities had to quickly move from a traditional on-site environment to a completely new terrain endowed with various digital facilities. The speed with which the events unfolded, corelated with the large variety of used technologies and the lack of face-to-face interconnection represent new factors that affected both the teachers and the learners. Statistics taken from universities and pre-university institutions revealed that many students encountered difficulties in maintaining a high interest level for the ongoing educational activities. New tools must be added to the current digital environments to motivate the students and to revive their learning drive. Our own approach to on-line or hybrid education has been materialized in a new software platform that, besides embedding all the tools needed by both the teachers and the learners, introduces a series of gamification and social media inspired techniques that helps maintaining the learners’ interest for the live or asynchronous educational material presentations, thus ensuring pedagogical continuity.

This paper introduces the ZevoMOS entry to the main track of the VoiceMOS Challenge 2022. The ZevoMOS submission is based on a two-step finetuning of pretrained self-supervised learning (SSL) speech models. The first step uses a task of clas-sifying natural versus synthetic speech, while the second step’s task is to predict the MOS scores associated with each training sample. The results of the finetuning process are then combined with the confidence scores extracted from an automatic speech recognition model, as well as the raw embeddings of the training samples obtained from a wav2vec SSL speech model. The team id assigned to the ZevoMOS system within the VoiceMOS Challenge is T01. The submission was placed on the 14th place with respect to the system-level SRCC, and on the 9th place with respect to the utterance-level MSE. The paper also introduces additional evaluations of the intermediate results.

Speaker embeddings represent a means to extract representative vectorial representations from a speech signal such that the representation pertains to the speaker identity alone. The embeddings are commonly used to classify and discriminate between different speakers. However, there is no objective measure to evaluate the ability of a speaker embedding to disentangle the speaker identity from the other speech characteristics. This means that the embeddings are far from ideal, highly dependent on the training corpus and still include a degree of residual information pertaining to factors such as linguistic content, recording conditions or speaking style of the utterance. This paper introduces an analysis over six sets of speaker embeddings extracted with some of the most recent and high-performing deep neural network (DNN) architectures, and in particular, the degree to which they are able to truly disentangle the speaker identity from the speech signal. To correctly evaluate the architectures, a large multi-speaker parallel speech dataset is used. The dataset includes 46 speakers uttering the same set of prompts, recorded in either a professional studio or their home environments. The analysis looks into the intra- and inter-speaker similarity measures computed over the different embedding sets, as well as if simple classification and regression methods are able to extract several residual information factors from the speaker embeddings. The results show that the discriminative power of the analyzed embeddings is very high, yet across all the analyzed architectures, residual information is still present in the representations in the form of a high correlation to the recording conditions, linguistic contents and utterance duration. However, we show that this correlation, although not ideal, could still be useful in downstream tasks. The low-dimensional projections of the speaker embeddings show similar behavior patterns across the embedding sets with respect to intra-speaker data clustering and utterance outlier detection.