Şeymanur Aktı

Şeymanur Aktı, Marwa Qaraqe, Hazım Kemal Ekenel

Food recognition is an important task for a variety of applications, including managing health conditions and assisting visually impaired people. Several food recognition studies have focused on generic types of food or specific cuisines, however, food recognition with respect to Middle Eastern cuisines has remained unexplored. Therefore, in this paper we focus on developing a mobile friendly, Middle Eastern cuisine focused food recognition application for assisted living purposes. In order to enable a low-latency, high-accuracy food classification system, we opted to utilize the Mobilenet-v2 deep learning model. As some of the foods are more popular than the others, the number of samples per class in the used Middle Eastern food dataset is relatively imbalanced. To compensate for this problem, data augmentation methods are applied on the underrepresented classes. Experimental results show that using Mobilenet-v2 architecture for this task is beneficial in terms of both accuracy and the memory usage. With the model achieving 94% accuracy on 23 food classes, the developed mobile application has potential to serve the visually impaired in automatic food recognition via images.

Sai Koneru, Fabian Retkowski, Christian Huber et al.

The globalization of education and rapid growth of online learning have made localizing educational content a critical challenge. Lecture materials are inherently multimodal, combining spoken audio with visual slides, which requires systems capable of processing multiple input modalities. To provide an accessible and complete learning experience, translations must preserve all modalities: text for reading, slides for visual understanding, and speech for auditory learning. We present \textbf{BOOM}, a multimodal multilingual lecture companion that jointly translates lecture audio and slides to produce synchronized outputs across three modalities: translated text, localized slides with preserved visual elements, and synthesized speech. This end-to-end approach enables students to access lectures in their native language while aiming to preserve the original content in its entirety. Our experiments demonstrate that slide-aware transcripts also yield cascading benefits for downstream tasks such as summarization and question answering. We release our Slide Translation code at https://github.com/saikoneru/image-translator and integrate it in Lecture Translator at https://gitlab.kit.edu/kit/isl-ai4lt/lt-middleware/ltpipeline}\footnote{All released code and models are licensed under the MIT License.

Dogucan Yaman, Fevziye Irem Eyiokur, Leonard Bärmann et al.

In the task of talking face generation, the objective is to generate a face video with lips synchronized to the corresponding audio while preserving visual details and identity information. Current methods face the challenge of learning accurate lip synchronization while avoiding detrimental effects on visual quality, as well as robustly evaluating such synchronization. To tackle these problems, we propose utilizing an audio-visual speech representation expert (AV-HuBERT) for calculating lip synchronization loss during training. Moreover, leveraging AV-HuBERT's features, we introduce three novel lip synchronization evaluation metrics, aiming to provide a comprehensive assessment of lip synchronization performance. Experimental results, along with a detailed ablation study, demonstrate the effectiveness of our approach and the utility of the proposed evaluation metrics.

Tuan Nam Nguyen, Seymanur Akti, Ngoc Quan Pham et al.

Previous approaches on accent conversion (AC) mainly aimed at making non-native speech sound more native while maintaining the original content and speaker identity. However, non-native speakers sometimes have pronunciation issues, which can make it difficult for listeners to understand them. Hence, we developed a new AC approach that not only focuses on accent conversion but also improves pronunciation of non-native accented speaker. By providing the non-native audio and the corresponding transcript, we generate the ideal ground-truth audio with native-like pronunciation with original duration and prosody. This ground-truth data aids the model in learning a direct mapping between accented and native speech. We utilize the end-to-end VITS framework to achieve high-quality waveform reconstruction for the AC task. As a result, our system not only produces audio that closely resembles native accents and while retaining the original speaker's identity but also improve pronunciation, as demonstrated by evaluation results.

Sai Koneru, Maike Züfle, Thai-Binh Nguyen et al.

The scope of the International Workshop on Spoken Language Translation (IWSLT) has recently broadened beyond traditional Speech Translation (ST) to encompass a wider array of tasks, including Speech Question Answering and Summarization. This shift is partly driven by the growing capabilities of modern systems, particularly with the success of Large Language Models (LLMs). In this paper, we present the Karlsruhe Institute of Technology's submissions for the Offline ST and Instruction Following (IF) tracks, where we leverage LLMs to enhance performance across all tasks. For the Offline ST track, we propose a pipeline that employs multiple automatic speech recognition systems, whose outputs are fused using an LLM with document-level context. This is followed by a two-step translation process, incorporating additional refinement step to improve translation quality. For the IF track, we develop an end-to-end model that integrates a speech encoder with an LLM to perform a wide range of instruction-following tasks. We complement it with a final document-level refinement stage to further enhance output quality by using contextual information.

Seymanur Akti, Tuan Nam Nguyen, Alexander Waibel

Expressive voice conversion aims to transfer both speaker identity and expressive attributes from a target speech to a given source speech. In this work, we improve over a self-supervised, non-autoregressive framework with a conditional variational autoencoder, focusing on reducing source timbre leakage and improving linguistic-acoustic disentanglement for better style transfer. To minimize style leakage, we use multilingual discrete speech units for content representation and reinforce embeddings with augmentation-based similarity loss and mix-style layer normalization. To enhance expressivity transfer, we incorporate local F0 information via cross-attention and extract style embeddings enriched with global pitch and energy features. Experiments show our model outperforms baselines in emotion and speaker similarity, demonstrating superior style adaptation and reduced source style leakage.

Şeymanur Aktı, Ferda Ofli, Muhammad Imran et al.

Detecting fights from still images shared on social media is an important task required to limit the distribution of violent scenes in order to prevent their negative effects. For this reason, in this study, we address the problem of fight detection from still images collected from the web and social media. We explore how well one can detect fights from just a single still image. We also propose a new dataset, named Social Media Fight Images (SMFI), comprising real-world images of fight actions. Results of the extensive experiments on the proposed dataset show that fight actions can be recognized successfully from still images. That is, even without exploiting the temporal information, it is possible to detect fights with high accuracy by utilizing appearance only. We also perform cross-dataset experiments to evaluate the representation capacity of the collected dataset. These experiments indicate that, as in the other computer vision problems, there exists a dataset bias for the fight recognition problem. Although the methods achieve close to 100% accuracy when trained and tested on the same fight dataset, the cross-dataset accuracies are significantly lower, i.e., around 70% when more representative datasets are used for training. SMFI dataset is found to be one of the two most representative datasets among the utilized five fight datasets.

Şeymanur Aktı, Doğay Kamar, Özgür Anıl Özlü et al.

Predicting the evolution of the brain network, also called connectome, by foreseeing changes in the connectivity weights linking pairs of anatomical regions makes it possible to spot connectivity-related neurological disorders in earlier stages and detect the development of potential connectomic anomalies. Remarkably, such a challenging prediction problem remains least explored in the predictive connectomics literature. It is a known fact that machine learning (ML) methods have proven their predictive abilities in a wide variety of computer vision problems. However, ML techniques specifically tailored for the prediction of brain connectivity evolution trajectory from a single timepoint are almost absent. To fill this gap, we organized a Kaggle competition where 20 competing teams designed advanced machine learning pipelines for predicting the brain connectivity evolution from a single timepoint. The competing teams developed their ML pipelines with a combination of data pre-processing, dimensionality reduction, and learning methods. Utilizing an inclusive evaluation approach, we ranked the methods based on two complementary evaluation metrics (mean absolute error (MAE) and Pearson Correlation Coefficient (PCC)) and their performances using different training and testing data perturbation strategies (single random split and cross-validation). The final rank was calculated using the rank product for each competing team across all evaluation measures and validation strategies. In support of open science, the developed 20 ML pipelines along with the connectomic dataset are made available on GitHub. The outcomes of this competition are anticipated to lead to the further development of predictive models that can foresee the evolution of brain connectivity over time, as well as other types of networks (e.g., genetic networks).

Fabio Valerio Massoli, Fabrizio Falchi, Alperen Kantarci et al.

Anomalies are ubiquitous in all scientific fields and can express an unexpected event due to incomplete knowledge about the data distribution or an unknown process that suddenly comes into play and distorts observations. Due to such events' rarity, to train deep learning models on the Anomaly Detection (AD) task, scientists only rely on "normal" data, i.e., non-anomalous samples. Thus, letting the neural network infer the distribution beneath the input data. In such a context, we propose a novel framework, named Multi-layer One-Class ClassificAtion (MOCCA),to train and test deep learning models on the AD task. Specifically, we applied it to autoencoders. A key novelty in our work stems from the explicit optimization of intermediate representations for the AD task. Indeed, differently from commonly used approaches that consider a neural network as a single computational block, i.e., using the output of the last layer only, MOCCA explicitly leverages the multi-layer structure of deep architectures. Each layer's feature space is optimized for AD during training, while in the test phase, the deep representations extracted from the trained layers are combined to detect anomalies. With MOCCA, we split the training process into two steps. First, the autoencoder is trained on the reconstruction task only. Then, we only retain the encoder tasked with minimizing the L_2 distance between the output representation and a reference point, the anomaly-free training data centroid, at each considered layer. Subsequently, we combine the deep features extracted at the various trained layers of the encoder model to detect anomalies at inference time. To assess the performance of the models trained with MOCCA, we conduct extensive experiments on publicly available datasets. We show that our proposed method reaches comparable or superior performance to state-of-the-art approaches available in the literature.