@inproceedings{KraljevskiDuckhornWolffetal., author = {Kraljevski, Ivan and Duckhorn, Frank and Wolff, Matthias and Hoffmann, R{\"u}diger}, title = {Multi-condition training and adaptation for noise robust speech recognition}, series = {Elektronische Sprachsignalverarbeitung 2012, Tagungsband der 23. Konferenz, Cottbus, 29. - 31. August 2012}, booktitle = {Elektronische Sprachsignalverarbeitung 2012, Tagungsband der 23. Konferenz, Cottbus, 29. - 31. August 2012}, editor = {Wolff, Matthias}, isbn = {978-3-942710-81-7}, pages = {73 -- 80}, language = {en} } @misc{TschoepeKraljevskiDuckhornetal., author = {Tsch{\"o}pe, Constanze and Kraljevski, Ivan and Duckhorn, Frank and Wolff, Matthias}, title = {Sprachtechnologie und akustische Mustererkennung in der medizinischen Anwendung}, series = {16. Landeskonferenz „Digitalisierung im Gesundheitswesen" 2021}, journal = {16. Landeskonferenz „Digitalisierung im Gesundheitswesen" 2021}, pages = {9}, language = {de} } @misc{SchmidtHildebrandKraljevskietal., author = {Schmidt, Ralph Rudi and Hildebrand, Jorg and Kraljevski, Ivan and Duckhorn, Frank and Tsch{\"o}pe, Constanze}, title = {A study for laser additive manufacturing quality and material classification using machine learning}, series = {2022 IEEE sensors}, journal = {2022 IEEE sensors}, publisher = {Institute of Electrical and Electronics Engineers (IEEE)}, address = {Piscataway, New Jersey}, isbn = {978-1-6654-8464-0}, issn = {2168-9229}, doi = {10.1109/SENSORS52175.2022.9967311}, pages = {1 -- 4}, abstract = {This paper demonstrates the use of acoustic emissions (AEs) to monitor the quality, and material used, for the laser additive manufacturing (LAM) process with steel and copper wire. Layers of deposited material (steel or copper) were created using LAM. The quality of these layers was either good or unstable. The AEs were recorded using three sensors, one microphone, and two structure-borne sound probes. The recorded signals were processed and transformed using the fast Fourier method. Then models were trained with the processed data and evaluated using a fivefold cross-validation. Results show that it is possible to accurately classify the materials used during LAM (up to a balanced accuracy [BAcc] score of 0.99). Also, the process quality could be classified with a BAcc score of up to 0.81. Overall, the results are promising, but further research and data collection are necessary for a proper validation of our results.}, language = {en} } @misc{KraljevskiDuckhornSobeetal., author = {Kraljevski, Ivan and Duckhorn, Frank and Sobe, Daniel and Tsch{\"o}pe, Constanze and Wolff, Matthias}, title = {Speech-to-text in upper sorbian : current state}, series = {Elektronische Sprachsignalverarbeitung 2025 : Tagungsband der 36. Konferenz Halle/Saale, 5.-7. M{\"a}rz 2025}, journal = {Elektronische Sprachsignalverarbeitung 2025 : Tagungsband der 36. Konferenz Halle/Saale, 5.-7. M{\"a}rz 2025}, editor = {Grawunder, Sven}, publisher = {TUDpress}, address = {Dresden}, isbn = {978-3-95908-803-9}, issn = {0940-6832}, pages = {109 -- 116}, abstract = {This study presents recent advancements in Upper Sorbian Speech-to-Text (STT) technology. We provide an overview of the Sorbian languages, the available speech and language resources, and the development of an STT system based on a traditional approach, which includes acoustic, pronunciation, and language modeling. Due to the scarcity of resources for Sorbian languages, our approach leverages sub-word and word-class modeling techniques. The word-class modeling is based on Finite-State Transducer definitions, which are applicable to both offline text parsing and integration into the decoding graph of the STT system. Word-class parsing is performed on the speech corpus and utilized for language modeling with complete words, sub-word units, or both. Additionally, the same definitions can be applied to Named Entity Recognition during the post-processing of recognized transcriptions. This approach significantly reduces out-of-vocabulary words and enables greater customization of the recognizer for domain-specific applications. The system was implemented for the real-time transcription of church sermon broadcasts in Upper Sorbian. The domain-specific system achieved performance comparable to fine-tuned OpenAI Whisper models developed also by other initiatives while also providing a resource-efficient solution with semantically tagged recognition results.}, language = {en} } @misc{KraljevskiDuckhornSobeetal., author = {Kraljevski, Ivan and Duckhorn, Frank and Sobe, Daniel and Tsch{\"o}pe, Constanze and Wolff, Matthias}, title = {Preserving Language Heritage Through Speech Technology: The Case of Upper Sorbian}, series = {Speech and Computer, SPECOM 2024, Belgrade, Serbia, 25-28 November 2024}, journal = {Speech and Computer, SPECOM 2024, Belgrade, Serbia, 25-28 November 2024}, editor = {Karpov, Alexey and Delic, Vlado}, publisher = {Springer Nature Switzerland, Cham}, isbn = {978-3-031-77960-2}, doi = {10.1007/978-3-031-77961-9_1}, pages = {3 -- 22}, abstract = {The modern world is facing a crisis with the rapid disappearance of endangered languages, which poses a serious threat to global cultural diversity. Speech Technologies and Artificial Intelligence present promising opportunities to address this crisis by supporting the documentation, revitalization, and everyday use of these vulnerable languages. However, despite recent and remarkable advancements in speech technology, significant challenges persist, particularly for languages with very limited resources and unique linguistic features. This paper details the development of Upper Sorbian speech technologies, focusing on the creation of a practical Speech-to-Text (STT) system as a versatile tool for language preservation. The study explores the current state of Sorbian languages and underscores collaborative efforts with the Foundation for the Sorbian People. Through a series of pilot and successive projects, each phase has contributed to the steady advancement of speech recognition modules and supporting tools, improving their performance, effectiveness and practical usability.}, language = {en} } @misc{KraljevskiDuckhornJuetal., author = {Kraljevski, Ivan and Duckhorn, Frank and Ju, Yong Chul and Tsch{\"o}pe, Constanze and Richter, Christian and Wolff, Matthias}, title = {Acoustic Resonance Recognition of Coins}, series = {2020 IEEE International Instrumentation and Measurement Technology Conference (I2MTC), 25-28 May 2020, Dubrovnik, Croatia}, journal = {2020 IEEE International Instrumentation and Measurement Technology Conference (I2MTC), 25-28 May 2020, Dubrovnik, Croatia}, publisher = {IEEE Xplore}, isbn = {978-1-7281-4460-3}, doi = {10.1109/I2MTC43012.2020.9129256}, pages = {6}, abstract = {In this study, we compare different machine learning approaches applied to acoustic resonance recognition of coins. Euro-cents and Euro-coins were classified by the sound emerging when throwing the coins onto a hard surface.The used dataset is a representative example of a small data which was collected in carefully prepared experiments.Due to the small number of coin specimens and the count of the collected observations, it was interesting to see whether deep learning methods can achieve similarly or maybe even better classification performances compared with more traditional methods.The results of the multi-class prediction of coin denominations are presented and compared in terms of balanced accuracy and Matthews Correlation Coefficient metrics. The feature analysis methods combined with the employed classifiers achieved acceptable results, despite the relatively small dataset.}, language = {en} } @misc{KraljevskiDuckhornJuetal., author = {Kraljevski, Ivan and Duckhorn, Frank and Ju, Yong Chul and Tsch{\"o}pe, Constanze and Wolff, Matthias}, title = {Acoustic Resonance Testing of Glass IV Bottles}, series = {Artificial Intelligence Applications and Innovations : 16th IFIP WG 12.5 International Conference, AIAI 2020, Neos Marmaras, Greece, June 5-7, 2020, Proceedings, Part II}, volume = {Cham}, journal = {Artificial Intelligence Applications and Innovations : 16th IFIP WG 12.5 International Conference, AIAI 2020, Neos Marmaras, Greece, June 5-7, 2020, Proceedings, Part II}, editor = {Maglogiannis, Ilias and Iliadis, Lazaros S. and Pimenidis, Elias}, publisher = {Springer International Publishing}, isbn = {978-3-030-49186-4}, issn = {1868-4238}, doi = {10.1007/978-3-030-49186-4_17}, pages = {195 -- 206}, abstract = {In this paper, acoustic resonance testing on glass intravenous (IV) bottles is presented. Different machine learning methods were applied to distinguish acoustic observations of bottles with defects from the intact ones. Due to the very limited amount of available specimens, the question arises whether the deep learning methods can achieve similar or even better detection performance compared with traditional methods.}, language = {en} } @misc{KraljevskiRjelkaDuckhornetal., author = {Kraljevski, Ivan and Rjelka, Marek and Duckhorn, Frank and Tsch{\"o}pe, Constanze and Wolff, Matthias}, title = {Cross-Lingual Acoustic Modeling in Upper Sorbian - Preliminary Study}, series = {Elektronische Sprachsignalverarbeitung 2021 : Tagungsband der 32. Konferenz Berlin, 3.-5. M{\"a}rz 2021}, journal = {Elektronische Sprachsignalverarbeitung 2021 : Tagungsband der 32. Konferenz Berlin, 3.-5. M{\"a}rz 2021}, editor = {Hillmann, Stefan and Weiss, Benjamin and Michael, Thilo and M{\"o}ller, Sebastian}, publisher = {TUDpress}, address = {Dresden}, isbn = {978-3-959082-27-3}, issn = {0940-6832}, pages = {43 -- 50}, abstract = {In this paper, we present a preliminary study for acoustic modeling in Upper Sorbian, where a model of German was used in cross-lingual transfer learning. At first, we define the grapheme and phoneme inventories and map the target phonemes from the most similar German source equivalents. Phonetically balanced sentences for the recording prompts were selected from a combination of general and domain-specific textual data. The speech corpora with a total duration of around 11 hours was collected in controlled recording sessions involving an equal number of females, males, and children. The baseline acoustic model was employed to force-align the speech corpora given the knowledge-based phoneme mappings. How well the mappings were, was evaluated by the phoneme confusions in free-phoneme recognition. The new derived data-driven model with a reduced phoneme set was included in the adaptation and evaluation along with the baseline acoustic model. The model adaptation performance was cross-validated with the "Leave One Group Out" strategy. We observed major improvements in phoneme error rates after adaptation for the knowledge-based and data-driven phoneme mappings. The study confirmed the feasibility of transfer learning for acoustic model adaptation in the case of Upper Sorbian, at the same time demonstrating practical usability with a small vocabulary speech recognition application (Smart Lamp).}, language = {de} } @misc{KraljevskiDuckhornTschoepeetal., author = {Kraljevski, Ivan and Duckhorn, Frank and Tsch{\"o}pe, Constanze and Wolff, Matthias}, title = {Machine Learning for Anomaly Assessment in Sensor Networks for NDT in Aerospace}, series = {IEEE Sensors Journal}, volume = {21}, journal = {IEEE Sensors Journal}, number = {9}, issn = {1558-1748}, doi = {10.1109/JSEN.2021.3062941}, pages = {11000 -- 11008}, abstract = {We investigated and compared various algorithms in machine learning for anomaly assessment with different feature analyses on ultrasonic signals recorded by sensor networks. The following methods were used and compared in anomaly detection modeling: hidden Markov models (HMM), support vector machines (SVM), isolation forest (IF), and reconstruction autoencoders (AEC). They were trained exclusively on sensor signals of the intact state of structures commonly used in various industries, like aerospace and automotive. The signals obtained on artificially introduced damage states were used for performance evaluation. Anomaly assessment was evaluated and compared using various classifiers and feature analysis methods. We introduced novel methodologies for two processes. The first was the dataset preparation with anomalies. The second was the detection and damage severity assessment utilizing the intact object state exclusively. The experiments proved that robust anomaly detection is practically feasible. We were able to train accurate classifiers which had a considerable safety margin. Precise quantitative analysis of damage severity will also be possible when calibration data become available during exploitation or by using expert knowledge.}, language = {en} } @misc{KraljevskiBissiriDuckhornetal., author = {Kraljevski, Ivan and Bissiri, Maria Paola and Duckhorn, Frank and Tsch{\"o}pe, Constanze and Wolff, Matthias}, title = {Glottal Stops in Upper Sorbian: A Data-Driven Approach}, series = {Proc. Interspeech 2021, 30 August - 3 September, 2021, Brno, Czechia}, journal = {Proc. Interspeech 2021, 30 August - 3 September, 2021, Brno, Czechia}, doi = {10.21437/Interspeech.2021-1101}, pages = {1001 -- 1005}, abstract = {We present a data-driven approach for the quantitative analysis of glottal stops before word-initial vowels in Upper Sorbian, a West Slavic minority language spoken in Germany. Glottal stops are word-boundary markers and their detection can improve the performance of automatic speech recognition and speech synthesis systems. We employed cross-language transfer using an acoustic model in German to develop a forced-alignment method for the phonetic segmentation of a read-speech corpus in Upper Sorbian. The missing phonemic units were created by combining the existing phoneme models. In the forced-alignment procedure, the glottal stops were considered optional in front of word-initial vowels. To investigate the influence of speaker type (males, females, and children) and vowel on the occurrence of glottal stops, binomial regression analysis with a generalized linear mixed model was performed. Results show that children glottalize word-initial vowels more frequently than adults, and that glottal stop occurrences are influenced by vowel quality.}, language = {en} } @misc{KraljevskiDuckhornBarthetal., author = {Kraljevski, Ivan and Duckhorn, Frank and Barth, Martin and Tsch{\"o}pe, Constanze and Schubert, Frank and Wolff, Matthias}, title = {Autoencoder-based Ultrasonic NDT of Adhesive Bonds}, series = {IEEE SENSORS 2021, Conference Proceedings, Oct 31- Nov 4, Sydney, Australia}, journal = {IEEE SENSORS 2021, Conference Proceedings, Oct 31- Nov 4, Sydney, Australia}, publisher = {IEEE}, isbn = {978-1-7281-9501-8}, doi = {10.1109/SENSORS47087.2021.9639864}, pages = {4}, abstract = {We present an approach for ultrasonic non-destructive testing of adhesive bonding employing unsupervised machine learning with autoencoders.The models are trained exclusively on the features derived from pulse-echo ultrasonic signals on a specimen with good adhesive bonding and tested on another specimen with artificially added defects.The resulting pseudo-probabilities indicating anomalies are visualized and presented along to the C-scan of the same specimen. As a result, we achieved improved representation of the defects, allowing their automatic and reliable detection.}, language = {en} } @misc{KraljevskiDuckhornTschoepeetal., author = {Kraljevski, Ivan and Duckhorn, Frank and Tsch{\"o}pe, Constanze and Wolff, Matthias}, title = {Convolutional Autoencoders for Health Indicators Extraction in Piezoelectric Sensors}, series = {2020 IEEE Sensors, 25-28 Oct. 2020, Rotterdam, Netherlands,}, journal = {2020 IEEE Sensors, 25-28 Oct. 2020, Rotterdam, Netherlands,}, address = {Rotterdam, Netherlands}, isbn = {978-1-7281-6801-2}, doi = {10.1109/SENSORS47125.2020.9323023}, pages = {1 -- 4}, abstract = {We present a method for extracting health indicators from piezoelectric sensors applied in the case of microfluidic valves. Convolutional autoencoders were used to train a model on the normal operating conditions and tested on signals of different valves. The results of the model performance evaluation, as well as, the qualitative presentation of the indicator plots for each tested component, showed that the used approach is capable of detecting features that correspond to increasing component degradation. The extracted health indicators are the prerequisite and input for reliable remaining useful life prediction.}, language = {en} } @misc{MaierKuhnDuckhornetal., author = {Maier, Isidor Konrad and Kuhn, Johannes Ferdinand Joachim and Duckhorn, Frank and Kraljevski, Ivan and Sobe, Daniel and Wolff, Matthias and Tsch{\"o}pe, Constanze}, title = {Word Class Based Language Modeling: A Case of Upper Sorbian}, series = {Proceedings of The Workshop on Resources and Technologies for Indigenous, Endangered and Lesser-resourced Languages in Eurasia within the 13th Language Resources and Evaluation Conference, Marseille, France}, journal = {Proceedings of The Workshop on Resources and Technologies for Indigenous, Endangered and Lesser-resourced Languages in Eurasia within the 13th Language Resources and Evaluation Conference, Marseille, France}, publisher = {European Language Resources Association}, isbn = {978-2-493814-07-4}, pages = {28 -- 35}, abstract = {In this paper we show how word class based language modeling can support the integration of a small language in modern applications of speech technology. The methods described in this paper can be applied for any language. We demonstrate the methods on Upper Sorbian. The word classes model the semantic expressions of numerals, date and time of day. The implementation of the created grammars was realized in the form of finite-state-transducers (FSTs) and minimalists grammars (MGs). We practically demonstrate the usage of the FSTs in a simple smart-home speech application, that is able to set wake-up alarms and appointments expressed in a variety of spontaneous and natural sentences. While the created MGs are not integrated in an application for practical use yet, they provide evidence that MGs could potentially work more efficient than FSTs in built-on applications. In particular, MGs can work with a significantly smaller lexicon size, since their more complex structure lets them generate more expressions with less items, while still avoiding wrong expressions.}, language = {en} } @misc{KraljevskiDuckhornTschoepeetal., author = {Kraljevski, Ivan and Duckhorn, Frank and Tsch{\"o}pe, Constanze and Schubert, Frank and Wolff, Matthias}, title = {Paper Tissue Softness Rating by Acoustic Emission Analysis}, series = {Applied Sciences}, volume = {13}, journal = {Applied Sciences}, number = {3}, issn = {2076-3417}, doi = {10.3390/app13031670}, abstract = {Softness is one of the essential properties of hygiene tissue products. Reliably measuring it is of utmost importance to ensure the balance between customer expectations and cost-effective tissue production. This study presents a method for assessing softness by analyzing acoustic emissions produced while tearing a tissue specimen. The aim was to train neural network models using the corrected results of human panel tests as the ground truth labels and to predict the tissue softness in two- and three-class recognition tasks. We also investigate the possibility of predicting some production parameters related to the softness property. The results proved that tissue softness and production parameters could be reliably estimated only by the tearing noise.}, language = {en} } @misc{MaierKuhnDuckhornetal., author = {Maier, Isidor Konrad and Kuhn, Johannes and Duckhorn, Frank and Kraljevski, Ivan and Sobe, Daniel and Wolff, Matthias and Tsch{\"o}pe, Constanze}, title = {Word Class Based Language Modeling: A Case of Upper Sorbian, LREC 2022, Marseille}, series = {Language Resources and Evaluation Conference 2022 (LREC 2022), Marseille, 2022-06-13}, journal = {Language Resources and Evaluation Conference 2022 (LREC 2022), Marseille, 2022-06-13}, doi = {10.5281/zenodo.7501145}, pages = {1}, language = {en} }