Voice cloning, an emerging field in the speech-processing area, aims to generate synthetic utterances that closely resemble the voices of specific individuals. In this study, we investigated the impact of various techniques on improving the quality of voice cloning, specifically focusing on a low-quality dataset. To contrast our findings, we also used two high-quality corpora for comparative analysis. We conducted exhaustive evaluations of the quality of the gathered corpora in order to select the most-suitable data for the training of a voice-cloning system. Following these measurements, we conducted a series of ablations by removing audio files with a lower signal-to-noise ratio and higher variability in utterance speed from the corpora in order to decrease their heterogeneity. Furthermore, we introduced a novel algorithm that calculates the fraction of aligned input characters by exploiting the attention matrix of the Tacotron 2 text-to-speech system. This algorithm provides a valuable metric for evaluating the alignment quality during the voice-cloning process. We present the results of our experiments, demonstrating that the performed ablations significantly increased the quality of synthesised audio for the challenging low-quality corpus. Notably, our findings indicated that models trained on a 3 h corpus from a pre-trained model exhibit comparable audio quality to models trained from scratch using significantly larger amounts of data.

This work introduces the design and assessment of a voice-controlled elevator system aimed at facilitating touchless interaction between users and hardware, thereby minimising contact and improving accessibility for individuals with disabilities. The research distinguishes three distinct deployment scenarios – on cloud, on edge and embedded – with the ultimate goal of integrating the entire system into a low-resource environment on a custom carrier board. An objective evaluation measured acoustic conditions rigorously using a dataset of 2900 audio files recorded inside a laboratory elevator cabin featuring two internal coatings, five audio input devices, and under four distinct noise conditions. The study evaluated the performance of two Automatic Speech Recognition systems: Google's Speech-to-Text API and a Kaldi model adapted for this task, deployed using Vosk. Additionally, latency times for these transcribers and two communication protocols were measured to enhance efficiency. Finally, two subjective evaluations on clean and noisy conditions were conducted simulating a real world scenario. The results, yielding 84.7 and 77.2 points respectively in a System Usability Scale questionnaire, affirm the reliability of the presented prototype for industrial deployment.

The growing need for secure and privacy-preserving identity management in industrial environments has exposed the limitations of traditional, centralized authentication systems. In this context, SIBERIA was developed as a modular solution that empowers users to control their own digital identities, while ensuring robust protection of critical services. The system is designed in alignment with European standards and regulations, including EBSI, eIDAS 2.0, and the GDPR. SIBERIA integrates a Self-Sovereign Identity (SSI) framework with a decentralized blockchain-based infrastructure for the issuance and verification of Verifiable Credentials (VCs). It incorporates multi-factor authentication by combining a voice biometric module, enhanced with spoofing-aware techniques to detect synthetic or replayed audio, and a behavioral biometrics module that provides continuous authentication by monitoring user interaction patterns. The system enables secure and user-centric identity management in industrial contexts, ensuring high resistance to impersonation and credential theft while maintaining regulatory compliance. SIBERIA demonstrates that it is possible to achieve both strong security and user autonomy in digital identity systems by leveraging decentralized technologies and advanced biometric verification methods.

Voice cloning, an emerging field in the speech processing area, aims to generate synthetic utterances that closely resemble the voices of specific individuals. In this study, we investigate the impact of various techniques on improving the quality of voice cloning, specifically focusing on a low-quality dataset. To contrast our findings, we also use two high-quality corpora for comparative analysis. We conduct exhaustive evaluations of the quality of the gathered corpora in order to select the most suitable audios for the training of a Voice Cloning system. Following these measurements, we conduct a series of ablations by removing audios with lower SNR and higher variability in utterance speed from the corpora in order to decrease their heterogeneity. Furthermore, we introduce a novel algorithm that calculates the fraction of aligned input characters by exploiting the attention matrix of the Tacotron 2 Text-to-Speech (TTS) system. This algorithm provides a valuable metric for evaluating the alignment quality during the voice cloning process. We present the results of our experiments, demonstrating that the performed ablations significantly increase the quality of synthesised audios for the challenging low-quality corpus. Notably, our findings indicate that models trained on a 3-hour corpus from a pre-trained model exhibit comparable audio quality to models trained from scratch using significantly larger amounts of data.