The main objective of this paper is the development of a speech and text recognition system to improve security in user identification. For the development of the system, deep learning methodologies and several Python libraries were implemented, including Speech_recognition, Pyttsx3, and Librosa, among others. The system was evaluated in a controlled environment using 50 speech samples, obtaining an accuracy of 74%. The results indicated that 61.53% of the errors were due to failures in voice identification and 30.76% were due to discrepancies in matching the generated text. These findings underscore the overall effectiveness of the system, although they also point to the need to adjust the similarity thresholds and improve the recognition algorithms to increase their accuracy and robustness. It is concluded that the system presents a promising solution for biometric voice authentication...

The impact of behavioral instructions on language models is a fundamental area of research in the field of processing for language that is human. This study focuses on analyzing how specific directions provided to language models affect their performance and efficiency on various tasks. It examines in detail the importance of instructions in the understanding for languages to be human and their influence on applications in activities such as machine translation, textual content creation, and document categorization. It discusses how behavioral instructions impact the configuration and training of models, as well as their predictive and generative capabilities. Concrete examples of how instructions can improve or limit the performance of linguistic models in different contexts are presented. The results obtained highlight the need to carefully consider behavioral instructions when develop...

The increasing use of QR codes across various sectors has facilitated the transfer of information but has also exposed users to new cybersecurity threats, such as quishing, a variant of phishing that leverages these codes to redirect users to malicious websites. To address this issue, the study aimed to implement an antimalware system that employs machine learning alongside the VirusTotal API to analyze and classify links embedded in QR codes in real time. The methodology was structured into four stages: capturing and decoding QR codes using OpenCV, analyzing extracted URLs with the VirusTotal API, issuing preventive alerts based on the link classification, and evaluating system performance with a dataset of 100 QR codes (50 safe and 50 malicious). The results showed 100 % accuracy, 95 % sensitivity, and an average response time of 48.95 ms. No false positives were detected, and only a s...