Challenge
Communication
Use cases
Fake voice detection on a large scale presents significant challenges. Some reasons why the use of artificial intelligence (AI)-based solutions is essential to address this problem:
Among other cases, successful implementation of speaker verification in the telecommunications domain opens up opportunities the following use cases:
1. Secure Access to Telephone Services:Ensuring that only authorised users have access to sensitive services throughvocal verification.
2. Financial Transactions: Ensure user authenticity in telephone transactions related to financial services.
3. Customer Services: Improve the security of telephone interactions with customer services.
Introduction
Challenge designed exclusively with UPF-BSM and the Telefónica Research team, to offer active, personalised and interdisciplinary learning models that prepare MsC in Data Analytics for Business students to become professionals capable of moving in constantly changing environments.
Relevance of the problem for telecommunications companies
User Security: Voice authentication is used to ensure security in telephone transactions, account access and sensitive services.
Fraud Prevention: Voice deception detection is essential to prevent fraud in telephone services and financial transactions.
Improving Customer Experience: Implementing robust speaker verification systems improves the customer experience by providing secure and efficient services.
Development of the Challenge
Fake voice detection on a large scale presents significant challenges. Some reasons why the use of artificial intelligence (AI)-based solutions is essential to address this problem:
1.Complexity of Spoofing Techniques
2.Scalability and Variability of Data
3.Adaptability to New Threats
4.Advanced Feature Extraction
5.Continuous Performance Improvement
In April 2024, the students start weeks of analysis and teamwork to present solutions to the different possible use cases.
The final presentation will take place at the beginning of May in front of a jury of experts from Telefónica and UPF-BSM professors.
Suggested resources
VoxCeleb: Dataset with celebrity voice samples.
ASVspoof 2019: Dataset for evaluating speaker verification systems and deception detectors.
LibriSpeech: Dataset for speech recognition research and speaker verification.
Custom datasets: Students can create their own datasets.