[Meet our Faculty] Massimiliano Todisco, expert in AI for biometrics at EURECOM

EURECOM Communication
7 min readAug 25, 2023

In the complex interplay between voice authenticity and the imperative of user privacy, Dr Massimiliano Todisco is at the forefront of security at EURECOM. Specialising in audio and voice technologies, he is dedicated not only to preserving authenticity but also the confidentiality of user data. Balancing the challenges of sound engineering and artificial intelligence, Dr. Todisco’s research seeks to answer a general question: How can we meticulously recognise the authenticity of digital audio while ensuring that user privacy remains intact in the vast domain of digital audio processing?

Q. What is your academic trajectory?

MT: My academic foundation is rooted in physics, but I pursued an engineering doctorate with a specialization in music signal processing and AI, concentrating on automated transcription of polyphonic piano compositions. This expertise set the stage for my diverse research interests, which include signal processing, pattern recognition, and machine learning.

Upon completing my PhD, I committed three additional years to Tor Vergata University of Rome and Fondazionne Bordoni as a postdoctoral researcher. I further honed my skills by securing a second master’s degree in sound engineering. Beyond the academic realm, my commitment to the field is also personal: I am a professional musician, having earned a trumpet degree from the Conservatory of Rome. This blend of academia and artistry drives my passion for sound and music engineering.

My time in Rome evolved from a focus on music signals to an interest in speech signals. This led to my collaboration on a successful EU H2020 project, named OCTAVE, providing an opportunity to work closely with Prof. Nick Evans from EURECOM, a distinguished figure in speech technologies. In 2015, I accepted EURECOM’s offer to join their team, which culminated in a rewarding 5-year postdoctoral position within Nick’s group, delving into speaker verification and anti-spoofing. By 2020, I was appointed as an assistant professor at EURECOM, and by 2023, I was promoted to the position of associate professor.

Q. What are the expertise you bring to the Digital Security department?

MT: While both the music and speech sectors work with audio signals, there’s historically been limited interaction between them. My background allowed me to bridge this gap, integrating tools and expertise from the music domain to enhance the speech domain.

At EURECOM, I introduced an innovative research direction. Within the Digital Security department, there’s the Crypto group and the Biometric Group. I suggested a fusion of these two, leveraging cryptography and biometrics synergistically. The objective? Implementing privacy-enhancing algorithms within biometric systems to protect sensitive data. Currently, we are co-mentoring two PhD students on interdisciplinary projects that intersect privacy, security and biometrics.

I’m fully aware of the prevalent privacy concerns, and while I’m not a cryptographer by training, my expertise in signal processing can contribute significantly to the refinement of privacy-enhancing algorithms. It’s worth noting that machine learning, which underpins biometric recognition, hasn’t traditionally been built with privacy and security at its core. As legislative mandates evolve, we are compelled to prioritize these aspects. The need for privacy-preserving methods that align with biometric signal processing is more urgent than ever. A critical hurdle we face is the noticeable decline in accuracy when integrating cryptography with machine learning. Our goal is to adapt and fine-tune privacy-enhancing techniques to ensure they complement machine learning systems effectively.

Q. What are the highlights of your research here at EURECOM?

MT: In the year 2016, our team at EURECOM broke new ground by crafting an advanced anti-spoofing framework. Its innovation and impact have since been acknowledged widely, making it a benchmark tool within the community. Our dedication was rewarded with the best paper award at the Odyssey 2016 conference held in Bilbao. Adding to our honors, our contribution was also celebrated as the standout paper in the Computer Speech and Language journal from 2015 to 2019. These recognitions, without a doubt, are milestones I cherish deeply!

Q. What are your current research projects that you are working on?

MT: One of my prominent involvements has been with the ANR PRCI RESPECT project, which focuses on privacy preservation, multi-biometric personal verification, and anti-spoofing. Now approaching its culmination, RESPECT is a collaborative effort between Germany and France and has been driven by the goal to balance robust security measures with unwavering privacy preservation. We have been striving to augment the accuracy and reliability of recognition, which we anticipate will fortify trust in the emergent multi-biometric audio-visual recognition technologies. These advancements aim to unlock significant societal and economic value.

Concurrently, I’m actively guiding the Marie Curie Early Training Network (ETN) initiative, TReSPAsS-ETN. Its collaborative effort, now nearing completion, unites seven universities with seven industry partners across France, Germany, the Netherlands, Switzerland, Spain, and Belgium. Our mission is to sculpt Early Stage Researchers at the nexus of biometrics, privacy, security, and the inherent legal and ethical considerations. On the research frontier, TReSPAsS-ETN is poised to pioneer a new-age security paradigm, fusing cutting-edge presentation attack detection methodologies with practical encryption solutions. Importantly, we’re ensuring that the project’s scope transcends just engineering, giving due regard to the legal and ethical underpinnings of security.

Q. TReSPAsS-ETN project has an interesting format combining research and training. Could you explain how it works in more detail?

MT. The TReSPAsS-ETN Marie Curie initiative is an integrated partnership of 7 academic and 7 industrial entities. Our primary objective is to provide comprehensive training to Early Stage Researchers (ESRs), which presently includes 14 PhD students. These ESRs actively collaborate with our industrial partners, even undergoing 3-month internships, to gain insights into the industrial research approach and contrast it with academic methodologies. This project is designed to foster synergy among researchers from varied fields and specializations.

Though TReSPAsS-ETN commenced on 1st January 2020, right before the pandemic, its progress was notably impacted, particularly concerning the mobility of our PhD students. Within this framework, EURECOM’s responsibility is to embed cryptography in biometric recognition and devise countermeasures, such as anti-spoofing techniques.

In today’s landscape, issues related to biometrics, attack detection, privacy, security, and legal-ethical concerns are escalating. Our most challenging task lies in seamlessly integrating cryptography algorithms into biometric systems designed for individual recognition. We’re keen on exploring efficient solutions by intertwining deep learning with cryptography and optimizing the use of biometric data for personal identification.

Q: Can you share any recent accomplishments or grants you’ve received in relation to your research?

MT. I am delighted to report that I was recently awarded the ANR JCJC grant for my project, P-SPIKE. This project envisions a future where speech technology inherently prioritizes privacy, aligning with regulations like the General Data Protection Regulation (GDPR). P-SPIKE aims to bolster trust in speaker authentication, with its methodologies finding applications beyond biometrics. Our exploration into neuromorphic hardware, designed to efficiently run Spike Neural Networks (SNNs), promises a revolution in artificial intelligence. With its energy-efficient, rapid spike-based processing, it’s a game-changer for real-time voice applications. While SNNs in privacy-centric speech technology is a nascent field, P-SPIKE embodies our dedication to pioneering sustainable and confidential voice technologies.

Q. What are the real-life applications for your research?

MT. My primary focus is on enhancing privacy and security in voice biometrics. Consider voice-activated home systems like Alexa. When making online purchases, it’s crucial that devices like Alexa or Google accurately recognize the account owner. These systems are vulnerable to threats from attackers using deepfake voice generation or simple audio playback. Beyond just security concerns, there are significant privacy implications. Voice data can inadvertently reveal sensitive details about an individual, such as their health condition, gender, and more. Addressing these vulnerabilities and ensuring the protection of such intimate information is at the heart of my research.

Q. Do you have common research projects with other departments in EURECOM?

MT. Absolutely. I often collaborate with Prof. Maria Zuluaga from the Data Science Department on projects related to voice and health. Maria’s team is an expert in integrating AI with healthcare solutions. During the pandemic, we started a project to detect Covid-19 through voice and cough patterns. We extended our collaboration beyond EURECOM to a company in Spain. This company actively collected voice data from patients with Covid-19 confirmed in hospitals, after verification of the PCR test. Our ultimate goal? To enable the detection of Covid-19 with a simple mobile phone and a conversation.

Additionally, I’ve recently embarked on another research journey, this time in collaboration with Prof. Chiara Galdi, focusing on ensuring fairness in biometrics. This initiative responds to the increasing demands and guidelines set by the GDPR to address and promote fairness and equity in biometric applications.

Q. What is the take-home message you would like to give?

MT. The landscape of technology, especially in the realm of biometrics and digital security, is evolving rapidly. As we navigate these advancements, it’s paramount that we prioritize privacy, security, fairness, and environmental sustainability at every juncture. The intersection of technology with our daily lives means that while we benefit from its conveniences, we must also be vigilant against potential threats and biases. My work, and the broader work at EURECOM, seeks to bridge this gap — ensuring that as we push the boundaries of what’s possible, we do so responsibly and ethically. In essence, as we embrace the future, let’s make certain it’s a secure and equitable one for all.

REFERENCES

http://www.respect-project.eu/

https://www.asvspoof.org/

https://www.massimilianotodisco.eu/

--

--

EURECOM Communication

Graduate school & Research Center in digital science with a strong international perspective, located in the Sophia Antipolis technology park.