»Alessandro Torres

About Alessandro Torres

This author has not yet filled in any details.
So far Alessandro Torres has created 23 blog entries.

Trials have been thoroughly designed

The work has defined the measurement parameters and the test plans to be executed. Test and evaluation aim both at tuning the single modules and at integrating the various modules that make up the Platform. Moreover, the evaluation strategy has specifically targeted the Platform and the services it deploys in the most possible realistic scenarios, in [...]

A “corpus of corpora” database has been designed and implemented

Availability of spoken language dataset has proved crucial to the progress of research and technology on speech processing, ever since. Bearing this in mind, we have designed and implemented a unified database, where metadata of virtually any speech-dataset can be represented. The project has defined and normalised a general format for the metadata. Open-source tools [...]

Operational environment and flows for online authentication

User enrolment procedure User acces procedure Multi-channel environment: real customer management scenario foresees processes managed by the TBAS through the multiple channels provided by Findomestic (phone, web, web chat) Prompting and messages, in plain language (no coded messages), immediately understandable by the user Trial target  and Productive Environment target (may differ from those of the [...]

Determine and develop hybrid SV systems in accordance with the selected modes of operation

The optimal Convenience-Security trade-off is reached by means of the fusion of three operation modes: Fixed passphrase, Prompted text-dependent, Prompted text-independent.

Determine the single/fused (hybrid) modes of SV for different classes of access control applications

Different modes of operation (single/fused -hybrid modes of SV for different classes of access control applications) have been assessed in order to verify accuracy and vulnerability.

Unprecedented solutions to noise-robust automatic speaker verification

multi-stage noise compensation, including SNR estimation, speech enhancement, spectral feature extaction,  voice activity detection model-domain acoustic normalisation modules multiple simulations to find the best perforing NR method at each stage in the system: the best from each stage is then selected as the best system text-dependent speaker recognition results in terms of EER (in %) [...]

Dissemination and communication actions

this site, created at project start and continously updated publications (20+) participation to events, conferences and workshops social media leaflets, posters, video presentations

Created awareness about OCTAVE scope and objectives

Several papers have been produced in the context of OCTAVE work and partially funded by the project. Among many others, the following prestigious targets have been reached: INTERSPEECH 2015 SPLINE 2015 ODYSSEY 2016 IEEE ICASSP 2016 IEEE INDICON 2016 IEEE Transactions on Cybernetics 2016  

WHAT DOES IT DO?

OCTAVE delivers an access control platform through a trusted voice authentication service for use in data-sensitive and mission-critical applications in actual business/commercial environments. OCTAVE relieves users from the inconvenience of textual passwords and reduces the economic and practical burden on service providers regarding password loss and recovery. The OCTAVE platform is based on state-of-the-art speaker verification technologies that offer effective robustness against [...]

HOW DOES IT WORK?

From a user's perspective, biometric systems perform a conceptually simple task: recognition of persons. However, their implementation is generally highly complex. The general approach involves the enrolment of a person in the system and the learning of a biometric model. At the time of use, a new biometric sample is compared to the model. The [...]