BATVOX

BATVOX is an advanced 1:1 Voice Biometrics tool designed for forensic experts and scientific police to perform speaker verification and compile expert reports as evidence in court.

Features

  1. Case management: BATVOX allows the organization of both audio and model, by case and session, in which one or more calculations can be run, thus facilitating the investigation.
  2. Speaker Identification (SID): BATVOX allows the identification of unknown voices against voices coming from known speakers, as well as identity verification of a speaker making 1:1 comparisons, independently of text, channel and language. BATVOX provides detailed speaker verifications with Likelihood Ratios (LR) computation based on Bayesian Networks. The LR gives a robust estimation of the verification task, in a probabilistic way, which can be easily presented and justified in legal processes, always with the support of an expert. The calculation process can be exported to an HTML file
    to be presented in court.
  3. BATVOX, 2 solutions:
    • BATVOX Basic: a single-user standalone solution for
      forensic experts based on a single machine.
    • BATVOX Pro: a multi-user and distributed solution,
      Client-Server ideally suited for larger organizations and
      laboratories, managing a central database, interconnected
      with users’ Client PCs through a local network.

Benefits

Based on the brand new iVector algorithm, BATVOX uses AGNITIO’s 4th generation technology and achieves the fastest and most accurate results ever, providing channel, text and language independent scores and the best accuracy for very short utterances.

  • New way to compute Likelihood Ratios (LR) based on the Bayesian Networks approach, thus dramatically reducing discrimination and calibration losses
  • BATVOX gives a graphic representation of the LR. A LR shows that it is the X (=LR) times more likely that the test and the suspects are the same person than they are not
  • BATVOX allows creation of target models, providing an automatic, accurate speaker separation technology
  • High precision, thanks to Reference Populations and Impostors (for normalization and Likelihood Ratio calculation purposes) – quality measures and detection of outliers are automatically provided
  • Creation of reports to present results to court
  • Proven in the most demanding and challenging environments (de facto standard Voice Biometrics forensic tool in more than 35 0countries worldwide), works even with low quality speech signals
  • Only 7s of net speech are needed to perform identification, and 30s to create a speaker model.
  • Small model size: under 10kB (smallest in the marketplace).