ACU-ASR
Pre-Forensic Automatic Speaker
Recognition for Experts
AUTOMATIC SPEAKER RECOGNITION
ACU-ASR is a voice biometric software designed to provide forensic examiners and investigators with the ability to accurately match an individual’s identity with content captured through any type of audio channel.
ACU-ASR analyses recorded speech and then compares that sample to a comprehensive speaker profile and a robust reference population to statistically assess whether an individual can be associated with that speech. Based on innovative industry-leading algorithms, this solution provides a complete set of tools to develop reference populations, build comprehensive speaker profiles, and assess the likelihood that speech of an unknown matches a target speaker profile.
Multi-language support and pre-defined reference populations segmented by dialect, gender and language expand the software’s applicability to assist in a wide range of investigations and prosecutions.
ADVANTAGES
1:1 AND 1:N ACCURATE PRE-FORENSIC SPEAKER IDENTIFICATION – associate speaker identities with recorded audio files, using cutting-edge, state-of-the-art voice biometric technology combining Gaussian Mixture Model (GMM) with powerful techniques (an iVector based system, etc.) to generate small but highly representative voiceprints.
EASILY PRODUCE A DETAILED FORENSIC REPORT containing a graphical representation of the Log Likelihood Ratio.
SUPPORT FOR ROBUST REFERENCE POPULATIONS. Reliably estimate Inter-Speaker Variation by defining a representative set of voice samples with common features (gender, language, dialect, etc.) to estimate the probability that a given voice sample belongs to a known speaker, rather than to a random individual.
VERIFY RELIABILITY WITH OBSERVED TIPPETT PLOTS. Measure the strength of voice evidence graphically with a system-generated Tippett Plot graph comparing the log likelihood ratio (LLR) and the residual error in the system. Tippett Plots represent all the certified results of previous comparisons, including matches and mis-matches, and demonstrate how strongly the present LLR supports the prosecution’s hypothesis.
SYSTEM CHECKS ENSURE QUALITY. Quickly and accurately check calibration similarity, audio quality, enrollment segment consistency, and reference population consistency to ensure quality forensic analysis.
LANGUAGE, ACCENT, TEXT & CHANNEL independent technology trained on wide range of audio sources (GSM, PSTN, VoIP, and other channels)
