January 18, 2024
By Pavel Jiřík in Blog
We are excited to unveil our latest version of Phonexia Voice Inspector—state-of-the-art voice comparison software specifically designed for forensic experts. Its fifth generation offers a significant leap in voice analysis accuracy, marking a significant milestone in the field of forensic voice comparison and speaker identification.
This cutting-edge voice biometrics technology can identify a person's voice in just three seconds of conversational speech with an exceptional level of accuracy, regardless of language and words spoken.
In 2021, the technology was evaluated based on the forensic_eval_01 methodology (introduced by Geoffrey and Ewald Enzingfer in 2016), achieving a 2.1% Equal Error Rate (EER) before calibration and a 1.2% EER after calibration.
Since then, our R&D has improved the speaker identification technology even further, enabling greater accuracy across a broader range of audio channels.
Meticulously Balanced Calculation of Voice Comparison Results
The Phonexia Voice Inspector's utilization of our most precise speaker identification technology can lead to an output of extremely high log-likelihood ratio scores. As a result, the discriminability of target and non-target scores in population sets can increase significantly, making the final log-likelihood ratio score used for evidence calculation very high and perhaps harder to justify in court.
Therefore, to ensure objective and unbiased analyses at all times, Phonexia Voice Inspector provides forensic experts with an option to choose between a generalized logistic distribution function or a traditional Gaussian distribution function for the calculation of a final voice comparison result.
The generalized logistic distribution function is the recommended option (used by default) as it is able to model a skewed distribution very accurately:
State-of-the-Art Supportive Technologies
Modern forensic voice analysis often needs to deal with various challenges, such as long audio files or mono audio recordings containing multiple speakers.
To make the life of forensic professionals easier, Phonexia Voice Inspector packs in a comprehensive set of advanced supportive technologies:
- Speaker diarization for effortless labeling and separating individual speakers in mono audio recordings based on their voice.
- Phoneme recognizer to help forensic experts discover similar sound patterns across audio recordings.
- Voice activity detection for the fast detection of speech and non-speech segments in audio.
- Spectrogram for further analysis of investigated audio.
Aligned with the ENFSI Guidelines
Performing forensic voice comparisons in accordance with internationally recognized standards is of utmost importance in order to provide meaningful and unbiased court-admissible reports.
Phonexia Voice Inspector aligns with the European Network of Forensic Science Institutes (ENFSI) guidelines, introduced in 2015 as Methodological Guidelines for Best Practice in Forensic Semiautomatic and Automatic Speaker Recognition.
The software results are output in a log-likelihood ratio, require population sets for unbiased analysis, and provide additional features necessary for detailed voice analysis, such as the spectrogram or phonetic recognizer.
One-Click Generation of Customizable Reports
One size does not fit all, which is why Phonexia Voice Inspector supports an easy customization of generated reports.
You can easily customize these reports to include your organization's brand, case descriptions, and forensic experts' conclusions.
Experience Phonexia Voice Inspector for Yourself
Are you ready to bring your voice comparison analysis to another level? Would you like to experience the cutting-edge features and capabilities of our latest version of Phonexia Voice Inspector for yourself? Request a live demonstration today.