Our activities

SIVE Speaker Identification and Verification Software Package

Speaker Identification and Verification Software Package (SIVE) is designed for Forensic voice comparison (Forensic Speaker Identification). The purpose of SIVE is to define and compare the voice parameters in objective way, to display a statistical distribution of the relevant parameters in diagrams and to calculate correlation coefficients between these diagrams. It also provides the forensic expert with a visual aid to prove the decision. It makes easier and faster to draw the objective conclusions in comparison with the traditional sonographic method. The analysis becomes independent from the text.

SIVE package comprises from two parts, which can be used independently of each other:

– SIVE BASE is a software package consisting of the separate modules for the semi-automatic speech signal analysis using the different methods.

– SIVE VOICE is a software package for the automated voice searching and comparison using the Likelihood Ratio based approach.

Methods of speech analysis used in SIVE BASE software package

Voiced speech signals energetically are the strongest and their energy is concentrated in the lower frequencies, that’s why they are least influenced by the noise and the transmission channel. Voiced speech signal is a convolution of the excitation signal (or the pitch) and its multiple harmonics and the impulse response of the speech tract. The linear mathematical model may describe such type of signals.

The pitch is one of the parameters of voiced signals, which has an important identification value, and is least dependent upon the quality of the recording conditions and channel. SIVE package uses a frequencyautocorrelation method for pitch estimation. Due to the physical differences in the specific features of the human speech tract, there are many harmonics of the pitch and their amplitudes and they go down sooner or later. That’s why additionally to the pitch estimation, there are calculated such derivative parameters: the highest harmonic of the pitch (MH), voice clearness (VCL) and timbre (T). The comparison of the pitch parameters of two different speech signals is implemented in SIVE package. The results of the analysis are presented as a list of the minimum, average and maximum values of parameters, their variance and variation coefficients, distribution diagrams and correlation coefficients, and a final coincidence coefficient of the pitch parameters.

If the good quality speech records are submitted for the analysis, but having the insufficient length, then it is possible to use the method of evaluation of the relative distances between the phonemes. This method is based on assumption, that having two phonemes spoken by the same person (e.g. “A”) and performing the identification according to the first four formants, depending on the pronunciation of the sound (especially first two) and the special features of the speakers voice tract (especially third, fourth, fifth), the relative distance should be the smallest.

Module for Speaker Verification is used for text-independent and text-dependent cases.In text-independent cases the requirement for the duration of the speech segment is 10-20 seconds. Software is working in two stages: training and verification (identification). For training 3 different speech segments are needed – not shorter than 10 seconds.

In text-dependent cases, Speaker Verification software can be successfully used for Voice identification, when the recordings are short (at least 5 words), but with one condition: both tested and voice print records must have the same phrase.

Automated Voice searching and Forensic Voice comparison system SIVE VOICE essentially is designed for the telephone quality signals (mobile phone conversation recordings) and acts as follows. First, the expert analyses the investigative recording and distinguishes (segments) the identifiable person’s speech example sound recording (20-60 seconds). Then the system is learning – this distinguished record is compared against all records of known voice recordings that are in the reference speech database. Next step – system compares all voices from the reference speech database with the investigative recordings and provides the alignment results in Bayesian metric LLR form, in descending order of LLR. Now the expert should check only those records, that have obtained LLR is in the range from -6 to +2 or to check those that are found in the first top ten (10)

SIVE international recognition

The first SIVE version was created in 1995. Reliability of SIVE was validated using TWINS voice base, imitative voice ENFSI Forensic Speech and Audio Work Group test (Fake Case) and doing thousands forensic speaker recognition examinations.

In year 2010 the main SIVE methods were tested in the NIST 2010 Speaker recognition test and the tests results were positive.

In year 2012 SIVE VOICE method ( Automated voice searching and voice comparison ) was tested in ENFSI FSAAWG test – FORENSIC SPEAKER RECOGNITION using a DATABASE of REAL CASES. SIVE VOICE result ( EER=0,07 and C_LLR=0,17 ) was the best.

In year 2015 SIVE program package was included in ENFSI METHODOLOGICAL GUIDELINES FOR BEST PRACTICE in FORENSIC SEMIAUTOMATIC and AUTOMATIC SPEAKER RECOGNITION as one of the software package for Forensic Semiautomatic Speaker Recognition.

Currently SIVE software package is used for Forensic Speaker Identification in 11 countries: Finland, Estonia, Latvia, Lithuania, Poland, Czech, Greece, Portugal, Romania, Ukraine and Croatia.

RIEGL, based in Horn / Austria, is a leading global supplier of 3D laser scanners and scanning systems for terrestrial, industrial, mobile, airborne and UAS applications.

With more than 40 years of experience in the research, development and production of laser rangefinders, distance meters, and scanners RIEGL delivers proven innovations in 3D. The company has been leading the market in pulsed time-of-flight laser radar technology in multiple wavelengths.

The combination of RIEGL‘s state-of-the-art hardware for terrestrial, industrial, mobile, airborne, bathymetric and UAV-based laser scanning with appropriate, equally innovative RIEGL software packages for data acquisition and processing results in powerful solutions for multiple fields of application in surveying.

RIEGL has always been committed to delivering the highest performance, quality, reliability, and longevity of all its products and services, and strict adherence to applicable international standards is a priority.

Net Insight is a global market leader in media networks and resource optimization. Net Insight provides the world’s most efficient media transport solution for Broadcast, Media, Digital Terrestrial Television, Mobile TV, IPTV, CATV with 100% Quality of service and network solutions, based on the NIMBRA MSR (Media Switch Routers) platform, to service providers and businesses for high-performance real-time IP networking needs.

NIMBRA equipment is used by Telcos, telecommunications companies, broadcasters and high-level media companies for the transport and distribution of high-quality video content in combination with high-performance data services. With more than 20 years of experience, Net Insight is a key partner for its customers and a leader in the ever-changing media market.