RU150244U1 - PHONOMIC AUTHENTICITY DETERMINATION SYSTEM - Google Patents

Abstract

1. A system for determining the authenticity of phonograms, including a module for inputting an audio signal, which is connected to a module for analog-to-digital conversion, which is connected to a module for determining the authenticity of phonograms, which performs multilevel wavelet transform of the signal and visualization of the analysis results. 2. The system of claim 1, wherein the audio input module is a microphone for recording voice signals of an identifiable speaker. The system of claim 1, wherein the audio input module is a telephone channel of a public network. The system of claim 1, wherein the audio signal input module is a cellular communication channel. The system of claim 1, wherein the audio signal input module is an analog magnetic recording reproducing device.

Description

Phonogram authentication system

The utility model relates to tools for the forensic investigation of sound information or phonoscopic examination of audio recordings, and more specifically to a system for determining the authenticity of phonograms using high-precision multi-level wavelet analysis of speech signals.

Phonoscopic examination is carried out in order to establish the identity of the speaker on the basis of voice and speech recorded on the phonogram; identifying signs of editing and other changes made to the phonogram during or after the end of sound recording; determination of the conditions, circumstances, means and materials of sound recording, as well as other facts of relevance to judicial evidence.

Especially acute in recent years in the practice of phonoscopists is the problem of detecting traces of violations of the reliability of phonograms. The development and widespread use of computer tools for processing and editing audio recordings, the availability of detailed information on the implementation of such actions, has led to a situation where the creation of a mounted phonogram with modified replica texts, the composition of speakers and the background acoustic environment introduced during installation is a relatively simple task, even for lay people. At the same time, digital phonograms are increasingly being attached to criminal and civil cases in domestic judicial practice, since, in accordance with the Civil Procedure Code and the Code of Criminal Procedure, audio and video recordings are included in the list of admissible evidence (Article 55 of the Code of Civil Procedure of the Russian Federation and Article 84 of the Code of Criminal Procedure RF). In this regard, the creation of new high-precision tools for the analysis of audio recordings that provide reliable determination of the authenticity of phonograms is an urgent scientific and technical task.

The prior art systems for detecting signs of mounting phonograms, which are based on random or special added constant background noise, not tied to the original structure of the controlled speech signal.

A known method of recording information on a medium, in which before recording the main and control data in the form of frames on the recording medium, a control data file is formed in the form of N-identical groups in each data packet (RF patent No. 2073913, 02.20.97).

The method is aimed at transferring the processing of recorded data from specialized material to software.

There is also a method of placing on the media verification characters for correcting errors in the information data, so that each component of the information and verification characters are recorded at intervals of each sector to disperse the above verification characters in each sector (RF patent No. 2154897, 08.20.00) .

This patent solves the problem of interference-free encoding of data. If failures occur when encoding a transmission or reading this data, they are corrected using the method proposed in this patent.

RF patent No. 2095857 (10.11.97) describes a method for transmitting audio and / or video information using a data carrier, in which data is formatted with digital sub-code data added to the main information, which is divided into information packets before formatting and interleaved. After reformatting, the selected packets are extracted from the subcode information.

The method according to this patent provides for the use of additional tracks of digital recordings with multilingual texts. This additional data is not mixed with the desired signal.

The disadvantage of the above methods is that the marking signals used in them can be quite simply identified, deleted, replaced or added during manipulations with the mounted phonogram using digital or analog recording techniques. In the case of a "watermark" during recording control, it is necessary to constantly store the reference marking signal.

The prior art system for detecting signs of mounting phonograms, "tied" to the original structure of the monitored speech signal (№ patent No. 082728, 08/29/2002. Eurasian patent organization). The present invention consists in creating a method for generating a phonogram of voice information recorded on a medium, which, when changing the recording, allows you to detect any corrections or editing of the phonogram, and if not available, confirm the authenticity of the original recording. In other words, the task of determining the authenticity of a phonogram is solved by introducing into the main speech signal a control marking signal modulated by the spectral power of the frames of the sonophile over harmonics with a random distribution of their phases. As a result, the phonogram signal with traces of speech is marked with a special additional signal structure located in the same frequency range as the speech signal itself, which does not allow changing or deleting fragments of the phonogram without a trace. A method for generating a phonogram of speech information is that voice and control marking signals are simultaneously recorded on a medium to determine the integrity of the phonogram based on the results of its time-frequency visualization (sonofilm).

The disadvantages of the considered methods for determining the authenticity of phonograms include the need at the stage of recording the speech of the speakers of the formation and recording of special marking signals.

In addition, IkarLab, Justiphone, OTExpert sound and software complexes using Fast Fourier Transform (FFT) algorithms are known from the prior art.

Known from the prior art, hardware and software for forensic investigation of phonograms use the Fourier transform, which does not provide the necessary resolution when the process parameters change over time (non-stationarity) because it gives average coefficients for the entire signal under study.

The claimed system for determining the authenticity of phonograms differs from the foregoing in that it does not require the recording of special markers at the stage of recording voice signals; the facts of editing are detected by means of a high-precision multi-level wavelet transform of the registered signal, isolation of the background signals of the power supply network and its low-level harmonics.

The utility model is the development of tools for forensic investigation of phonograms based on the new technology of multilevel wavelet analysis of signals with an increased frequency-time resolution.

The technical result, which the utility model aims to achieve, is to increase the resolution of the time-frequency analysis of audio signals in a given frequency band.

The technical result is achieved due to the fact that the system for determining the authenticity of phonograms includes an audio signal input module that is connected to a module for analog-to-digital conversion, which is connected to a phonogram authentication module that performs multilevel wavelet signal conversion and visualization of analysis results.

The figure 1 shows a structural diagram of a system for determining the authenticity of phonograms.

The figure 2 presents the appearance of a four-channel module analog-to-digital conversion of audio signals N1 9234 (National Instruments, USA); 24 bits, 51.2 kS / s, dynamic range 102 dB.

Figures 3-7 show examples of revealing the facts of editing phonograms using a multilevel wavelet analysis.

The system for determining the authenticity of phonograms (Fig. 1) includes a module for inputting an audio signal (1), a module for analog-to-digital conversion (2), and a module for determining the authenticity of phonograms (3).

The audio input module can be an analog magnetic recording device, a public telephone channel, a GSM standard cellular channel, and a microphone for recording voice signals of an identifiable speaker.

The analog-to-digital conversion module performs analog-to-digital conversion of phonogram signals. Figure 2 shows an example of an ADC module that can receive and process signals from four different audio input modules at the same time.

The phonogram authenticity determination module performs multilevel wavelet transform of the selected signal section and visualization of the analysis results.

To perform an analysis of a non-stationary process (for example, for an acoustic signal), it is necessary to use basic functions that have the ability to detect both frequency and time characteristics in the analyzed signal. In other words, the functions themselves must have the properties of time-frequency localization. To a greater extent, wavelets meet the above requirements. Wavelet spectrograms are much more informative than conventional Fourier spectrograms, and unlike the latter, they reveal the finest local features of acoustic signals.

The result of the wavelet transform of some signal will be the convolution of this signal with the basic wavelet function. As a basic wavelet function, when receiving a time-frequency representation of a speech signal, for example, Morlet wavelet can be selected. Wavelet spectrograms (the result of the transformation) can be obtained using sets of low-frequency and high-frequency discrete filters (their number depends on the number of levels of the wavelet transform). Discrete filters can be implemented on the FPGA, visualization of the results can be carried out on a liquid crystal display. Ways to visualize this information may be different. The most widespread is the representation of the wavelet transform spectrum in the form of a projection onto the “time-frequency plane” with contours that allow us to trace changes in the intensity of the wavelet transform amplitudes at different scales and in time, as well as patterns of lines of local extrema of these surfaces that clearly reveal the structure of the analyzed process .

When the “background” signal of the 50 Hz power supply and its harmonics are extracted, the frequency range of 40–200 Hz is set, the analysis interval is 1 second.

Based on the results of a high-precision time-frequency analysis, presented in the form of visualized wavelet sonograms, a decision on the integrity of a phonogram is made by an expert phonoscopist. Figure 3 (phonogram 1) shows a wavelet sonogram of an audio recording with two editing sections (corresponding to a break in the signal of the "background" of the 50 Hz power supply network). The speaker’s speech was recorded indoors and outdoors. When registering a speech signal in the room, a digital Labtec Verse-704 USB microphone was used, sampling frequency 8 kHz, 16 bits; two inserts of a speech signal are registered using the Gnome-2M voice recorder. Recording is processed by an equalizer, frequency suppression in a band of 0-100 Hz - 60 dB.

In figure 4 (phonogram 2) is a wavelet sonogram of an audio recording with a editing section (in the center). The low-frequency part of the sonogram of the “insertion” of the speech signal recorded in the room contains “traces” of the “background” of the 50 Hz power supply and its harmonics. Recording of the speaker’s speech was performed on the street, “insertion” - indoors. When recording a speech signal on the street, the OLYMPUS DS-2000 Voice Recorder was used, in the room - a Labtec Verse-704 USB digital microphone, a sampling frequency of 44.1 kHz, 16 bits. Recording is processed by an equalizer, frequency suppression in a band of 0-100 Hz - 60 dB.

In figure 5 (phonogram 3) is a wavelet-sonogram of an audio recording with a transition point: you can see the gap signal "background" of the power supply 50 Hz - 442.5 seconds; analyzed frequency band 5 Hz - 100 Hz.

In figure 6 (phonogram 4) is a wavelet-sonogram of an audio recording of a connection establishment signal in a PSTN channel: tone sending of a “call” and the beginning of a voice message. The duration of the tone call "call" is 0.768 seconds. (start - 6.51 sec., end - 7.278 sec.), which is 0.232 sec. less than its standard length of 1 second. At the end of the tone parcel, a section of the transition is revealed; signal in the interval of 7.22 seconds. - 7.278 seconds It is not its true continuation, but “docked” with a tone and voice message taken from another communication session.

In figure 7 (phonogram 5) is a wavelet-sonogram of an audio recording with a editing section: discontinuities in the “background” signal of a 50 Hz power network are visible.

Thus, the use of a phonogram authenticity determination module that performs multilevel wavelet transform of a signal and visualization of analysis results makes it possible to increase the resolution of the time-frequency analysis of audio recording signals in a given frequency band, which makes it possible to more accurately determine the editing areas (integrity violations) of a phonogram.

Claims (5)

1. The system for determining the authenticity of phonograms, including a module for inputting an audio signal, which is connected to a module for analog-to-digital conversion, which is connected to a module for determining the authenticity of phonograms, which performs multilevel wavelet transform of the signal and visualization of the analysis results. 2. The system of claim 1, wherein the audio input module is a microphone for recording voice signals of an identifiable speaker. 3. The system of claim 1, wherein the audio input module is a telephone channel of a public network. 4. The system of claim 1, wherein the audio signal input module is a cellular communication channel. 5. The system of claim 1, wherein the audio signal input module is an analog magnetic recording reproducing device.

Images