Amelia Ciobanu

Electrical network signal's waveform and frequency logging for forensic

In this paper we develop a framework for collecting the ENF (electric network frequency) variations in a database, which can be used in digital media authentication and time stamping. The proposed framework allows the recording of the ENF variations and of the hum signal. The discontinuities introduced in the hum signal by the probe used for data acquisition are studied and methods for minimizing these discontinuities are proposed. In the end a method for time stamping audio recordings is used, based on the created ENF database.

Analysis of reverberation time blind estimation used in audio forensics

This paper proposes an analysis of the possibility to use the reverberation time as a reliable tool in audio forensics, for detecting audio editing traces in a recording. We begin with an optimized method for the blind estimation of the reverberation time, in the presence of additive noise and we further adapt it to better serve the forensic context. Next we present several experiments which validate, through objective measurements, the estimation method we used. Finally we propose different scenarios in which we underline the opportunity to use reverberation time as a mean of detecting audio editing in a recording. The results of our incipient study showed that, within controlled conditions, the reverberation time parameter can be successfully used for detecting the existence of audio editing traces in a recording.

Low cost electric network signal and frequency recorder

This paper presents a low cost solution for a framework used for recording the electric network signal and for logging the fundamental frequency variations. The resulted database is used for time stamping audio recordings based on the Pearson correlation coefficient. Furthermore, we propose a fast algorithm for computing the Pearson correlation coefficient by emplying filters to implement efficiently mathematical operations involved in the process. The overall cost of the proposed framework is reduced ten times than its previous version.

Fast algorithm for detecting butt-spliced edits

In this paper we conduct an in-depth analysis of a reference method used for the detection of butt-spliced edits. First, we bring forward a solution for significantly reducing the involved time resources. Next, we reveal the methods limitations in terms of variable analysis window and we test two methods for differentiating between tampered audio signals and clean signals. We also develop a relevant experimental setup which allows us to quantify for the first time the methods performances, in the context of audio forensic.

A hum removal algorithm used for audio restoration purposes

An automatic hum removal system is presented and characterized in the paper, which could be used in audio restoration applications. Its design starts from hum modeling. Two hum models were proposed, then the importance of high precision hum noise fundamental frequency estimation along with a comparative study of common estimators is presented. The problematic of filter design, represented by the choice of the quality factors depending on the spectral band in which each filter will work, is described afterwards, concluding with objective and subjective test results for the algorithm. The results clearly indicate the strong preference of the listeners for the signals resulted after the hum removal, in the high precision frequency estimation case.

Low computational method for siren detection

A driving assistance system which can detect the sound produced by sirens and warn the driver has indisputable advantages: diminishes the numbers of accidents involving emergency vehicles, saves lives by influencing the drivers to take early actions in order to clear the way for these vehicles. The United States National Highway Traffic Safety Administration reported in 2012 a number of 60 accidents involving emergency vehicles travelling with the specific signals (sounding siren, blinking lights) in use in which at least one person died preceded by 51 in 2011 and 67 in 2010. There are situations when hearing the siren plays a critical role in avoiding an accident, for example at crossroads. In such cases the visual detection of the emergency vehicle can happen too late to avoid an accident. Also there are hearing-impaired persons who drive and have difficulties in hearing the siren. Siren detection systems exist but they use complex signal processing algorithms, demanding very high processing power and, in consequence, they are expensive. The paper presents a cost effective microcontroller based siren detection system which uses analog electronics blocks to decrease the needed processing power. The detailed system along with the MATLAB and SPICE simulation results are illustrated and discussed in the paper.

Algorithm for the calculus of virtual time instants used in prosodic modifications

The topic addressed by this paper is related to a very specific operation in prosodic modifications, the calculus of the virtual time instants, for speech models based on a pitch synchronous analysis/synthesis. We propose and describe in detail a straightforward algorithm for the calculus of these moments. Also we present the results we obtained with the proposed algorithm, which show that the pitch synchronism is kept, without altering the quality of the time/pitch scaled signal.

Time-frequency processing of partials for high-quality speech synthesis

Based on the particularities offered by the chosen signal model, we introduce a novel approach regarding the chain of actions pursued in the analysis stage of the speech signal, which succeeds the level of partial extraction. According to the Harmonic plus Noise Model (HNM) a number of successive estimation and synthesis operations are performed. The present paper proposes a method to enhance the harmonic parameters estimation. This new algorithm proves to have good behavior offering support in selecting an appropriate subset of partials. In addition to reducing the arithmetic complexity of the harmonic synthesis (which is known to be the most resource consuming module), this optimized selection of the partials allows us to perform a specific frequency correction, which enables the possibility of a simple and coherent future pitch manipulation. The performed experiments confirmed the expected complexity reduction. Moreover, we applied the proposed algorithm for partial selection and tracking followed by a frequency aligning of the harmonic components. The reconstructed signal compared to the original speech proved to be a perceptually indistinguishable replica.