Solving a Crime using Java,
Signal Processing, and Sound Source Localization
SPARCS Research Group1
Department of Computer Science
University of Victoria
Victoria, B.C. V8W 3P6
jmurdoch@cs.uvic.ca
Foreward:
Motivated by the popularity of the "CSI: Crime Scene Investigation" T.V. drama, the success of the Media Computation approach to introductory Computer Science education, as well as numerous statistics citing forensics as a field in which women predominate, we use a CSI-inspired scenario to inspire first-year (CS1) students' learning of core concepts. Simultaneously, students explore a real-world application of Computer Science while advancing scientific/analytical thinking and interdisciplinary problem-solving skills.
First-year and introductory courses in Computer Science (CS) represent a critical interface between today's youth and the dynamic and innovative pool of IT professionals affiliated with science, industry, and academia. Professionals today work in highly interdisciplinary environments where specialized domain knowledge must intersect rigourous software development practices. Scientists trained in the traditional physical sciences are increasingly becoming knowledgeable about and exploiting the versatility of technology to solve problems, but the benefits are greatest when they team with software development experts skilled in the development, rigourous testing, and deployment of innovative solutions. The opportunities for computer scientists within traditional scientific domains are growing wider every day, and students will therefore benefit from curriculum that identifies the potential of software development to provide tools for the empirical investigation of our physical world. At the same time, students become motivated by curricula which explore issues and real-world applications of Computer Science that are relevant to them and to the current trends of technical innovation in society. Finally, motivating curricula have the potential to dissolve stereotypes and perceived barriers, allowing more students to identify with and appreciate the field of Computer Science.
This website describes and provides instructional materials for an item of innovative curriculum developed at the University of Victoria. The "CSI: Audio Forensics project" illustrates a real-world application of core learning objectives. Simultaneously, a connection is made between the development of software tools, and scientific and empirical analysis of these tools and their use to solve a real-world problem relevant to the students of today.
Introduction and Impact:
Audio forensics refers to the reconstruction and retrieval of evidence related to a crime through the analysis of audio recordings. Inspired by the popular television program CSI: Crime Scene Investigation, this project will have students reconstruct the scene of a crime through the use of the Java language and fundamental digital signal processing techniques.
As audio analysis software advances, so too has the potential for crime investigators to extract important clues from audio-recorded evidence. Acoustical gunshot analysis, audio enhancement, voiceprint analysis, voice elimination, and acoustic simulation for accident reconstruction are some of the current applications for state-of-the-art audio analysis and processing techniques.
 |
Apart from the perceptible content of audio recordings, more subtle clues may be hidden in the timing of events recorded by multiple microphones. If, for instance, a sound (or "acoustic event") is recorded by several microphones placed at known locations around the periphery of a room, as shown in Figure 1, a time disparity between the simultaneous recordings of the event will exist depending upon the position of the sound source relative to each mic. That is, a sound will be recorded first by the microphone closest to the source of the sound, and last by the microphone furthest from the sound source due to the finite speed of sound (approx. 340 m/s at sea level).
If the time disparity, known as the Time Difference Of Arrival (TDOA), between pairs of microphones is computed, it is possible to estimate the position of the sound source. This process of determining position based on sensor readings (in this case microphone recordings) is similar in concept to position 'triangulation', and the position tracking performed by Global Positioning Systems (GPS). Finally, estimating the position of a moving sound source over time may be used to reconstruct the path taken by the (noisy!) perpetrator of a crime through a crime scene.
Hands-On Project Overview:
Aided by a simple software framework for digital audio manipulation, students will use the Java programming language to reconstruct the movements of two suspects in a bank robbery. Students will implement two algorithms for computing the TDOA between simultaneous digital audio recordings. The first (simpler) algorithm is threshold-based and is limited in its ability to properly distinguish an acoustic event from background noise. The second (more complex) cross-correlation-based algorithm performs better but with longer running times. Students will compare their algorithms to reveal the tradeoffs between algorithm complexity, run-time, and computation accuracy. In the process, students will gain an understanding of 1- and 2-dimensional arrays, and will be exposed to fundamental concepts of multimedia and digital signal processing, such as analog-to-digital conversion and digital audio representation within computer memory.
Students will then be introduced to the concept of position localization using sound TDOA values. A Java class to perform the localization computations based on TDOA values is provided, and students must refer to Java API documentation to learn the usage of class methods. A trajectory over time for two crime suspects will be computed, and the trajectories visualized graphically using another provided Java class. Students will then investigate the relationship between the accuracy of computed position coordinates and the level of noise present in the microphone recordings. Students will learn the concepts of signal-to-noise ratio and the statistical measure of relative error. Finally, students will support the guilt or innocence of the two crime suspects with scientific data derived from their Java programs!
|
1SPARCS (Solving Problems with Algorithms Robotics and ComputerS) strives to enlighten and inform students from grade school to first year university about the possibilities within Computer Science at the University of Victoria and beyond, using fun and innovative methods while contributing original research to the Computer Science Education community. For more information about programs run by SPARCS, please email us. |
 TM |
