Impact Sound Based Identification Of Sound Source Physical Attributes And Acoustic Cues Extraction

In acoustic target recognition, how to extract effective features to improve therecognition rate had been the focus of researches. For a long time, researchers mainly use twofeature extraction methods. One is focusing on the time domain, frequency domain andnonlinear features based on signal processing and transform. And the other is mainly aboutthe auditory perceptual features (such as loudness, pitch and timbre) through the simulation ofthe auditory system. Although these two kinds of features can provide with the description ofthe sounds from different angles, they did not establish direct contact with the sound source.Therefore, the recognition objects could only be "voice" instead of "source", which affects theacuracy of the target perception ability. In recent years, the sound source identification hasgiven us a new feature extraction method which extracting information related to physicalproperties of a sound source, such as material, size and shape. The features obtained by thismethod had explicit physical meaning and more meticulous description of the sound source,so they were help to improve the ability of acoustic target automatic recognition.This dissertation focused on relationships between sound source, sound signal and theauditory perception. Based on the synthetic impact sound and subjective evaluationexperiments, we simulated the impact vibration and sound radiation in time-domain, andexplored the performances and the information integration characteristics of the auditorysystem when identifying materials and size of sound sources, and summarized the constantacoustic cues for material identification finaly. The main contents included:A ball-plate collision model and corresponding fast computing method were given intime-domain. Firstly, a hybrid method combined finite-difference time-domain (FDTD) withmodal expansion method (MEM) was used to solve the vibration equations of the plate.Secondly, some key issues for the hybrid method were considered, including modal truncationand consistency of different damping representation between the two methods. Finally, theefficiency and effectiveness of the method was verified by simulation and experiment.In order to simulate the structural vibration and sound field in time-domain, we discussedthe theoretical calculation and experimental measurement for the structural modal damping,and provided an integrated numerical method for impact sound synthesis based on finiteelement method (FEM), MEM and boundary element method (BEM). Results proved thevalidity of such an integrated numerical method with basically the same temporal envelope,spectral structure and decay trend in both recordings and synthetic sounds.A series of subjective evaluation experiment were designed and implemented to investigate the ability of the auditory system to identify the sound sources. Firstly, weanalyzed the effects of three different sounds, live sounds, recordings and synthetic sounds, onmaterial identification and their corresponding physical mechanism. Secondly, we exploredthe effects of different size and boundary conditions on material identification, and reportedthe performance of identifying the plate size under synthetic sounds and recordings.The sound continua were synthesized with varied material properties throughinterpolation functions among aluminum, glass and wood. We obtained the perceptual spaceof materials by use of the similarity evaluation experiment and multidimensional scaling andacquired mechanical parameters characterizations of plate materials through the analysis ofthe relationship between space perception and mechanical parameters. Then, a series ofacoustic features were extracted respectively from time-domain, frequency-domain andauditory perception, and the concept of information accuracy and perceived weight wereintroduced to quantify the capacity of sound characteristics which described material changesfrom subjective and objective perspectives respectively. Finally, we analyzed the role ofinformation accuracy and perceived weight in automatic recognition of materials, anddiscussed the information integration strategy in sound source material perception.In the identification of impacted plate sizes, we designed dissimilary evaluationexperiments for aluminum and wood plates, and determined mechanical parameterscharacterization. Afterwards, we reported the information accuracy and perceptual weightingof different sound features, and provided with reliable acoustic cues in size perceptionstrategies and the strategies listeners used in the process.In the material identification of complex structure, we compared the informationaccuracy of sound features for flat, stiffened plates and cylindrical shells, and provided withconstant sound cues for material identification. Both results of information accuracy andperceptual weighting showed that: During the auditory perception, listeners would like tocomplete the perception task on bases of constant acoustic cues not acoustic informationwhich were easily affected by source types.