Stephen Lakatos

A common perceptual space for harmonic and percussive timbres

The goal of a series of listening tests was to better isolate the principal dimensions of timbre, using a wide range of timbres and converging psychophysical techniques. Expert musicians and nonmusicians rated the timbrai similarity of three sets of pitched and percussive instruments. Multidimensional scaling analyses indicated that both centroid and rise time comprise the principal acoustic factors across all stimulus sets and that musicians and nonmusicians did not differ significantly in their weighting of these factors. Clustering analyses revealed that participants also categorized percussive and, to a much lesser extent, pitched timbres according to underlying physical-acoustic commonalties. The findings demonstrate that spectral centroid and rise time represent principal perceptual dimensions of timbre, independent of musical training, but that the tendency to group timbres according to source properties increases with acoustic complexity.

The representation of auditory source characteristics: Simple geometric form

Two experiments examined listeners’ ability to discriminate the geometric shape of simple resonating bodies on the basis of their corresponding auditory attributes. In cross-modal matching tasks, subjects listened to recordings of pairs of metal bars (Experiment 1) or wooden bars (Experiment 2) struck in sequence and then selected a visual depiction of the bar cross sections that correctly represented their relative widths and heights from two opposing pairs presented on a computer screen. Multidimensional scaling solutions derived from matching scores for metal and wooden bars indicated that subjects’ performance varied directly with increasing differences in the width/height (WIH) ratios of both sets of bars. Subsequent acoustic analyses revealed that the frequency components from torsional vibrational modes and the ratios of frequencies of transverse bending modes in the bars correlated strongly with both the bars’ WIH ratios and bar coordinates in the multidimensional configurations. The results suggest that listeners can encode the auditory properties of sound sources by extracting certain invariant physical characteristics of their gross geometric properties from their acoustic behavior.

Time—distance relations in shifting attention between locations on one’s body

Does the time to discriminate the presence or absence of a tactile stimulus depend on its distance from the bodily location to which one has been attending? Subjects were tested by the presentation of air puffs at four of eight possible locations on the body. In each trial of Experiment 1, (1) subjects attended auditorily to an announcement of one of the eight locations; (2) 2 sec later, a second location was announced and, simultaneously, air puffs were presented at a randomly selected four of the eight locations; (3) the subjects reported as quickly as possible on whether or not air puffs had been presented at the second-announced location. Overall, reaction times increased with the distance between the just-attended and the tested locations. Experiment 2 was undertaken to determine whether the operative distance was distance through the subject’s body or directly through three-dimensional space. The subjects were tested as in Experiment 1, but with their arms or legs either side-by-side in front or spread out to each side. Reaction times then depended more on straight-line distance in space than on distance through the body.

Selective attention to the parameters of a physically informed sonic model

Two experiments tested listeners’ ability to attend selectively to the properties of a physical model comprising collisions between multiple independent sound-producing objects. A probe signal paradigm measured attention to two properties—resonant frequency and number of colliding objects. Listeners completed a baseline task measuring absolute sensitivity at each stimulus against a background noise. Subsequently, stimuli served as both cues and targets; cue validity was probabilistic. When cue and target were generated by the same object (Experiment 1), greater detectability occurred with valid cues for both resonant frequency and object number, implying the presence of attentional mechanisms for these properties. When cue and target were generated by different objects (Experiment 2), selective attention persisted for object number but not for resonant frequency.

Temporal constraints on apparent motion in auditory space

The hypothesis that the extent of spatial separation between successive sound events directly affects the perception of time intervals between these events was tested using an apparent motion paradigm. Subjects listened to four-tone pitch patterns whose individual tones were sounded alternately at one of two loudspeaker positions, and they adjusted the alternation rate until they could no longer distinguish the four-tone ordering of the pattern. Four horizontal and two vertical loudspeaker separations were tested. Results indicate a direct relation between horizontal separation and the critical stimulus onset asynchrony (SOA) between successive tones within a pattern. At the critical SOA, subjects reported hearing not a four-tone pattern, but two pairs of two-note groups overlapping in time. The findings are discussed in the context .of auditory spatial processing mechanisms and possible sensory-specific representational constraints.

Loudness‐independent cues to object striking force

To what degree can listeners rely on auditory cues besides loudness to correctly judge the striking forces used to excite an object? Listening experiments used a discrimination task to assess whether individuals could correctly map loudness relations onto striking forces for stimuli derived from tam–tam, tympani, and xylophone‐bar recordings. On each trial, listeners heard two sound pairs representing an instrument struck with differing strike forces using a mallet; one pair had its loudness relations intact while the other had its loudness relations inverted. Listeners determined which pair preserved its correct loudness relations. Both striking force and mallet properties served as independent variables. Discrimination varied in general with the extent to which the mallet and striking force excited the characteristic vibrational modes of the instruments: the more acoustic modes available, the higher the discrimination performance. Discrimination was significantly poorer for instruments with fewer acoustic cues generated by such modes, like the xylophone, than it was for instruments with a rich set of cues, like the tam–tam. Mallet density and weight were also important factors in modulating discrimination. The results indicate source‐based cues besides loudness can convey information about striking force, albeit with less effectiveness. [Research supported by Air Force Grant F49620‐99‐1‐0293.]

Knowledge acquisition by listeners in a source learning task using physical models

Two experiments examined listeners’ ability to learn about the acoustic properties of physical models by varying the amount of structured information listeners received about the models. In experiment 1, 60 listeners participated in a repeated learning‐testing paradigm, using a graphical interface to permit listeners to manipulate three acoustic properties for each of eight simulated percussive instruments. Information about the properties of the instruments varied across six conditions, ranging from highly structured (e.g., pictures of instruments, description of physical properties, sliders grouped by instrument with properties labeled) to weakly structured (e.g., unlabeled sliders arranged randomly without reference to instruments). Performance on discrimination and memory tasks suggests that learning can occur even with weakly structured acoustic information. To determine whether learning can occur even when acoustic information is obtained ‘‘passively,’’ 30 listeners in experiment 2 were yoked to lis...

Extended perceptual spaces for pitched and percussive timbres

A series of listening tests attempted to better isolate higher‐order perceptual dimensions of timbre by building on past findings showing that spectral centroid and rise time represent principal acoustic correlates of the primary timbral dimensions. Listeners with varying levels of musical training rated the timbral similarity of three sets of pitched and percussive instrument sounds that were equated for centroid and rise time using signal processing techniques. Multidimensional scaling analyses yielded two‐ and three‐dimensional perceptual spaces for the three stimulus sets. Several aspects of the spectral fine structure of the timbres correlated with the dimensions of these spaces, but the nature of the acoustic correlate varied somewhat depending on the stimulus set in which the timbre was presented. The results suggest that additional perceptual dimensions of timbre exist, but that their precise acoustic correlates are context dependent and therefore less perceptually ‘‘primary’’ than centroid and ri...

An interactive similarity rating program for large timbre sets

A Linux‐based program is described for collecting similarity data for large numbers of sounds in order to train an automated audio classifier. Most multidimensional‐scaling (MDS) programs require similarity ratings for all pairwise comparisons in order to compute an optimal low‐dimensional space, thereby discouraging their use for timbre sets comprising more than 25 or 30 sounds. The current program permits the testing of much larger stimulus sets (e.g., 150–200 sounds) through two principal innovations. First, the program initially assumes that the optimal perceptual space is two dimensional, and adds additional dimensions only if they are warranted. Second, it provides flexibility to listeners by allowing them to choose stimuli for comparison and to indicate their relative similarity in a highly interactive graphical format. The program has been used in listening experiments to study the perceptual criteria listeners apply in their timbral ratings for real‐world sounds, and to assess the role that menta...

Psychoacoustic evaluation of broadband and conventional reversing alarms localization accuracy

Three psychoacoustic tests assessed whether a recently developed class of vehicle reversing alarms with broadband signal characteristics would be more accurately localized than conventional equivalents. Vehicle reversing conditions were simulated by mounting sample alarms on a linear slide whose velocity and displacement could be controlled by computer. Reversals of 4 m displacement were recorded at two velocities (1 m/s and 2 m/s) and at azimuthal orientations varying in 50 increments. The first two tests presented direct hit trajectories recorded with a binaural microphone (Crown SASS‐P) and Kemar Mannequin, respectively; a third test presented near miss trajectories with the recording centerline perpendicular to the alarm trajectory and at linear displacements of 0.5 m and 1 m. Participants listened to trajectories over supra‐aural headphones and selected the closest match from computer depictions of all possible trajectories. Results for all three tests revealed no significant advantage in localization acuity for the broadband alarm.