Upon hearing objects collide, humans can estimate physical attributes such as material and mass. Although the physics of sound generation is well established, the inverse problem that listeners solve – of inferring physical parameters from sound – remains poorly understood. Here we show that human perceptual inferences are enabled by intuitive knowledge of acoustical physics, allowing listeners to disambiguate multiple object properties from a single impact sound. To study this problem, we built a generative model of impact sounds, combining theoretical acoustics with statistics of object resonances measured from hundreds of everyday objects, and used it to synthesize and manipulate experimental stimuli. Humans accurately judged object properties from sound alone, even when both of the underlying objects varied. However, performance was impaired if simulated physical regularities deviated from those measured in real-world objects. The results suggest that listeners use internal physical models to separate the acoustic contributions of objects in the world
This web page contains examples of the stimuli used in the associated paper. All of the sounds provided here represent a small, hard ball impacting objects made of different materials. Because the ball is small relative to the impacted object, we assume the ball itself radiates no sound.
These demos should be listened to over headphones to ensure that the relevant audio properties are audible. Note also that some sound presentation devices may not faithfully convey the sounds, for instance if they have poor reproduction of low frequencies. Use wired headphones for best results.
Impulse responses are drawn from the Small size category (10-30 cm) of our impulse response survey
Sounds were synthesized assuming that the impacting ball had a mass of 50 gm. The equivalent stiffness was set to theoretically derived values used in Experiment 1 (see paper for details).
Impulse responses are sampled from the Small size category (10-30 cm) distributions
Sounds were synthesized assuming that the impacting ball had a mass of 50 gm. The equivalent stiffness was set to theoretically derived values used in Experiment 1 (see paper for details).
These example sounds are synthesized from a recorded impulse response of a wood object, for different simulated ball masses. The sound level increases with mass, but the balance of high to low frequencies also shifts. To make this apparent we have included two sets of examples, with the second normalized so that all sounds have the same root-mean-square level. Sounds were synthesized with a recorded impulse response (wood) and the equivalent stiffness set to 10^7 N/m.
These example sounds illustrate the experiment stimuli used to assess the perception of damping (Experiment 4 in the paper). We found in acoustic experiments that external damping (e.g. holding a mug in your hand vs. letting it rest freely) primarily affects the decay time of an object's resonant modes. We thus synthesized sounds with different degrees of object damping by scaling the decay times of the impulse response modes. These examples present impact sounds from two different impulse responses sampled from our full generative model, with different degrees of simulated damping. We quantified simulated damping with a damping coefficient, given in parentheses in the labels below (see paper for details).
These example sounds illustrate the intact and lesioned generative models used in Experiments 1-4 of the paper. It is difficult to summarize the distributions defined by each model with a small set of randomly drawn samples, so we have hand-picked single examples that to our ears contain some of the distinctive features of the lesioned models. The examples are all drawn from the distributions for Wood objects in the Medium size category. Sounds were synthesized assuming that the impacting ball had a mass of 50 gm. The equivalent stiffness was set to theoretically derived values used in Experiment 1 (see paper for details).