Parameters of Echoic Memory

Since the days of the multiple-components theory of memory it has become common practice to characterize and/or differentiate memory systems by specifying their lifetime, capacity and susceptibility to interference. Periodic noise was used to study these parameter of echoic memory for random waveforms and to compare them to those of short-term memory. In a first experiment, it was examined up to which cycle length it is possible to perceive the periodicity. With only a small amount of training periodicities as long as 10 s could be detected. In two further experiments it was quantified how much of each cycle was memorized in order to detect the periodicity. Independently of the cycle length, a surprisingly small amount of the cycle, on average 130 ms, served as a cue. The fourth experiment demonstrated that echoic memory is to a certain degree protected against interference. The similarity of the parameters of echoic memory to those of short-term memory strengthens the view of echoic memory as a modality-specific module of short-term memory. It is now well established that storage of sensory information is at least two-fold. Massaro and Loftus (1996) differentiate sensory and perceptual storage, with the latter lasting much longer than the former. The first classification into short and long sensory stores was done by Cowan (1984) in the auditory realm. He reviews many studies relevant to auditory memory and classifies them into two groups: those revealing time constants of 200 ms or less, and those revealing time constants of 10 to 20 s. For the short store Cowan cites data from masking experiments, auditory persistence, and temporal integration. These phenomena form part of sensation. The long store accounts for phenomena of up to 20 s, which are perceived as memory. The auditory partial report falls into this category as well as dichotic listening experiments and the perception of periodic random waveforms (Guttman and Julesz, 1963). According to Cowan (1988, 1995) long sensory storage and short-term memory are both activated parts of long-term memory. While the lifetime of traces in long sensory stores is compatible with the assumption of a close relation to short-term memory, the capacity of sensory stores is often thought of as being much higher than that of short-term memory, and sensory stores are thought of as being much more susceptible to interference. While this is true for short sensory stores, it needs not be so for long sensory stores. The goal of the present study was to examine these three parameters for the long auditory store using a single class of stimuli, by this means avoiding unjustified synopses across tasks and material. Periodic random waveforms represent an excellent test of auditory sensory memory. They do not offer clues to categorical storage such as most other auditory materials do. The stimulus is produced by seamlessly connecting repetitions of a single segment of white noise. At the connection points, no artifacts are introduced that could give rise to clicks or other artificial percepts. However, even naive listeners perceive a striking difference between periodic and continuous noise: periodic noise is perceived as rhythmically structured, and filled with perceptual events such as “clanks” and “rasping”. For the same noise segment, these perceptions are reproducible across different sessions of the same listener, and to a lesser degree correlated across listeners (Kaernbach, 1992). The temporal extent of the basis of these perceptual events is restricted to about 100 ms (Kaernbach, 1993). Periodic noise has been used as signal in masking experiments (Pollack, 1990), its perception has been compared with pitch perception (Warren & Bashford, 1981; Warren & Wrightson, 1981; Warren, Wrightson, & Puretz, 1988) and it has been used to study time order processing (Warren & Bashford, 1993). For a review on periodic noise research see Warren (1998). A demonstration of periodic noise stimuli can be found at www.periodic-noise.de Experiment 1. Lifetime Guttman and Julesz (1963) reported that for periods longer than 2 seconds periodicity detection would become difficult. Cowan, however, ascribes periodic noise perception to the long auditory store (10 to 20 s). Warren, Bashford, Cooley and Brubaker (2001) showed that cross-modal cueing helps experienced listeners to detect periodicity in cycles up to 20 s. In pilot studies it became obvious that only a small amount of training is needed for naive listeners to perceive long cycles. Experiment 1 was conducted to quantify the relation between training and maximum cycle length in naive listeners. The noise was generated as a sequence of Gaussian random numbers with a standard deviation of 10% of the conversion range. These were converted at 20 kHz and presented via headphones at 60 dB hearing level. To make this periodic, the random number sequence was recycled. For each of the 20 naive participants and each single trial a different noise sample was generated. The cycle lengths ranged from 0.5 to 20 s. The participants started the experiment without practice. The trial started when the participant hit the space bar. Participants were instructed to tap the space bar once per period. If the participant did not start tapping to a noise sample, this trial was considered a failure and the next trial started. Participants were randomly assigned to one of two groups. Group A passed through the cycle lengths in ascending order, group B in descending order. From trial to trial the cycle length was increased or decreased regardless of the success of the previous trial. All participants performed 3 blocks. From the obtained tapping data it was determined whether the participant had perceived the correct periodicity. The results are presented in Figure 1, summed over all participants and monotonized. Monotonizing was applied to all points except the first point of the first block of group A. The low performance here is due to the fact that it was the very first trial of the experiment so the participants were not well prepared for what they were to hear. This clearly demonstrates that the participants were naive. In the first block of group A, cycles up to 2.8 s are correctly tapped by half of the participants. But also cycles of lengths of 10-20 s can be detected, despite practice being limited to the presentation of previous, shorter cycles. Due to the effects of practice, in the third block cycles of 7 s are correctly tapped by more than 50% of the participants. Group B shows similar data. There is a smaller training effect. Apparently, descending periodicities are less efficient for training participants on periodicity detection than ascending periodicities. It is all the more remarkable that two participants of group B were able to tap correctly with cycles as long as 12 s in the first block. Their only training had consisted in listening for six minutes to even longer cycles in which they did not notice any periodicity. For comparison, Figure 1 shows data by Peterson and Peterson (1959) on the retention of consonant trigrams as a function of time (black diamonds, cf. Discussion).