An Associative Memory for Auditory Recall

A model of auditory temporal reca l l based on correlographic recording and Interrogation by convolution is presented. Complex tree and loop structured memory sequences can be recorded, with selective reca l l based on fragmentary cues. The sequences are " re membered" by proceeding through a recorded net of associations u n t i l the coherence of the retr ieved signals f a l l s to a su f f i c ien t l y low level . This indicates no further s igni f icant reca l l is possible without an addit ional external cue. A computer simulation of this model using d ig i t ised speech has produced variable-length and cycl ic sound-segment reca l l from fragmentary cues. Introduction. The problem of recording information in a manner so as to associate it with other recently received information and/or old information recalled by the present environment, appears to be at least a prerequisite to I n t e l l igent behavior. * We are concerned here with presenting a model of paired-associate recording and recal l that may have the necessary potent ia l . The poss ib i l i t y of reca l l based upon holographic techniques was presented by Stroke (1) as a compensation or f i l t e r e f fect , and later by Col l ier and Pennington (2) as reca l l based upon fragments (possibly displaced) of the o r ig inal scene. This ver i f ied Van Heerden's (3) ear l ier predict ion made on a theoretical basis. The experimental setup for displaying this phenomena is shown In Figures 1 and 2. An interest ing poss ib i l i t y of using this as a model of an associative memory comes from the fact that mult iple exposures can be made on the same hologram and, af ter development, i l luminat ion by a fragment from one of the scenes w i l l reca l l that or ig ina l scene provided su i t able circumstances are insured. In the fol lowing discussion we shal l be pr imar i ly concerned with auditory signals rather than v isual , but thay are generally Just a res t r i c t i on to one dimension rather than two. The poss ib i l i t y of recording the auditory analog to Figures 1 and 2 was suggested by Longuet-Hlggins (4) and extended to a neural model by Chopping (5). In the auditory model a given time signal is considered to be made up of two successive fragments, a( t ) and b ( t ) . Then the magnitude of the power spectral density is