Amplitude Envelope Encoding as a Feature for Temporal Information Processing in the Auditory Cortex of Squirrel Monkeys

Abstract

Auditory signals represent complex patterns of information in the frequency and intensity domain over time. How do neurons encode the complex frequency-time varying features of a natural vocalization? Where and how are the complex acoustic signals transferred and coded in the animal’s brain? Are there specialized areas for the extraction of communicatively meaningful sound parts? For the neurophysiological experimentalist the situation seems at first sight frustrating. The answers obtained with single unit recordings are only simple yes/no codes (i.e, action potentials). First, the interpretation of the action-potential frequencies with respect to different complex stimuli seemed to be impossible at the single neuron level. Therefore, when the first monkey studies started, the only promising concept seemed the search for highly specialized cortical neurons which may act as “vocalization detectors”. Such neurons should answer only to a very small set of vocalizations but not — or not in the same way — to noise or tone bursts. Neurons with such feature specificity have never been clearly identified. Even complex vocalizations appeared to evoke neural responses only by producing spike trains following the time structure of the transient elements in the stimulus (Newman, 1988). Consequently, it is not surprising that previous electrophysiological research with species-specific sounds gave rise to more questions than answers. Moreover, the theoretical background for these experiments was poorly described. Highly specialized “vocalization detector” neurons are only to be expected if one assumes one of the following mechanisms: an innate neural network specialized for filtering the species’ own vocalizations, or, alternatively, a sensory serial information processor leading to the extraction of information-bearing sounds. Both mechanisms could produce higher-order neurons exclusively restricted in their activity to vocalizations as adequate stimuli. An argument against a simple innate neural network comes from behavioral studies. The meaning of an animal’s vocalization can vary with context, even though the acoustic morphology of the sound remains unchanged (Symmes, 1981; Cheney and Seyfarth, 1990). An argument against the second alternative is the anatomically established centrifugal pathways which parallel the centripetal auditory pathways (Pickles, 1982). On the other hand, electrophysiologists have documented several neurons that are, for example, specialized to small intensity or frequency ranges. This can be interpreted as support for the view that neurons are able to extract features, which may be information-bearing parameters in a complex stimulus. Such feature encoding seems an essential framework for the brain in complex sound recognition. To offer an example, it is an every day experience that in speech perception the meaning of a word is often known despite an unclear pronunciation or against a noisy environment. We complete a familiar word if we detect some features or meaningful parts of the word. Therefore, many studies regarding the auditory perception of sounds focus on experiments with relatively simple artificial sounds, which may act as information-bearing elements in a complex sound.