31: The Auditory Central Nervous System

• Multiple Types of Information Are Present in Sounds

• The Neural Representation of Sound Begins in the Cochlear Nuclei

  • The Cochlear Nerve Imposes a Tonotopic Organization on the Cochlear Nuclei and Distributes Acoustic Information into Parallel Pathways

  • The Ventral Cochlear Nucleus Extracts Information About the Temporal and Spectral Structure of Sounds

  • The Dorsal Cochlear Nucleus Integrates Acoustic with Somatosensory Information in Making Use of Spectral Cues for Localizing Sounds

• The Superior Olivary Complex of Mammals Contains Separate Circuits for Detecting Interaural Time and Intensity Differences

  • The Medial Superior Olive Generates a Map of Interaural Time Differences

  • The Lateral Superior Olive Detects Interaural Intensity Differences

• Efferent Signals from the Superior Olivary Complex Provide Feedback to the Cochlea

  • Brain Stem Pathways Converge in the Inferior Colliculus

  • Sound Location Information from the Inferior Colliculus Creates a Spatial Map of Sound in the Superior Colliculus

  • Midbrain Sound-Localization Pathways Are Sensitive to Experience in Early Life

• The Inferior Colliculus Transmits Auditory Information to the Cerebral Cortex

  • The Auditory Cortex Maps Numerous Aspects of Sound

  • Auditory Information Is Processed in Multiple Cortical Areas

  • Insectivorous Bats Have Cortical Areas Specialized for Behaviorally Relevant Features of Sound

  • A Second Sound-Localization Pathway from the Inferior Colliculus Involves the Cerebral Cortex in Gaze Control

  • Auditory Circuits in the Cerebral Cortex Are Segregated into Separate Processing Streams

  • The Cerebral Cortex Modulates Processing in Subcortical Auditory Areas

• Hearing Is Crucial for Vocal Learning and Production in Both Humans and Songbirds

  • Normal Vocal Behavior Cannot Be Learned in Isolation

  • Vocal Learning Is Optimal During a Sensitive Period

  • Both Humans and Songbirds Possess Specialized Neural Networks for Vocalization

  • Songbirds Have Feature Detectors for Learned Vocalizations

• An Overall View

Because of its role in the understanding and production of speech, auditory perception is one of the most important sensory modalities in humans. In most animals hearing is crucial for localizing and identifying sounds; for some species, hearing additionally guides the learning of vocal behavior.

Once sounds have been transformed into electrical responses in the cochlea, a rich hierarchy of auditory circuits analyzes and processes these signals to give rise to auditory perception. The auditory system differs from most other sensory systems in that the location of stimuli in space is not conveyed by the spatial arrangement of the afferent pathways. Instead, the localization and identification of sounds is constructed from patterns of frequencies mapped at the two ears as well as from their relative intensity and timing. The auditory system is also notable for its temporal sensitivity; time differences as small as 10 μs can be detected. Auditory pathways resemble other sensory systems, however, in that different features of acoustic information are processed in discrete circuits that eventually converge to form complex representations of sound.

In addition to studies of primates and mammals such as cats and rodents, research on animals with especially acute or specialized auditory capacities—frogs, bats, barn owls, and songbirds—has provided a wealth of information about auditory processing. Many of the principles learned from the study of such auditory specialists have proven to be generally applicable. They are simply easier to detect in animals with specialized mechanisms.

Hearing helps to alert animals to the presence of unseen dangers or opportunities, and in many species also serves as a basis for communication. Information about where sounds arise and what they mean must be extracted from the representations of the physical characteristics of sound at each of the ears. To understand how animals process sound, it is useful first to consider what cues are available.

Most vertebrates take advantage of having two ears for localizing sounds in the horizontal plane. Sound sources at different positions in that plane affect the two ears differentially: Sound arrives earlier and is more intense at the ear nearer the source (Figure 31–1A). The size of the head determines how interaural time delays are related to the location of sound sources; the neuronal circuitry determines the precision with which time delays are resolved. Because sound travels at roughly 340 m/s in air, the maximal interaural delay in humans is approximately 600 μs; in small birds the greatest delay is only 35 μs. Humans can resolve the location of a sound source directly ahead to within approximately 1 degree, corresponding to an interaural time difference of 10 μs. Interaural time differences are particularly well conveyed by neurons that encode relatively low frequencies, for groups of these neurons can fire at the same position in every cycle of the sound and in this way encode the interaural time difference as an interaural phase difference.

Sounds of high frequencies produce sound shadows or intensity differences between the two ears. For many mammals with small heads, high-frequency sounds provide the primary cue for localizing sound in the horizontal plane.

Spectral filtering in mammals allows sounds to be localized in the vertical plane and with a single ear (Figure 31–1B). High-frequency sounds, with wavelengths that are close to or smaller than the dimensions of the head, shoulders, and external ears, interact with those parts of the body to produce constructive and destructive interference, introducing broad spectral peaks and narrow, deep spectral notches whose frequency changes with the location of the sound. High-frequency sounds from different origins are filtered differently because in mammals the shape of the external ear differs back-to-front as well as top-to-bottom. Animals learn to use these spectral cues to locate sound sources. If the shape of the ear is experimentally altered, even adult humans can learn to make use of a new pattern of spectral cues. If animals lose hearing in one ear, they lose interaural timing and intensity cues and must depend completely on spectral cues for localizing sounds.

How do we make sense of the sounds that we hear? Most natural sounds contain energy over a wide range of frequencies and change rapidly over time. The information used to recognize sounds varies among animal species, and depends on listening conditions and experience. Human speech, for example, can be understood in the midst of noise, over electronic devices that distort sounds, and even through cochlear implants. One reason for its robustness is that speech contains redundant cues: The vocal apparatus produces sounds in which multiple parameters covary. At the same time, this makes the task of understanding how animals recognize patterns a complicated one. It is not clear which cues are used by animals under what conditions.

Music is a source of pleasure to human beings. Musical instruments and human voices produce sounds that have energy at the fundamental frequency that corresponds to the perceived pitch as well as at multiples of that frequency that give sounds a quality that allows us, for example, to distinguish a flute from a violin when their pitch is the same. Musical pitches are largely in the low frequency range in which auditory nerve fibers fire in phase with sounds. Musical sounds when combined simultaneously produce chords, and chord progressions produce melodies. Euphonious, pleasant chords elicit regular, periodic firing in the auditory nerve in which the most common interval between action potentials corresponds to the period of the perceived pitch. In dissonant sounds there is less regularity both in the sound itself and in the firing of auditory nerve fibers; the component frequencies are so close that they interfere with one another instead of periodically reinforcing one another.

The neural pathways that process acoustic information extend from the ear to the brain stem, through the midbrain and thalamus, to the cerebral cortex (Figure 31–2). Acoustic information is conveyed from the cochlea by the central processes of cochlear ganglion cells (see Figure 30–17) that terminate in the cochlear nuclei in the brain stem. There information is relayed to several different types of neurons, most of which are arrayed tonotopically.

The axons of each of these types of neurons take a different route and terminate on separate targets in the brain stem and midbrain, ultimately bringing acoustic information mainly to the contralateral inferior colliculus. Some of the pathways from the cochlear nuclei to the inferior colliculus are direct; others involve one or two synaptic stages in brain stem auditory nuclei. From the inferior colliculi acoustic information flows two ways: to the ipsilateral superior colliculus, where it participates in orienting the head and eyes in response to sounds, and to the ipsilateral thalamus, the relay to auditory areas of the cerebral cortex. The afferent auditory pathways from the periphery to higher brain regions include efferent feedback at many levels.

The Cochlear Nerve Imposes a Tonotopic Organization on the Cochlear Nuclei and Distributes Acoustic Information into Parallel Pathways

The afferent nerve fibers from cochlear ganglion cells are bundled in the cochlear or auditory component of the vestibulocochlear (eighth cranial) nerve and terminate exclusively in the cochlear nuclei. The cochlear nerve in mammals contains two groups of fibers: a large contingent (95%) of myelinated fibers that receives input from inner hair cells, and a small number (5%) of unmyelinated fibers that receives input from outer hair cells.

The larger, more numerous, myelinated fibers are much better understood than the unmyelinated fibers. Each myelinated fiber detects energy over a narrow range of frequencies; together these fibers carry information about the moment-to-monent variation in the frequency content of sounds. The unmyelinated fibers terminate on the large neurons in the ventral cochlear nuclei and on the small granule cells that surround the ventral cochlear nuclei. These fibers integrate information from a relatively wide region of the cochlea and are therefore unlikely to be as sharply tuned as the myelinated fibers. Because it is difficult to record from these tiny fibers, the information they convey to the brain is unknown. Indirect evidence suggests that they encode the intensity of sounds over a wide dynamic range.

Two features of the cochlear nuclei are important. First, the cochlear nerve fibers terminate in these nuclei in a tonotopic organization. Fibers that carry information from the apical end of the cochlea, which detects low frequencies, terminate ventrally in the ventral and dorsal cochlear nuclei; those that carry information from the basal end of the cochlea, which detects high frequencies, terminate dorsally (Figure 31–3). Second, each cochlear nerve fiber innervates several different areas within the cochlear nuclei, contacting various types of neurons that have distinct projection patterns to higher auditory centers. As a result, the auditory pathway is split into at least four parallel ascending pathways that simultaneously extract different facets of acoustic information from the representation of sound carried by cochlear nerve fibers.

The Ventral Cochlear Nucleus Extracts Information About the Temporal and Spectral Structure of Sounds

The principal cells of the unlayered ventral cochlear nucleus sharpen timing and spectral information and convey it to other auditory nuclei in the brain stem. Three types of neurons are intermingled and form separate pathways through the brain stem.

Bushy cells project bilaterally to the superior olivary complex. This pathway has two parts, one through the medial superior olive and the other through the lateral superior olive and medial nucleus of the trapezoid body. Large spherical bushy cells sense low frequencies and project bilaterally to the medial superior olive, forming a circuit that detects interaural time delay and permits the localization of low-frequency sounds in the horizontal plane. Small spherical bushy cells and globular bushy cells sense high frequencies and are associated with the lateral superior olive. Small spherical bushy cells probably excite the lateral superior olive ipsilaterally. The globular bushy cells, through calyceal axonal endings that surround the postsynaptic neurons, excite neurons in the contralateral medial nucleus of the trapezoid body that in turn inhibit principal cells of the lateral superior olive. The pathways through the lateral superior olive are involved in the detection of interaural intensity differences and contribute to the localization of high-frequency sounds in the horizontal plane.

Stellate cells excite neurons in the ipsilateral dorsal cochlear nucleus, probably in the ipsilateral lateral superior olive, in the periolivary nuclei, and in the contralateral ventral nucleus of the lateral lemniscus through collaterals of axons that project to the contralateral inferior colliculus. The tonotopic array of stellate cells encodes the spectra of sounds.

Octopus cells excite targets in the contralateral periolivary region and the ventral nucleus of the lateral lemniscus. These neurons detect onset transients and periodicity in sounds and may be involved in the recognition of sound patterns.

The differences in the integrative tasks performed by these pathways are evident in the synaptic structures and shapes of the three types of neurons. The shapes of their dendrites differ, reflecting differences in the way they collect information from cochlear nerve fibers (Figure 31–4A, B). The dendrites of bushy and stellate cells span only a small range of the tonotopic array of auditory nerve fibers, receive input from relatively few auditory nerve fibers, and are consequently sharply tuned. Many of the inputs to bushy cells are from unusually large terminals that surround the cell bodies, meeting the need for large synaptic currents. Octopus cells, in contrast, have dendrites that span a large proportion of the array of afferent fibers, receiving input from many cochlear nerve fibers, and thus are broadly tuned. Their need for large synaptic currents is met by summing inputs from large numbers of small terminals.

The biophysical properties of neurons determine how synaptic currents are converted to voltage changes and over how long a time synaptic inputs are integrated. Octopus and bushy cells in the ventral cochlear nucleus are able to respond with exceptionally rapid and precisely timed synaptic potentials. These neurons have a prominent, low-voltage-activated K⁺ conductance that confers a low input resistance and rapid responsiveness and prevents repetitive firing (Figure 31–4C). The large synaptic currents that are required to trigger action potentials in these leaky cells are delivered through rapidly gated, high-conductance, AMPA-type (α-amino-3-hydroxy-5-methylisoxazole-4-propionate) glutamate receptors found at many synaptic release sites. In contrast, stellate cells, in which even relatively small depolarizing currents produce large, protracted voltage changes, generate slower excitatory postsynaptic potentials (EPSPs) in response to synaptic currents. NMDA-type (N -methyl-D-aspartate) glutamate receptors enhance the slow responses.

Neurons in the ventral cochlear nucleus are able to encode different features of sounds because of differences in the pattern of input and their biophysical properties. Octopus cells detect synchronous firing in cochlear nerve fibers with exceptional temporal precision. Large, low-voltage-activated K⁺ and hyperpolarization-activated conductances give octopus cells exceptionally low input resistances so that individual cochlear nerve fibers can produce synaptic responses of only approximately 1 mV and only 1 ms in duration. To fire action potentials octopus cells require summation of the rising phases of numerous synaptic inputs. Individual octopus cells detect coincident firing in the relatively large number of cochlear nerve fibers (more than 60) that contact them. Coincident firing in large numbers of cochlear nerve fibers is produced by periodic sounds such as vowels and musical sounds and by the onset of broadband sounds found in consonants or clicks.

Compared to octopus cells, bushy cells convey a more sharply tuned but less temporally precise version of the firing patterns of cochlear nerve fibers. The roughly 10 cochlear nerve fibers that terminate on each bushy cell deliver relatively large synaptic currents that require summation of only a few inputs to trigger an action potential. Two properties of bushy cells enable these neurons to encode with precision the detailed temporal structure of sounds. The cells' low input resistance shortens the voltage changes produced by the incoming synaptic currents; their need to summate several inputs removes variability in the timing of firing in cochlear nerve fibers by averaging.

The temporal fine structure of sounds that bushy cells encode provides information about the relative time of arrival of inputs to the two ears and is used at the next synaptic stage to form a map of the interaural time differences that underlie the ability to localize sound sources in the horizontal plane. The detection of musical pitch also requires the encoding of the temporal fine structure of sounds, but whether that information is carried through octopus or bushy cells or through a combination of pathways is not known.

Individual stellate cells detect intensity over a narrow frequency range and as a population they provide a continuous representation of the spectrum of sound energy. They are sharply tuned, being driven by only approximately 10 cochlear nerve fibers. Feedforward excitation through other, similarly tuned stellate cells and enhancement of inputs through NMDA-type receptors compensates for adaptation and obscures the fine structure of sound, encoded by cochlear nerve inputs, but allows the tonic firing rate of the cells to reflect the intensity of sounds to which they are tuned. Sideband inhibition enhances the cells' encoding of spectral peaks and troughs. Stellate cells track modulations in intensity on a timescale of tens of milliseconds that is known to be important for understanding speech.

The Dorsal Cochlear Nucleus Integrates Acoustic with Somatosensory Information in Making Use of Spectral Cues for Localizing Sounds

In mammals cochlear nerve fibers extend into a layered dorsal cochlear nucleus (Figure 31–4A, B). The dorsal cochlear nucleus receives input from two systems of neurons that project to different layers.

The outermost molecular layer is the terminus of a system of parallel fibers, the unmyelinated axons of granule cells that are scattered in and around the cochlear nuclei. The parallel fibers terminate on interneurons that bear a strong resemblance to those in the cerebellum. This system transmits somatosensory, vestibular, and auditory information from widespread regions of the brain to the molecular layer.

The deep layer is the terminus of cochlear nerve fibers, which convey acoustic information directly and indirectly through stellate cells of the ventral cochlear nucleus. The cochlear nerve inputs are tonotopically organized in isofrequency laminae that run at right angles to parallel fibers.

Inputs from both systems of fibers are combined in the principal neurons of the dorsal cochlear nucleus, the fusiform cells. Parallel fibers excite fusiform cells on their spiny apical dendrites in the molecular layer at plastic synapses, whereas cochlear nerve fibers and stellate cells of the ventral cochlear nucleus excite fusiform cells on their smooth basal dendrites in the deep layer at synapses that show little plasticity. Both fiber systems also inhibit fusiform cells through inhibitory interneurons.

Eric Young and his colleagues proposed that the dorsal cochlear nucleus helps mammals interpret spectral cues for localization of sound sources. When animals move their head or ears or walk, they affect the angle of incidence of sounds to the ears even when a sound source stays in one place. To make full use of spectral cues, animals must differentiate the predictable cues that are produced by their own movements from the more interesting, unpredictable ones that inform them about the location of external sound sources. The somatosensory and vestibular information about the position of the head and ears, as well as descending information from higher levels of the nervous system about the animal's own movements, can serve to identify the predictable cues. Fusiform cells thus relate the spectral cues they receive through the deep layer with predictable cues received through the molecular layer.

Among vertebrates only mammals have dorsal cochlear nuclei and only mammals have been shown to make use of spectral cues. Most birds and other reptiles hear to only 5 kHz, which corresponds to a wavelength of 6.8 cm. Sounds with wavelengths that exceed 7 cm do not interact with heads whose diameters are less than 2 cm wide and thus do not provide spectral cues. Conversely, humans hear to 20 kHz, corresponding to a wavelength of 1.7 cm; cats and dogs hear to approximately 50 kHz; and mice and bats hear to 100 kHz. The wavelengths of these high-frequency sounds are small with respect to the dimensions of the heads and ears and therefore provide useful spectral cues.

In many vertebrates, including mammals and birds, neurons in the superior olivary complex compare the activity of cells in the bilateral cochlear nuclei to locate sound sources. Separate circuits detect interaural time and intensity differences.

The Medial Superior Olive Generates a Map of Interaural Time Differences

Differences in arrival times at the ears are not represented at the cochlea. Instead, a map of interaural phase is created in the medial superior olive by a comparison of the timing of firing in responses to sounds from the two ears. Sounds arrive at the near ear before they arrive at the far ear, with interaural time differences being directly related to the location of sound sources in the horizontal plane.

Cochlear nerve fibers tuned to frequencies below 4 kHz and their bushy cell targets encode sounds by firing in phase with the pressure waves. This property is known as phase-locking. Although individual neurons may fail to fire at some cycles, the population of neurons represents the fine structure of sound waves by firing with every cycle. In so doing, these neurons carry information about the timing of inputs with every cycle of the sound. Sounds arriving from the side evoke phase-locked firing that is consistently earlier at the near ear than at the far ear, resulting in consistent interaural phase differences (Figure 31–5A).

In 1948 Lloyd Jeffress suggested that an array of detectors of coincident inputs from the two ears, transmitted through delay lines comprising axons with systematically differing lengths, could form a map of interaural time differences and thus a map of the location of sound sources (Figure 31–5B). In such a circuit conduction delays compensate for the early arrival at the near ear. Interaural time delays increase systematically as sounds move from the midline to the side, resulting in coincident firing further toward the edge of the neuronal array.

Such neuronal arrays have indeed been found in the barn owl in the homolog of the medial superior olivary nucleus. Mammals and chickens use a variant of this neuronal arrangement. The principal neurons of the medial superior olive form a sheet of one or a few cells' thickness on each side of the midline. Each neuron has two tufts of dendrites, one extending to the lateral face of the sheet, the other projecting to the medial face of the sheet (Figure 31–5C). The dendrites at the lateral face are contacted by large spherical bushy cells from the ipsilateral cochlear nucleus, whereas the dendrites at the medial face are contacted by large spherical bushy cells of matching best frequency from the contralateral cochlear nucleus. The axons of bushy cells terminate in the contralateral medial superior olive with delay lines just as Jeffress had suggested, but the branches that innervate the ipsilateral medial superior olive are of equal length (see Figure 31–5C).

Each medial superior olive receives coincident inputs from the two ears only when sounds come from the contralateral half of space. As sound sources move from the midline to the most lateral point on the contralateral side of the head, the earlier arrival of sounds at the contralateral ear needs to be compensated by successively longer delay lines. This results in inputs from the two ears coinciding at increasingly posterior regions of the medial superior olive, thereby forming a map of interaural phase. Recent findings indicate that inhibition that is superimposed on these excitatory inputs plays a significant role in sharpening the map of interaural phase.

In responding to interaural phase, individual neurons in the medial superior olive provide ambiguous information about interaural time differences. Phase ambiguities are resolved when sounds have energy at multiple frequencies, as natural sounds almost always do. The sheet of neurons of the medial superior olive forms a representation of interaural phase along the rostrocaudal dimension but its tonotopic innervation by bushy cells imposes a tonotopic organization in the dorsoventral dimension. Sounds that contain energy at multiple frequencies evoke maximal coincident firing in a single dorsoventral column of neurons that localizes sound sources unambiguously.

Each medial superior olive thus forms a map of the location of sound sources in the contralateral hemifield. The striking difference between this spatial map and those in other sensory systems is that this map is not the result of the spatial arrangement of inputs, like retinotopic or somatosensory maps, but is inferred by the brain from computations made in the afferent pathways.

The Lateral Superior Olive Detects Interaural Intensity Differences

Sounds with wavelengths that are similar to or smaller than the head are deflected by the head, causing the intensity at the near ear to be greater than that at the far ear. In humans interaural intensities can differ in sounds that have frequencies greater than about 2 kHz. Interaural intensity differences produced by such head shadowing are detected by a neuronal circuit that includes the medial nucleus of the trapezoid body and the lateral superior olive.

Although the lateral superior olive does not form a map of the location of sounds in the horizontal plane, it performs the first of several integrative steps that use interaural intensity differences to localize sounds. Neurons in this nucleus balance excitatory input from small spherical bushy cells and stellate cells in the ipsilateral ventral cochlear nucleus with inhibitory input from a disynaptic pathway that includes globular bushy cells in the contralateral ventral cochlear nucleus and principal neurons of the ipsilateral medial nucleus of the trapezoid body (Figure 31–6). Sounds that arise ipsilaterally generate relatively strong excitation and relatively weak inhibition whereas those that arise contralaterally generate stronger inhibition than excitation. Differences in the location of the sound sources that generate differences in the balance between excitation and inhibition are reflected in the firing rates of neurons in the lateral superior olive. Neurons in the lateral superior olive are thus activated more strongly by sounds from the ipsilateral than from the contralateral hemifield.

The relative strength of inhibitory and excitatory inputs varies in the population of cells in the lateral superior olive, and thus in the degree of interaural level difference that causes them to stop firing. Although no individual neuron is sharply tuned to a particular interaural intensity difference, the combination of firing from a variety of neurons with different cutoffs conveys the necessary information through a population code.

In order to balance excitation and inhibition stimulated by the same sound, the ipsilateral excitation and contralateral inhibition must arrive at neurons in the lateral superior olive at the same time. Thus excitation that arises monosynaptically from the ipsilateral ventral cochlear nucleus must arrive at the same time as inhibition that arises disynaptically from the contralateral cochlear nucleus. It is probably for this reason that the axons of globular bushy cells are exceptionally large; their synapses with the principal cells of the medial nucleus of the trapezoid body are the large calyces of Held that produce synaptic responses with short and consistently timed delays. The axons of small spherical bushy cells and cochlear nuclear stellate cells that carry ipsilateral excitation are smaller than those of globular bushy cells and thus conduct more slowly.

The terminals of the globular bushy cells, the calyces of Held, engulf the cell bodies of trapezoid-body neurons so dramatically that they caught the attention of early anatomists and modern biophysicists. A single somatic terminal releases neurotransmitter at numerous release sites and generates large synaptic currents. The capability of making reliable recordings pre- and postsynaptically has allowed the mechanisms of synaptic transmission to be studied in detail at this synapse (see Chapter 12).

Although sensory systems are largely afferent, bringing sensory information to the brain, recent studies have led to an appreciation of the importance of efferent regulation at many levels of the auditory system, including even efferent signaling to the cochlea from the superior olivary complex in the brain stem.

Olivocochlear neurons form a feedback loop from the superior olivary complex to hair cells in the cochlea. Their cell bodies lie around the major dense clusters of cell bodies in the olivary nuclei. In mammals two groups of olivocochlear neurons have been distinguished. The medial olivocochlear neurons have myelinated axons that terminate on the outer hair cells bilaterally; the lateral olivocochlear neurons have unmyelinated axons that terminate ipsilaterally on the afferent fibers associated with inner hair cells.

Most medial olivocochlear neurons, with cell bodies that lie ventral and medial within the olivary complex, send their axons to the contralateral cochlea (Figure 31–7), but many also innervate the ipsilateral cochlea. These cholinergic neurons act on hair cells through a special class of nicotinic acetylcholine receptor-channels formed from α9 and α10 subunits. Calcium ions that enter through these channels activate K⁺ channels that hyperpolarize outer hair cells. These neurons thus mediate negative feedback and are binaural, being driven predominantly but not exclusively by stellate cells of the contralateral ventral cochlear nucleus. Collateral branches of olivocochlear neurons terminate on stellate cells in the cochlear nucleus, acting on conventional nicotinic and muscarinic acetylcholine receptors, forming an excitatory feedback loop. Activity in these efferent fibers increases the representation of signals in noise, reduces the sensitivity of the cochlea, and protects it from damage by loud sounds.

Lateral olivocochlear neurons, with cell bodies that lie in and around the lateral superior olive, send their unmyelinated axons exclusively to the ipsilateral cochlea, where they terminate on the afferent fibers from inner hair cells. These efferents balance the excitability of cochlear nerve fibers at the two ears.

Brain Stem Pathways Converge in the Inferior Colliculus

The inferior colliculus occupies a central position in the auditory pathway of all vertebrate animals because all auditory pathways ascending through the brain stem converge there (Figure 31–7). The most important sources of excitation are stellate cells of the contralateral ventral cochlear nucleus, fusiform cells of the contralateral dorsal cochlear nucleus, principal cells of the ipsilateral medial superior olive and contralateral lateral superior olive, principal cells of ipsi- and contralateral dorsal nuclei of the lateral lemniscus, commissural connections from the contralateral inferior colliculus, and pyramidal cells in layer V of the auditory cortex. Important sources of inhibition include the nuclei of the lateral lemniscus, the ipsilateral lateral superior olive, the superior paraolivary nucleus, and the contralateral inferior colliculus.

The inferior colliculus of mammals is subdivided into the central nucleus, dorsal cortex, and external cortex. The central nucleus is tonotopically organized. All neurons in each lamina have similar best frequencies. Low frequencies are represented dorsolaterally and high frequencies ventromedially. Fine mapping has shown that the tonotopic organization is discontinuous; the separation between best frequencies corresponds to psychophysically measured critical bands of approximately one-third octave. Although the central nucleus is organized tonotopically, the spectral range of inputs to these neurons is broader than at earlier stages in the auditory pathway. Inhibition sharpens the frequency tuning of inferior colliculus excitatory neurons, and tuning can be further modulated by descending inputs from the cortex.

Many neurons in the central nucleus carry information about the location of sound sources. The majority of these cells are sensitive to interaural time and intensity differences, which are known to be essential cues for localizing sounds in the azimuth. Neurons are also sensitive to spectral cues that localize sounds in the vertical plane (see Figure 31–1B). To localize sounds accurately, animals must ignore the reflections of sounds from surrounding surfaces that arrive after the initial direct wave front. Psychophysical experiments have shown that mammals suppress all but the earliest versions, a phenomenon termed the precedence effect. Physiological correlates of the precedence effect have been measured in the inferior colliculus, where inhibition suppresses simulated reflections of sounds.

Sound Location Information from the Inferior Colliculus Creates a Spatial Map of Sound in the Superior Colliculus

The inferior colliculus is not only a convergence point but also a branch point for ascending or outflow pathways. Central nucleus neurons project to the thalamus and also to the external cortex of the inferior colliculus and the nucleus of the brachium of the inferior colliculus, both of which then project to the superior colliculus (or the optic tectum in birds).

The superior colliculus is critical for reflexive orienting movements of the head and eyes to acoustic and visual cues in space. By the time they reach the superior colliculus, binaural sound cues and the monaural spectral cues that underlie mammalian sound localization merge to create a spatial map of sound, an auditory map, in which neurons are unambiguously tuned to specific sound directions. This convergence is critical, for the binaural level and timing differences alone cannot unambiguously code for a single position in space. The spectral cues that provide information about vertical location are essential. Different locations in the vertical plane can give rise to identical interaural time or intensity differences. Such a spatial map is formed both in birds and in some mammals (Figure 31–8). In ferrets and guinea pigs topographic representations of the location of sound in the horizontal plane are found in the external cortex and the nucleus of the brachium of the inferior colliculus.

Within the superior colliculus the auditory map is congruent with maps of visual space and the body surface. Unlike the visual and somatosensory maps, the auditory map is computed from a combination of cues that identify the specific position of a sound source in space, and is not based on the peripheral receptor surface.

Auditory, visual, and somatosensory neurons in the superior colliculus all converge on output pathways in the same structure that controls orienting movements of the eyes, head, and external ears. The motor circuits of the superior colliculus are mapped with respect to motor targets in space, and are aligned with the sensory maps. Such sensory-motor correspondence facilitates the sensory guiding of movements.

Midbrain Sound-Localization Pathways Are Sensitive to Experience in Early Life

Cues for sound localization vary both within and across individuals as a result of differences in the size and shape of the head and ears. Moreover, the neural representation of these cues can also vary during development and with changes of the brain as a result of aging. The neural system for sound localization, especially in barn owls, has become an important model for the study of synaptic plasticity in the brain.

In addition to studies of normal development of sound localization, artificial manipulations of sensory experience have been particularly revealing. In adult mammals and birds occlusion of one ear alters the binaural acoustic cues and causes mislocalization of sounds toward the open ear. When such manipulations of sound cues are conducted in young owls, the animals initially misorient but recover accurate orienting responses over a period of weeks. This is accompanied by a shift in the tuning of auditory neurons in the colliculi toward the abnormal cues that result from use of the earplug. If the earplug is removed after adaptation, owls initially make orienting errors in the opposite direction, but gradually recover accurate behavior as well as a normal spatial map of sound in the brain.

Because visual and auditory cues are normally in alignment in the external world, their representations are also congruent in the superior colliculus (Figure 31–8B). Alterations of visual inputs, and thus of the normal match between acoustic and visual cues, also affect sound localization. When carried out early in development, such visual manipulations cause striking changes in the spatial mapping of sound in the colliculi. For example, when an eye is deviated laterally in a young ferret by removal of an extraocular muscle, the auditory map in the contralateral superior colliculus undergoes a shift so that the visual and auditory maps remain in alignment.

In barn owls, which cannot move their eyes, the visual field has been experimentally misaligned by raising owls with prisms that shift the visual world horizontally (Figure 31–9A). Over the course of weeks owls adjust their auditory orienting responses so that their movements correspond to the location of a target in the shifted visual space. Although the response to sounds is incorrect in terms of purely auditory cues, this is an adaptive response because it causes the animal to look at the source of the sound. In parallel with the changes in behavior, neurons in the colliculi gradually shift their response to interaural time differences by the amount of displacement in the visual field (Figure 31–9B).

The mechanisms of this plasticity, which may be very general, have been extensively studied by Eric Knudsen and his colleagues. The auditory map in the central nucleus of the inferior colliculus does not change in young owls reared with prisms. However, orderly axonal connections from the central nucleus to the external nucleus gradually sprout, forming a pathway that represents the newly learned interaural timing difference (Figure 31–10). The original inputs from the central nucleus persist but are actively suppressed by GABA-ergic inhibitory inputs. The retention of the old functional connections likely underlies the rapid restoration of normal tuning after the prisms have been removed.

The prism experiments illustrate that vision plays a dominant role in changes in synaptic function in the auditory system. As further evidence for this idea, a small lesion in the superior colliculus prevents synaptic changes in the corresponding area of the inferior colliculus to accommodate prism displacement. Presumably the same instructive signals from the superior colliculus guide the transformation of auditory cues into a well-aligned spatial map of sounds during normal development.

A striking aspect of plasticity in both birds and mammals is the sensitivity of the spatial map to an animal's age. As in many other systems there is a critical period in which altered sensory experience can alter the brain and behavior. Marked adaptive changes in orienting behavior are seen in juvenile owls, ferrets, and guinea pigs but ordinarily not in adults. However, learning in young animals increases the capacity for plasticity in adults. If juvenile owls are subjected to a period of prism displacement but then returned to normal vision, they are capable of reacquiring the shifted map and the corresponding behavior when exposed to the same prism displacement as adults. However, their inferior colliculus cannot acquire a map to which they were not exposed to as juveniles. The period of early learning thus leaves a specific and permanent trace in the nervous system.

An animal's motivational state is also important in how malleable its brain is. If adult barn owls fitted with prisms are allowed to hunt for live mice, a more arousing situation than typical laboratory conditions, they display behavioral and neural plasticity intermediate between that of normally housed adults and juveniles. Similar increases in sensory plasticity have been seen in mammals actively involved in tasks that require specific sensory information. This suggests the clinically important point that more plasticity can be elicited from the adult brain if attention and motivation can be engaged appropriately.

Auditory information ascends to the cerebral cortex through the thalamus. Neurons of the inferior colliculus project to the medial geniculate body, where principal cells in turn project to the auditory cortex. The pathways from the inferior colliculus include a lemniscal or core pathway and extralemniscal or belt pathways. Descending projections from the auditory cortex to the medial geniculate body are prominent both anatomically and functionally.

The Auditory Cortex Maps Numerous Aspects of Sound

Ascending auditory pathways terminate in the auditory cortex, which includes multiple distinct areas on the dorsal surface of the temporal lobe. The most prominent projection is from the ventral division of the medial geniculate nucleus to the primary auditory cortex (A1, or Brodmann area 41). As in the lower relays of auditory processing, this cytoarchitectonically distinct region contains a tonotopic representation of characteristic frequencies: Neurons are arrayed in a systematic map reflecting the frequencies that best stimulate them. Neurons tuned to low frequencies are found at the rostral end of A1, and those responsive to high frequencies in the caudal region (Figure 31–11). Thus, like the visual and somatosensory cortices, the primary auditory cortex contains a map reflecting the pattern of peripheral sensors.

Because the cochlea encodes only frequency, however, a one-dimensional map from the periphery is spread across the two-dimensional surface of the cortex, with a smooth frequency gradient in one direction and iso-frequency contours along the other direction. In many animals subregions of the auditory cortex representing biologically significant frequencies are enlarged because of extensive inputs, similar to the large area in the primary visual cortex devoted to inputs from the fovea.

In addition to frequency, other features of auditory stimuli are mapped in the primary auditory cortex, although the overall organization is less clear and precise than for vision. Auditory neurons in A1 are excited by input from both ears (EE), with the contralateral input usually stronger than the ipsilateral contribution, or by unilateral input (EI). The EI neurons are inhibited by stimulation of the opposite ear. Summation columns of EE neurons alternate with suppression columns of EI neurons, especially in the high-frequency portion of A1, creating a map of interaural interactions at right angles to the axis of tonotopic mapping. This partitions the auditory cortex into columns responsive to every audible frequency and each type of binaural interaction.

Certain neurons in A1 also seem to be organized according to bandwidth, that is, according to their responsiveness to a narrow or broad range of frequencies. Distinct subregions of A1 form clusters of cells with narrow or broadband tuning within individual iso-frequency contours. Synaptic connections within the cortex respect these clusters, with neurons receiving intracortical input primarily from neurons with similar bandwidths and characteristic frequencies. This modular organization of bandwidth selectivity may allow redundant processing of incoming signals through neuronal filters of varying bandwidths as well as center frequencies, which could be useful for the analysis of spectrally complex sounds such as species-specific vocalizations, including speech.

Several other parameters are mapped on the surface of A1. These include neuronal response latency, loudness, modulation of loudness, and the rate and direction of frequency modulation. Although it remains to be seen how these various maps intersect, this array of parameters clearly endows each neuron and each location in A1 with the ability to represent many independent variables of sound and generates a great diversity of neuronal selectivity.

As is true for visual and somatosensory areas of the cortex, sensory representation in A1 can change in response to altered input. After peripheral hearing loss, tonotopic mapping can be altered so that neurons originally responsive to sounds within the range of the hearing loss begin to respond to adjacent frequencies. The work of Michael Merzenich and others has shown that behavioral training of adult animals can also result in large-scale reorganization of auditory cortex, so that the most behaviorally relevant frequencies—those specifically associated with attention or reinforcement—come to be overrepresented.

The auditory areas of young animals are particularly plastic. In rodents the frequency organization of A1 emerges gradually during development from an early, crude frequency map. Raising animals in acoustic environments in which they are exposed to repeated tone pulses of a particular frequency results in a persistent expansion of cortical areas devoted to that frequency, accompanied by a general deterioration and broadening of the tonotopic map. This result not only suggests that the development of A1 is experience-dependent, but raises the possibility that early exposure to abnormal sound environments can create long-term disruptions of high-level sensory processing. A greater understanding of how this happens and whether it is also true for human fetuses and infants may provide insights into the origin and remediation of disorders in which auditory processing is centrally impaired, such as many forms of dyslexia. Moreover, the ability to induce synaptic changes in adult auditory cortex by engaging attention or reward raises new hopes for brain repair even in adulthood.

Auditory Information Is Processed in Multiple Cortical Areas

The primary auditory area of mammals is surrounded by multiple distinct regions, many of which are tonotopic. These highly tonotopic areas resemble A1 in that they receive direct input from the ventral division of the medial geniculate nucleus. Like the major thalamocortical visual projection, this input primarily contacts cortical layers IIIb and IV. Adjacent tonotopic fields have mirror-image tonotopy: The direction of tono-topy reverses at the boundary between fields.

As many as 7 to 10 secondary (belt) areas surround 3 to 4 primary or primary-like (core) areas (see Figure 31–13). These cortical areas receive input from the core areas of auditory cortex and in some cases from thalamic nuclei. Electrophysiological and imaging studies have confirmed that the primary auditory cortex in humans lies on Heschl's gyrus, in the temporal lobe, medial to the sylvian fissure. In addition, recent functional magnetic-resonance imaging studies have revealed that in humans, just as in monkeys, pure tones activate primarily core regions, whereas the neurons of belt areas prefer complex sounds such as narrow-band noise bursts.

Insectivorous Bats Have Cortical Areas Specialized for Behaviorally Relevant Features of Sound

Although it is generally assumed that the upstream auditory areas perform increasingly specialized functions related to hearing, our knowledge about the functions of serial relays is much less in the auditory system than the visual system. In humans one of the most important aspects of audition is its role in processing language, but we know relatively little about how speech sounds are analyzed by neural circuits. New techniques for imaging the human brain are gradually providing insights into the localization of cortical areas associated with language (see Chapter 60).

The best evidence for specialized analysis of complex auditory signals in auditory cortical areas comes from studies of insectivorous bats. These animals find their prey almost entirely through echolocation, emitting ultrasonic pulses of sound that are reflected by flying insects. Bats analyze the timing and structure of the echoes to help locate and identify the targets, and discrete auditory areas are devoted to processing different aspects of the echoes.

Many bats, such as the mustached bat studied by Nobuo Suga and his collaborators, emit echolocating pulses with two components. An initial constant-frequency (CF) component consists of several harmonically related sounds or harmonics. These harmonics are emitted stably for tens to hundred of milliseconds, akin to human vowel sounds. The constant-frequency component is followed by a sound that steeply decreases in frequency, the frequency-modulated (FM) component, which resembles the rapidly changing frequencies of human consonants (Figure 31–12A).

The FM sounds are used to determine the distance to the target. The bat measures the interval between the emitted sound and the returning echo, which corresponds to a particular distance, based on the relatively constant speed of sound. Neurons in the FM-FM area of auditory cortex (Figure 31–12B) respond preferentially to pulse-echo pairs separated by a specific delay. Moreover, these neurons respond better to particular combinations of sounds than to the individual sounds in isolation; such neurons are called feature detectors (Figure 31–12C). The FM-FM area contains an array of such detectors, with preferred delays systematically ranging from 0.4 ms to 18 ms, corresponding to target ranges of 7 cm to 310 cm, mapped along the cortical surface (see Figure 31–12B). These neurons are organized in columns, each of which is responsive to a particular combination of stimulus frequency and delay. In this way the bat, like the barn owl in its inferior colliculus, is able to represent an acoustic feature that is not directly represented by sensory receptors.

The CF components of bat calls are used to determine the relative speed of the target with respect to the bat and the acoustic image of the target. When an echolocating bat is flying toward an insect, the sounds reflected from the insect are Doppler-shifted to a higher frequency at the bat's ear, for the bat is moving toward the returning sound waves from the target, causing a relative speeding up of these waves at its ear. Similarly, a receding insect yields reflections of lowered frequency at the bat's ear. Neurons in the CF-CF area (see Figure 31–12B) are sharply tuned to a combination of frequencies close to the emitted frequency or its harmonics. Each neuron responds best to a combination of a pulse of a particular fundamental frequency and an echo corresponding to the first or second harmonic of the pulse, Doppler-shifted to a specific extent. As in the FM-FM area, neurons do not respond to the pulse or echo alone, but rather to the combination of the two CF signals.

CF-CF neurons are arranged in columns, each encoding a particular combination of frequencies. These columns are arranged regularly along the cortical surface, with the fundamental frequency along one axis and the echo harmonics along a perpendicular axis. This dual-frequency coordinate system creates a map wherein a specific location corresponds to a particular Doppler shift and thus a particular target velocity, ranging systematically from –2 m/s to 9 m/s.

The CF components of returning echoes are also used for detailed frequency analysis of the acoustic image, presumably important in its identification. The Doppler-shifted constant-frequency area (DSCF) of the mustached bat is a dramatic expansion of the primary auditory cortex's representation of frequencies between 60 kHz and 62 kHz, corresponding well to the set of returning echoes from the major CF component of the bat's call (see Figure 31–12B). Within the DSCF area individual neurons are extremely sharply tuned to frequency, so that the tiny changes in frequency created by fluttering moth wings are easily detected.

Transient inactivation of some of these specialized cortical areas, while the bat performs a discrimination task, strikingly supports the importance of their functional specialization in behavior. Silencing of the DSCF selectively impairs fine frequency discrimination while leaving time perception intact. Conversely, inactivation of the FM-FM area impairs the bat's ability to detect small differences in the time of arrival of two echoes, while leaving frequency perception unchanged.

Investigation of this auditory system was greatly facilitated by knowledge of the stimuli relevant to bats. It remains to be seen whether these cortical areas are functionally or anatomically analogous to particular fields in cats, monkeys, and humans. Regardless, ethologically significant stimuli are likely to be as useful in studying these other species as they have been in studies of bats.

A Second Sound-Localization Pathway from the Inferior Colliculus Involves the Cerebral Cortex in Gaze Control

Many auditory neurons in the cerebral cortex are sensitive to interaural time and level differences and therefore to the location of sounds in space. Most of these cells have large receptive fields and broad tuning. In contrast to the auditory relays in the midbrain, however, there is no evidence for an organized spatial map of sound in any of the cortical areas sensitive to sound location.

The sound-localization pathways in the cortex originate in the central nucleus of the inferior colliculus and ascend through the auditory thalamus, area A1, and cortical association areas, eventually reaching the frontal eye fields involved in gaze control. Eye or head movements can be elicited by stimulating the frontal eye fields, which connect directly to brain stem tegmentum premotor nuclei that mediate gaze changes, as well as to the superior colliculus. But the midbrain pathway from location-sensitive neurons in the inferior colliculus to the superior colliculus to gaze-control circuitry can directly regulate orientation movements of the head, eyes, and ears. Why should there be a second sound-localization pathway connected to gaze-control circuitry?

Behavioral experiments shed light on this question. Although lesions of A1 can result in profound sound-localization deficits in a monkey, no deficiency is seen when the task is simply to indicate the side of the sound source by pushing a lever. The deficit becomes apparent only when the animal must approach the location of a brief sound source, that is, when the task is the more complex one of forming an image of the source, remembering it, and moving toward it.

Experiments in barn owls have produced particularly compelling evidence. The ability of owls to orient to sounds in space is unaffected by inactivation of the avian equivalent of the frontal eye fields. Similarly, when the midbrain localization pathway is disrupted by pharmacological inactivation of the superior colliculus, the probability of an accurate head turn is decreased but animals still respond correctly more than half of the time. In contrast, when both structures are inactivated, animals are completely unable to orient accurately to acoustic stimuli on the contralateral side. Thus cortical and subcortical sound-localization pathways have parallel access to gaze control centers, perhaps providing some redundancy. Moreover, when only the frontal eye fields are inactivated, birds lose their ability to orient their gaze toward a target that has been extinguished and must be remembered, just as is seen with mammalian A1 lesions. Thus in both mammals and birds cortical pathways are required for more complex sound-localization tasks.

This appears to be a general difference between cortical and subcortical pathways. Subcortical circuits are important for rapid and reliable performance of behaviors that are critical to survival. Cortical circuitry allows for working memory, complex recognition tasks, and selection of stimuli and evaluation of their significance, resulting in slower but more differentiated performance. Examples of this also exist in auditory pathways not involved in localization. Conditioned fear responses to simple auditory stimuli are mediated by direct rapid pathways from the auditory thalamus to the amygdala; they can be elicited after cortical inactivation. However, fear responses that require more complex discrimination of auditory stimuli require pathways through the cortex, and are accordingly slower but more specific.

Auditory Circuits in the Cerebral Cortex Are Segregated into Separate Processing Streams

In the visual system the output from the primary visual cortex is segregated into separate dorsal and ventral streams concerned respectively with object location in space and object identification. A similar division of labor is thought to exist in the somatosensory cortex, and recent evidence suggests that the auditory cortex follows this plan.

Anatomical tracing studies of the three most accessible belt areas in primates show that the more rostral and ventral areas connect primarily to the more rostral and ventral areas of the temporal lobe, whereas the more caudal area projects to the dorsal and caudal temporal lobe. In addition, these belt areas and their temporal lobe targets both project to largely different areas of the frontal lobes (Figure 31–13).

The frontal areas receiving anterior auditory projections are generally implicated in nonspatial functions, whereas those that are targets of posterior auditory areas are implicated in spatial processing. Electrophysiological and imaging studies provide support for this. Caudal and parietal areas are more active when a sound must be located or moves, and ventral areas are more active during identification of the same stimulus or analysis of its pitch. Consistent with this segregation, inactivation of the posterior auditory field in cats impairs performance of a sound-localization task, whereas inactivation of the anterior auditory field interferes with a pattern-discrimination task, but not vice versa. Thus anterior-ventral pathways may identify auditory objects by analyzing spectral and temporal characteristics of sounds, whereas the more dorsal-posterior pathways may specialize in sound-source location, detection of sound-source motion, and spatial segregation of sources.

Although the idea that all sensory areas of the cerebral cortex initially segregate object identification and location is attractive, it is likely an oversimplification. It is clear that the medial-belt areas of the auditory cortex project to both dorsal and ventral frontal cortices, and neurons with broad spatial responsiveness are distributed throughout caudal and anterior areas. Imaging studies with more complex stimuli suggest that parietal pathways are involved in additional functions, including analysis of the temporal properties of acoustic stimuli such as spectral motion. The latter property might explain the clear role of the posterior pathway in speech perception in humans. Nonetheless, although the details may differ between systems, the basic concept holds that sensory systems decompose stimuli into features and analyze these in discrete pathways.

The Cerebral Cortex Modulates Processing in Subcortical Auditory Areas

An intriguing feature of all mammalian cortical areas, and one shared by the auditory system, is the massive projection from the cortex back to lower areas. There are almost 10-fold as many corticofugal fibers entering the sensory thalamus as there are axons projecting from the thalamus to the cortex. Projections from the auditory cortex also innervate the inferior colliculus, olivocochlear neurons, some basal ganglionic structures, and even the dorsal cochlear nucleus.

Insights into possible functions of this feedback have come from the bat's auditory system. Silencing of frequency-specific cortical areas leads to decreased thalamic and collicular responses, whereas activation of cortical projections increases and sharpens the responses of some neurons. The auditory cortex can therefore actively adjust and improve auditory signal processing in subcortical structures.

A variety of evidence suggests that cortical feedback also occurs in the visual and somatosensory components of the thalamus. This challenges the view that ascending sensory pathways are purely feedforward circuits, and suggests that we should regard the thalamus and cortex as reciprocally and highly interconnected circuits in which the cortex exercises some top-down control of perception.

Virtually all animals use hearing for localization and recognition of sounds, including behaviorally important communication sounds of individuals of the same species (conspecifics), such as mating and warning calls. For animals known as vocal learners, hearing is also necessary for learning to mimic sounds produced by others.

Humans are consummate vocal learners. Human speech involves fantastically complex and variable communication sounds, and we depend critically on hearing the speech both of ourselves and of others for normal speech development. Much is being learned about the auditory areas associated with human speech recognition and production from imaging and electrophysiological techniques. To understand basic brain mechanisms of vocal learning and its disorders, however, it is important to study animals with related behaviors. Surprisingly, there are relatively few other vocal learners.

Although the vocalizations of nonhuman primates can be complex, none have been shown to be learned. Among other mammals there is evidence for vocal learning only in cetaceans (whales and dolphins) and some bats. In striking contrast to this paucity of mammalian vocal learners, the many thousands of songbird species, as well as the parrot and hummingbird groups, all must learn to produce their complex songs. Songbirds have provided a particularly useful model, and there is a wealth of information on their vocal behavior and the underlying brain substrates, with some striking parallels to human speech learning.

Normal Vocal Behavior Cannot Be Learned in Isolation

What do we mean by vocal learning, and how can we demonstrate it? One indication of vocal learning is great variation in the vocal output of a species, such as the multitude of human languages and dialects. Similarly, many songbird species have local dialects (Figure 31–14A), and songs also vary between individual songbirds.

Because such variability could in theory be genetically encoded, a critical test of vocal learning is whether animals develop normal vocal behavior in the absence of hearing other individuals. It is well established that deaf children do not learn to speak normally. In addition, the rare cases of so-called wild children raised without exposure to human speech, such as the girl called Genie, show that speech does not develop normally when hearing is intact but no speech has been heard. In songbirds as well deafness early in life prevents normal song learning and production. Birds reared without exposure to conspecific songs sing highly simplified isolate songs, which nevertheless display some features characteristic of the species (Figure 31–14B, C). In contrast, monkeys reared without exposure to the vocalizations of other monkeys display essentially normal vocal behavior, as do non-songbirds, such as chickens or flycatchers, deafened or raised in isolation. Clearly both birdsong and speech are learned, with a strong dependence on hearing.

Hearing serves two independent functions in vocal learning. For one, hearing the vocalizations of others allows imitation. This aspect of learning is perceptual in both humans and birds; in songbirds a sensory memory of the tutor's song, often called the template, is rapidly formed and stored during a period called the sensory learning phase. However, neither humans nor songbirds can directly translate their sensory memory of others into a faithful vocal copy. Rather, they initially produce rambling and immature vocalizations, called babbling in humans and subsong in songbirds. They must then be able to hear these vocalizations to gradually refine them and make them match their desired vocal output during a process of sensorimotor learning.

Second, hearing provides vocal learners with sensory information about the accuracy of their motor performance, in the form of auditory feedback. This is especially clear in songbirds. They can memorize adult tutor songs with a relatively small amount of song exposure, and if isolated after such exposure will nevertheless produce a good copy of the tutor song so long as they are able to hear themselves. In contrast, if they are deafened after sensory learning, they produce highly abnormal songs, without even the innate structure usually present in isolate songs (Figure 31–14C). The same dependence on hearing of one's own voice is evident in humans. For instance, the speech of children who become deaf during childhood, even after significant speech learning, deteriorates markedly.

Of course, some aspects of birdsong are clearly not analogous to human speech. Although birdsong is used for communication, it does not seem to be language in the sense of conveying complex meaning. What it shares with speech is the learned sensorimotor control of an elaborate vocal system, with a pronounced dependence on hearing.

Vocal Learning Is Optimal During a Sensitive Period

A variety of evidence makes it clear that neither human nor avian vocal learners are born as "blank slates." Human babies as early as 3 weeks of age can make perceptual distinctions between categories of sounds in languages to which they have never been exposed. For instance, newborn babies in both the United States and Japan can distinguish the sounds "r" and "l," a distinction that does not exist in Japanese and is not well perceived by adult monolingual Japanese speakers (see Chapter 60). Similarly, newly hatched songbirds show innate recognition of and preference for all songs of their own species, relative to other species.

Acoustic experience quickly shapes the nervous system after birth. By 1 year of age children exposed only to Japanese no longer distinguish "r" and "l" sounds, and children being raised in English no longer distinguish the several classes of "h"-like sounds in Hindi or "shi"-like sounds in Mandarin. Similarly, young songbirds quickly begin to learn the songs to which they are actually exposed.

The ability of sensory experience to alter perceptual ability varies with age, as we saw earlier in connection with sound localization. During sensitive or critical periods for learning young children exposed to a language are able to speak the language fluently and without accent. However, their ability to do so decreases as they grow older. If not exposed to a foreign language by adolescence, most humans cannot learn to speak that language in a manner that is indistinguishable from native speakers, primarily in accent and grammatical usage. This inability seems to reflect decreased perceptual abilities, and not simply limits on the ability to produce the sounds of speech. Songbirds also have clear sensitive periods. In most species birds not exposed to song in the first several months of life go on to sing isolate song even if subsequently exposed to normal adult song; such birds are known as closed learners.

An animal model with sensitive periods for vocal learning like our own can provide insights into the factors that control the synaptic changes, and ultimately the neural mechanisms involved and thus potentially into remediation. For instance, the sensitive period for song learning does not have a strict age limit. Rather, experience itself is centrally involved in closing the sensitive period. Songbirds tutored only with songs from birds of other species (heterospecifics) can incorporate new songs from their own species at a time when birds raised with conspecifics no longer can. In some species birds reared in complete isolation can still incorporate new song elements as adults. Thus a lack of normal experience leaves the brain open to be shaped by the appropriate input for a longer than usual period.

Attentional or motivational factors also influence the timing of the sensitive period. Birds learn from live tutors for longer than they learn from taped tutors. Hormonal factors may play a role as well. Manipulations that delay the onset of singing and decrease testosterone levels appear to extend the sensitive period. In light of this it is intriguing that the human critical period for speech appears to end in adolescence.

Both Humans and Songbirds Possess Specialized Neural Networks for Vocalization

Along with the many behavioral similarities between speech and song learning, both humans and songbirds have evolved specialized neural systems for vocal learning and production. At first glance the very different organization in birds and humans, especially of forebrain areas, seems to make drawing any direct parallels between the two brain systems difficult.

One major difference between birds and humans is that birds do not have the multilayered cortex seen in all mammals, including humans. Rather, the evolutionarily related avian forebrain areas—derived like mammalian cortex from the embryonic region known as the pallium—are organized in nuclei, just as many lower areas are in both mammals and birds. A closer look at the vocal control systems of humans and songbirds, however, reveals numerous anatomical and functional similarities in the organization of neural pathways for vocal production and processing.

All primates and birds, and many mammals as well, have pathways from midbrain areas to lower respiratory and vocal motor centers as well as medullary centers that integrate respiratory and vocal tract control. Stimulation in these midbrain areas, such as the periaqueductal grey in primates and the dorsomedial intercollicular area in birds, can elicit well-formed vocalizations. However, a critical evolutionary step in the adaptation of learned vocalizations was the development of high-level forebrain areas that control the lower motor pathways for vocalization; both songbirds and humans have such areas whereas nonlearners do not.

In humans numerous perisylvian and parieto-temporal cortical areas are critical for speech production, as shown by the effects of lesions and the fact that stimulation of many of these areas can elicit vocalizations or disrupt ongoing speech. In striking contrast, there are no neocortical areas in monkeys from which vocalizations can be elicited by stimulation or whose ablation affects calls. Among birds, only songbirds and other vocal learners such as parrots have evolved an elaborate forebrain system, often called the song system, for control of the lower brain areas involved in vocalization (Figure 31–15A).

The motor control portion of this circuit consists of a chain of nuclei, including in part the premotor nucleus HVC, an erstwhile acronym used as a proper name, which contains a central pattern generator for song. HVC projects to a motor cortex-like area called the robust nucleus of the arcopallium (RA), which then connects to all the lower nuclei involved with vocal motor and respiratory control. Just as in the analogous human areas, stimulation or ablation of HVC or RA dramatically affects song. Moreover, neural recordings demonstrate directly the sequential premotor activity of neurons in these areas during singing.

It has become apparent that, in addition to neocortex or its avian equivalents, the thalamus, basal ganglia, and cerebellum are important in vocal production. Lesion, stimulation, and imaging studies of humans suggest that these areas are involved in fluency, volume, articulation, and rhythm of speech. Whether the cerebellum contributes to birdsong is still unknown, but it is very clear that a specialized loop known as the anterior forebrain pathway connects HVC and RA and is critical for song learning and plasticity throughout life. Numerous recent studies of this circuit are providing insights into the function of the basal ganglia not only in song plasticity but also in motor learning more generally. These functions include acting as a source of motor variability and biasing signals important for trial-and-error learning and modulating motor output in response to social cues.

Finally, both humans and songbirds have numerous high-level auditory areas associated with processing and recognition of complex vocalizations. Many of the areas described earlier receive such auditory inputs and in fact encompass sensorimotor rather than purely motor circuits. A crucial step in the evolution of specialized vocal control areas may have been the acquisition of auditory input by preexisting motor control areas in the forebrain, giving those auditory inputs the ability to change the vocal motor map. The close relatives of songbirds, the suboscine birds such as flycatchers and phoebes, can sing but display no vocal learning; these birds also show no evidence of a specialized forebrain song-control system. In humans the capacity to learn speech and the development of specialized cortical systems for its control may have resulted from close interaction of motor control areas for orofacial movements with a variety of areas involved in processing and memorizing complex sounds.

Songbirds Have Feature Detectors for Learned Vocalizations

Reflecting the critical importance of hearing in song learning, the song system contains some of the most complex auditory neurons known. The use of behaviorally relevant stimuli, such as the bird's own song and the songs of conspecifics, was critical to the discovery of these remarkable neurons. Song-selective neurons are found throughout the adult male song system and respond more strongly to an animal's own song, and in some cases to the tutor's song, than to other equally complex auditory stimuli, such as conspecific songs or the animal's own song played in reverse or with the syllables out of order (Figure 31–15B).

In addition, many of these neurons are combination-sensitive: They show a highly nonlinear increase in firing when the component sounds of the bird's own song are combined and played in the correct sequence, compared to those sounds played alone (Figure 31–15C). Feature detectors with such marked spectral and temporal sensitivity could provide feedback to song motor areas, in the form of their firing rate or pattern, about how well sounds match the bird's goal and when the correct sequence has been sung. It is tempting to speculate that analogous neurons with responsiveness to well-learned speech sequences exist in humans.

The song selectivity of these songbird neurons is not present in young birds, but emerges during the course of song learning. Early in the sensory learning phase the neurons respond equally well to all conspecific song stimuli. Over time their response to the bird's own song increases while responsiveness to other stimuli declines. In this way the neurons parallel the initial broad acoustic sensitivity of human infants, which is subsequently narrowed and shaped by individual auditory experience.

Song-selective neurons not only respond to complex sensory signals, but can also be active during motor production. For instance, the same neurons that respond to a particular song are also active during singing of the song. These neurons are thus another example of mirror neurons (Chapter 38). Because adult song is such a stereotyped motor act, song neurons may provide insight into the function of mirror neurons. For instance, there is a remarkable correspondence between these song neurons' auditory responses and their premotor activity: playback of one set of syllables triggers an auditory response that resembles the premotor activity for the next syllable in the song. Thus the auditory response could be considered a prediction of the motor command for the following syllable.

The sensorimotor, mirror properties of song-selective neurons raise the possibility that they are critically involved in linking sensory and motor representations in the song system; such action-perception coupling may be the function of mirror neurons more generally. In this intermixing of sensory and motor processing, birdsong is again reminiscent of human speech. Electrical stimulation of a single language area in humans can affect both production and perception of speech, and some cortical neurons respond differently to the same word depending on whether it is spoken by the subject or by someone else. In songbirds it should be possible to explore the mechanisms underlying this sensory-motor interaction.

The central auditory system transforms the firing patterns of eighth-nerve fibers into the neural signals required for the localization and recognition of sounds, and in some animals for shaping vocal output. As in other sensory systems, this processing is accomplished by detecting and extracting different features of sound and analyzing these in separate pathways, only later merging the results.

Sound localization, which is best understood, involves representing information about interaural timing and level differences and about the amplitude spectrum of sounds that reflects filtering by the ear, and associating these with locations in space in an experience-dependent way. The ability of visual cues to alter sound localization is a potent reminder that sensory systems evolved together to guide behavior: Animals localize sounds in order to find their sources.

Sound recognition is much less understood, and the multiplicity of cortical auditory areas involved suggests that analyzing it will be a challenge. The study of auditory specialists such as bats and their use of natural auditory stimuli has, however, taught us a great deal about what different auditory areas accomplish. This suggests that the continuing use of specialized animals and behaviorally relevant stimuli will be important in future discoveries. For primates, including humans, this will include the study of their own vocalizations. Songbirds, which show complex vocal learning with similarities to human speech acquisition, provide an additional source of insights into neural mechanisms of learning and the developmental regulation of plasticity.

The auditory system is the site of several critical periods, likely including that for speech learning, but can be coaxed into plasticity in adult animals. Demonstrated originally in experimental animals but now applied in humans as well, this result raises hope for new strategies to remediate deficits in human brain function.