From a phonatory point of view, voice is produced when the expiratory airflow from the lungs sets the closed vocal folds of the larynx into vibration, converting aerodynamic power generated by the thoracic and abdominal muscles (subglottal component) into the basic sound wave (e.g., acoustic power). This sound wave is further filtered and amplified by oral articulators, such as pharynx, tongue, palate, lips, and jaw (supraglottal component), and is emitted from the mouth and nose as sound of voice (Fig. 1A,B).

Voice onset is required for production of vowels and voiced consonants (e.g., b, d), while voice offset is necessary for production of voiceless consonants (e.g., p, t). During speech production, voice onset is precisely timed, which allows linguistic distinctions between voiced and voiceless consonants, such as /d/ versus /t/. Changes in the subglottal pressure due to changes in lung volume, the elastic properties of the chest wall and the active contraction of the intercostal and abdominal muscles lead to modulations of voice intensity, whereas the resonance characteristics of the supraglottal region (e.g., oral and pharyngeal cavities) influence the spectral properties of the sound.

Vocal fold movements are controlled by intrinsic and extrinsic laryngeal muscles. The intrinsic laryngeal muscles are confined to the larynx and participate in vocal fold closure (thyroarytenoid, TA, lateral cricoarytenoid, LCA, and interarytenoid muscles, IA), opening (posterior cricoarytenoid muscle, PCA), and lengthening (cricothyroid muscle, CT) (Fig. 1C,D). The extrinsic muscles connect the larynx with surrounding structures, such as the hyoid bone, sternum and pharynx, and raise or lower the larynx within the neck relative to the spine to modulate vocal fold length, fundamental frequency, oro-pharyngeal resonance frequencies and formant structure.

Voice control develops gradually throughout childhood for the control of speech production. As we mentioned earlier, the first vocalization occurs at birth as a shriek of the newborn. This and other types of non-verbal vocalizations in humans, such as an infant's cry and laughter, represent innate vocal reactions whose acoustic structure is genetically preprogrammed (Scheiner et al., 2004). This means that infants do not need to hear the sounds of cry or laughter from others in order to produce them. These basic innate vocalizations are controlled by the sensory and motor nuclei of the lower brainstem (e.g., ambigual, trigeminal, facial, hypoglossal, solitary tract nuclei) and spinal cord (thoracic and lumbar ventral horn), which are responsible for the basic coordination of laryngeal, respiratory and articulatory muscle activity (Fig. 2, subsystem I). The reticular formation of the pons and lower brainstem is another important structure for this type of vocal control. It plays a dual role in establishing connections between different phonatory nuclei and in coordinating basic vocal motor activities. Single-unit recording studies in the squirrel monkey have found that some vocalization-correlated neurons in the reticular formation are active only during specific vocalizations, while other neurons fire during various types of vocalizations or change their discharge rate based on the rhythmic frequency modulations of produced vocalizations (Kirzinger and Jurgens, 1991; Luthe et al., 2000). Furthermore, the reticular formation dorsal to the superior olive contains a vocalization pattern generator, the neuronal activity of which co-occurs with the neuronal firing of the phonatory ambigual, facial and motor trigeminal nuclei (Hage and Jurgens, 2006).

The involvement of the forebrain is not essential for production of innate vocalizations. It has been reported that even anencephalic infants, who lack the entire forebrain but have an intact brainstem, are still able to vocally react to painful stimuli (Monnier and Willi, 1953). However, in older children, innate vocalizations come under voluntary control, sometimes involving mimicking of vocal utterances. For example, cry can be produced in the absence of pain or suppressed in the presence of pain. For this higher level of vocal control, that is, voluntary initiation of innate vocalizations and control of their emotional status, the brainstem phonatory nuclei and the vocal pattern generator require an input from the higher brain regions, such as the periaqueductal gray (PAG) and ACC (Fig. 2, subsystem II). The PAG receives direct projections from the ACC as well as from other cortical and subcortical regions controlling limbic, sensory, motor, cognitive and arousal systems (Dujardin and Jurgens, 2005). The PAG projects to the reticular formation of the lower brainstem, thus representing an neuroanatomical and functional relay station within the ACC-PAG-brainstem pathway. The PAG plays primarily a gating role in triggering a vocal response and modulating its intensity, while the ACC is involved in voluntary control of voice initiation and its emotional intonation. Destruction of the ACC results in a loss of voluntary control of emotional intonations during speaking, while lesions in the PAG lead to mutism. Interestingly, destruction of the ACC does not interfere with the production of vocalizations elicited at the level of the PAG and brainstem reticular formation; however, destruction of the PAG abolishes vocalizations from the ACC but not from the brainstem reticular formation (Jurgens, 2002). Thus, while the ACC and PAG are important in shaping voluntary control of voice initiation and emotional modulation, the reticular formation represents the basic executing level of this pathway.

The highest level within the hierarchy of voice controlling system is represented by the LMC and its input and output structures (Fig. 2, subsystem III). The ability to learn new vocalizations is very limited, perhaps impossible, in non-human primates and other mammals (except whales and dolphins). Their vocalizations are exclusively emotional, with each type of vocalization conveying only one meaning. In humans, in contrast, the development of the ability to produce innumerable learned vocal utterances for speech and song depends on the fine motor control by the LMC. While lesions in this region have almost no effects on monkey vocalizations, they abolish speech production in human patients (Jurgens, 2002). Such patients are occasionally able to initiate phonation, such as grunts, wails and laughs, but do not succeed in voluntary modulations of pitch, intensity and the harmonious quality of their vocalizations. The preservation of non-verbal vocalizations in patients with bilateral damage to the LMC may be due to parallel organization of the LMC and ACC-PAG pathways, one controlling voluntary voice production and the other controlling the initiation of basic vocal reactions.

First observations of the larynx representation within the motor cortex were made in the 1930s. Shortly after the report by Oscar Foester that bilateral vocal fold movements in humans can be produced with electrical stimulation of the motor strip of one hemisphere (Foester, 1936), Wilder Penfield and colleagues described the representation of vocalization in the inferior portion of the motor cortex above the jaw and below the lip muscles representations (Penfield and Bordley, 1937) (Fig. 3A). Nearly the same somatotopy was observed in the monkey brain with the representation of the larynx region just above the Sylvian fissure (Woolsey et al., 1952) (Fig. 3B). On the other hand, electrical stimulation studies failed to identify the larynx representation in the motor cortex of lower mammals, such as the dog and cat (Milojevic and Hast, 1964).

Although the implications of the LMC in the control of voice production is significantly different between humans and non-human primates, much of our understanding about the organization of the LMC comes from research in non-human primates, primarily in the rhesus and squirrel monkeys. The LMC in these species is situated between the inferior branch of the arcuate sulcus anteriorly and the subcentral dimple posteriorly (Hast et al., 1974) (Fig. 3D). This region contains small populations of neurons that are selectively active during either conditioned or spontaneous vocalizations (Coude et al., 2009). Among these, the majority of neurons specific for conditioned vocalization discharge before voice onset, while a smaller number of neurons are time-locked with voice onset.

The larynx area is surrounded by the representations of the tongue, lip and masticatory muscles and is considered to be part of the primary motor cortex, although cytoarchitectonically it corresponds to premotor cortex, area 6, in non-human primates (Jurgens, 1974; Simonyan and Jurgens, 2002). Electrical stimulation of this region with simultaneous laryngeal electromyographic (EMG) recordings in the rhesus monkey revealed that the intrinsic and extrinsic laryngeal muscles have separate representations within the LMC, i.e. topographical organization (Hast et al., 1974) (Fig. 3D).

More than half a century later after Foester's and Penfield's seminal observations, studies of the LMC in humans have been recently revived with the advent of non-invasive neuroimaging techniques. Recent investigations in humans have shown that the LMC is located more dorsally from the Sylvian fissure (Khedr and Aref, 2002; Rodel et al., 2004) than originally proposed (Penfield and Bordley, 1937). Similar to non-human primates, the human LMC is organized topographically, although, compared to the monkey's brain, the representation of the laryngeal muscles is reversed, with the CT muscle located more medially than the TA muscle (Rodel et al., 2004). However, in contrast to the monkey, recent studies have substantially revised the location of the LMC in humans. From the earlier works of Penfield and colleagues, it is notable that the motor homunculus identifies a representation of a behavior (vocalization) rather than an organ (larynx) in the inferior portion of the motor cortex (Fig. 3A). As behavior, vocalization requires not only laryngeal muscle activity but also orchestrated activity of the respiratory and orofacial movements. Hence, it is not surprising that, on the map of the motor homunculus, vocalization as behavior, in contrast to the single organ representation, encompasses quite a large region within the motor cortex, overlapping with the lip, jaw and tongue representations, without, however, a specific reference to the larynx. Thus, the location of the human LMC remained largely unknown until recently. Neuroimaging studies using different vocal tasks were finally able to separate and localize the laryngeal region within the primary motor cortex (Brown et al., 2008; Loucks et al., 2007; Rodel et al., 2004; Simonyan et al., 2009; Wilson et al., 2004), predominantly in the area 4p (Fig. 3C).

A series of recent investigations have shown that both humans and non-human primates share a common network of extensive cortical and subcortical connections with the LMC (Fig. 4). Nearly all cortical connections of the LMC are reciprocal, that is the LMC both receives and sends information to these regions. The regions reciprocally connected with the LMC are the surrounding ventral and dorsal premotor, primary motor and somatosensory cortices at their orofacial and trunkal representations, inferior frontal gyrus (IFG), ventrolateral and dorsolateral prefrontal cortex, insula, cingulate cortex (CC), supplementary motor area (SMA), angular (AG), supramarginal (SMG), middle (MTG) and superior temporal (STG) gyri, claustrum, ventral and mediodorsal thalamus, medial parabrachial nucleus, deep mesencephalic nucleus and locus coeruleus. The only cortical region projecting to the LMC without receiving an input from it is the orbital cortex, which is assumed to have only a very indirect involvement in vocal control (Price, 1996), as well as subcortically the ventral tegmental area, substantia nigra and raphe nucleus. On the other hand, the cortical regions that receive projections from the LMC but do not send connections back are the inferior parietal cortex, posterior parietal operculum and cortex within the intraparietal sulcus, as well as subcortically the putamen, caudate nucleus, globus pallidus and brainstem nuclei (i.e., reticular formation and spinal trigeminal, solitary tract and facial nuclei).

Functionally, the LMC is connected with most of these structures to fulfill its main task - the voluntary control of voice production. Interconnections of the LMC with the surrounding somatosensory cortex and inferior parietal cortex are important for integration of proprioceptive and tactile feedback from the orofacial, respiratory and laryngeal regions during voice production. Neuroimaging studies of overt and covert voice and speech production have consistently reported activation of the primary sensorimotor cortex (e.g., (Bohland and Guenther, 2006; Horwitz et al., 2003; Loucks et al., 2007; Riecker et al., 2000). Activation in this region is organized somatotopically with the orofacial and laryngeal activation occupying the ventral portion of the sensorimotor cortex and the respiratory activation represented as two discrete loci in the dorsal region of the trunk representation and in the ventral region overlapping with the laryngeal and orofacial activation (Brown et al., 2009; Loucks et al., 2007; Simonyan et al., 2007). Interestingly, only voluntary exhalation, which is required for voice production, but not inhalation elicits activation of both dorsal and ventral sensorimotor cortex (Ramsay et al., 1993). The orchestration of simultaneous laryngeal, orofacial and respiratory muscle activity at the cortical sensorimotor level becomes apparent when one of its components is voluntarily modulated or altered. A recent study of functional LMC networks during syllable production showed that when subjects were asked to produce voice with minimal orofacial movements and at the similar breathing depth, that is the modulation of two out of three components of voice production system, the LMC displayed reduced functional connectivity with both orofacial and respiratory sensorimotor cortices to maintain specific level of task production (Simonyan et al., 2009). The influences of the somatosensory cortex on the motor cortex become even more apparent when the auditory feedback, another online monitoring system of correct voice and speech production, is altered and the encoding of auditory input is challenged, for example, in the presence of very loud noise (Guenther, 2006; Peschke et al., 2009). These observations suggest that the ventral sensorimotor region contains not only a simple ensemble of motor and sensory components of the voice controlling system (e.g., laryngeal, orofacial and respiratory), but it also acts as a higher-level centralized region, coordinating exhalatory airflow necessary for setting the vocal folds into vibration and modulating the orofacial articulators for voluntary production of different types of vocal behaviors.

With respect to the inferior parietal cortex, including the SMG and AG, this region represents one of a few higher-order sensorimotor centers for coordination of both speech production and comprehension. The inferior parietal cortex is known to be involved in complex phonological and semantic processing and monitoring of the verbal response to suppress phonemic errors (e.g., (Fiebach et al., 2007; Hocking et al., 2009; Zheng et al., 2010)).

The LMC requires an input from the IFG for motor planning of voice and speech production. The IFG is another brain region where the control of speech production and comprehension converge. Its pars orbitalis is involved in the retrieval of semantic information (de Zubicaray and McMahon, 2009; Tyler et al., 2010) and its pars opercularis is responsible for hierarchical sequencing of linguistic and non-linguistic sequences (Tettamanti et al., 2009; Willems et al., 2009) and articulatory preparation to speech production (Papoutsi et al., 2009; Zheng et al., 2010). Because the IFG is prominent in the processing of information necessary for long-term preparation of learned oro-motor sequences, it is not surprising that the IFG is usually active during production of long sequences of syllables and words (e.g., (Horwitz et al., 2003; Ozdemir et al., 2006; Wise et al., 1999) and only rarely during production of single syllables (Brown et al., 2008; Ghosh et al., 2008; Loucks et al., 2007). Despite the fact that the IFG activation is not always observed in studies of differential complexity of voice and speech production, the strong functional and structural connectivity between the IFG and LMC seems to be responsible for motor preparation and processing of all components of speech production (e.g., laryngeal, orofacial, or respiratory) (Greenlee et al., 2004; Simonyan et al., 2009).

The reciprocal connections of the LMC with the SMA are needed for preparation for vocal motor command execution. Accordingly, it has been shown that the offset of the late vocal-related cortical potentials in the SMA precedes those in the LMC pointing to the involvement of the LMC only after the motor preparatory phase in the SMA (Galgano and Froud, 2008). The SMA is active during various voluntary laryngeal tasks (e.g., (Ghosh et al., 2008; Loucks et al., 2007; Ozdemir et al., 2006; Simonyan et al., 2007). More specifically, it has been shown that activation in the anterior pre-SMA was related to effortful word selection, activation in the posterior pre-SMA was related to the control of syllable sequencing, and activation in the SMA proper was related to overt articulation (Alario et al., 2006). Recently, the SMA has also been identified as a motor component of the speech monitoring network (van de Ven et al., 2009). Furthermore, electrical stimulation of this region has been reported to elicit vocalizations in humans (Penfield and Welch, 1951) but not in monkeys (Jurgens, 2002). Similarly, bilateral lesions in the SMA have no effect on monkey vocalizations. In contrast, in humans, such lesions severely reduce motivation to employ propositional speech, whereas nonpropositional (automatic) speech remains almost intact (so-called transcortical motor aphasia). Functional connections between the LMC and SMA are left-lateralized and are particularly stronger with the pre-SMA during learned voice production compared to innate laryngeal behavior, such as breathing (Simonyan et al., 2009). It, thus, appears that the connectivity between the LMC and the pre-SMA, a region responsible for processing of higher order motor plans for subsequently ordered movement execution (Matsuzaka et al., 1992; Shima and Tanji, 1998; Tanji and Shima, 1996), plays a central role in sequencing and initiation of complex learned vocal movements during speech production.

It is important to note that the LMC has a reciprocal structural connection with the ACC but not with the PAG as identified based on neuroanatomical tract tracing studies in non-human primates. The lack of connections between the LMC and PAG fits well with the separation between the two descending pathways controlling voice production, limbic and motor cortical. Similarly, in the monkey, the PAG blockade has no effect on vocal fold movements elicited from the LMC as oppose to the drastic effects on vocalization elicited by limbic structures (Jurgens, 2002). The LMC, thus, seems to control the voluntary voice production via a pathway by-passing the PAG, although it establishes connections with the ACC for shaping emotional intonations during speech production. As noted earlier, the connection between the LMC and ACC serves as one of the two links at which two pathways controlling innate emotional and learned vocalizations converge.

Another link between the limbic and motor cortical vocal pathways is present at the level of the brainstem reticular formation, specifically in its dorsal and parvocellular reticular nuclei, which are further directly connected with laryngeal motoneurons in the nucleus ambiguus, the articulatory motoneurons in the trigeminal motor, facial and hypoglossal nuclei, and the expiratory motoneurons in the thoracic and upper lumbar spinal cord (Bernard et al., 1990; Thoms and Jurgens, 1987; VanderHorst et al., 2001). Combined electrical stimulation and lesioning studies have shown that bilateral lesions in this region block vocalizations elicited from the PAG and vocal fold movements elicited from the LMC (Jurgens and Ehrenreich, 2007; Shiba et al., 1997). Because both limbic and motor cortical pathways come together in the reticular formation of the brainstem, it has been suggested that this region is involved in vocal motor coordination of both innate and learned voice production (Jurgens and Ehrenreich, 2007). Its functional properties are more important in vocal motor control in non-human primates than in humans due to the lack of direct projections from the LMC to the nucleus ambiguus in the former species and, therefore, their reduced ability to directly modulate activity of brainstem laryngeal motoneurons (Jurgens, 1976; Simonyan and Jurgens, 2003). In humans, in contrast, direct connections do exist between the LMC and the nucleus ambiguus (Iwatsubo et al., 1990; Kuypers, 1958). This direct LMC-ambigual connection together with the LMC representation in the primary motor cortex seem to be very recent acquisitions in hominid evolution and may be regarded crucial as prerequisites for speech control.

In addition to direct LMC-brainstem laryngeal motoneurons connections, the LMC establishes a widespread network of subcortical connections, most of which project further down to the brainstem phonatory motoneurons. The putamen receives the strongest of all telencephalic subcortical projections from the LMC, thus representing the main basal ganglia output structure of the LMC. Putaminal lesions cause dysarthria and dysphonia in humans but have no effect on monkey's vocalizations (Jurgens, 2002), suggesting the involvement of the putamen only in learned voluntary voice and speech production but not in production of innate vocalizations. Motor cortical connections to the putamen are somatotopically organized with the leg area projecting to the rostrodorsal putamen, the arm and trunk occupying its central part, the face area projections to the caudoventral putamen, and the larynx area connecting with the ventral putamen over a large anterio-posterior extent (Kunzle, 1975; Simonyan and Jurgens, 2003). It appears that there is an overlap of projections from the face and larynx areas in the posterior (postcommissural) putamen (i.e., part of the sensorimotor functional striatal loop), whereas the putamen rostral to the anterior commissure (i.e., part of the associative and limbic striatal loops) receives input only from the LMC. Direct connections of the LMC with all striatal functional subdivisions, including the sensorimotor, associative and limbic loops, may represent important factors in the integrative control of different aspects of speech production, ranging from motor control to motivation and cognitive processing of speech.

Another subcortical region that is directly connected with the LMC is the thalamus, specifically its ventral lateral, ventral posterior, medial, and centromedian groups of nuclei. Single-unit recordings in the ventral thalamus have shown voice and speech-related neuronal activity (Farley, 1997; McClean et al., 1990). Furthermore, the ventral thalamus has reciprocal connections with the LMC, IFG and SMA, electrical stimulation of which also produces vocalization in humans (Penfield and Roberts, 1959). Besides the direct role of the ventral thalamus in vocal sensorimotor coordination of learned voice production, it represents an important relay station of the cortico-striato-pallido-thalamo-cortical and cortico-ponto-cerebello-thalamo-cortical loops for integration of sensorimotor information from the basal ganglia and cerebellum. Both pathways, finally, connect the LMC with the laryngeal motoneurons after synapsing in the reticular formation of the brainstem.

This work was supported by the Extramural and Intramural Programs of the National Institute on Deafness and Other Communication Disorders, National Institutes of Health (R00DC009620 to K.S.).