Stimulus Complexity and Categorical Effects in Human Auditory Cortex: An Activation Likelihood Estimation Meta-Analysis

Investigations of the functional organization of human auditory cortex typically examine responses to different sound categories. An alternative approach is to characterize sounds with respect to their amount of variation in the time and frequency domains (i.e., spectral and temporal complexity). Although the vast majority of published studies examine contrasts between discrete sound categories, an alternative complexity-based taxonomy can be evaluated through meta-analysis. In a quantitative meta-analysis of 58 auditory neuroimaging studies, we examined the evidence supporting current models of functional specialization for auditory processing using grouping criteria based on either categories or spectro-temporal complexity. Consistent with current models, analyses based on typical sound categories revealed hierarchical auditory organization and left-lateralized responses to speech sounds, with high speech sensitivity in the left anterior superior temporal cortex. Classification of contrasts based on spectro-temporal complexity, on the other hand, revealed a striking within-hemisphere dissociation in which caudo-lateral temporal regions in auditory cortex showed greater sensitivity to spectral changes, while anterior superior temporal cortical areas were more sensitive to temporal variation, consistent with recent findings in animal models. The meta-analysis thus suggests that spectro-temporal acoustic complexity represents a useful alternative taxonomy to investigate the functional organization of human auditory cortex.

frequency sinusoidal wave (pure tone), constant over time, can be classified as simple, and a sound containing multiple components can be classified as complex with respect to the frequency domain. Examples of sounds with high spectral complexity are musical notes or sustained vowels. Similarly, a sound with acoustical structure varying over time can be classified as complex with respect to the time domain. Examples of stimuli with high temporal complexity are frequency or amplitude modulated sounds or sound sequences. Natural sounds can be complex with regards to both their frequency composition and temporal variation. Phonemes, the basic units of speech, contain multiple frequency components, called formants, which may be combined over time to produce syllables and words. Similarly, musical sequences are composed of complex changes in fundamental frequency and harmonic structure that unfold over time. Additionally, speech processing is mainly dependent on temporal information (Shannon et al., 1995), while spectral composition is most relevant for music perception (Warrier and Zatorre, 2002). Hence, acoustic complexity is not independent of sound categories and the two classification methods explored here should not be considered as mutually exclusive.
As previously proposed, an auditory stimulus can be categorized in more than one way; either based on a priori knowledge about the characterizing features of the sound source or on the basis of a sound's acoustic pattern in the frequency and time domain (Griffiths and Warren, 2004). Additionally, some studies suggest

INTRODUCTION
Current accounts of the functional organization of auditory cortex, mostly based on response specificity to different sound categories, describe an organizational structure that is both hierarchical and hemispherically specialized (Rauschecker, 1998;Hackett, 2008;Rauschecker and Scott, 2009;Woods and Alain, 2009;Recanzone and Cohen, 2010).
Characterizing responses to stimuli from typical auditory categories such as music, voices, animal, or environmental sounds have provided important information about the cortical specialization for auditory processing. However, this classification may not fully account for the range of stimulus variability encountered across neuroimaging studies, as most stimuli do not fit neatly into one auditory category. For instance, an amplitude modulated tone can vary in ways that cannot be adequately characterized using typical categories. However, its characteristics can easily be described in terms of variations in time (temporal dimension) and frequency (spectral dimension), suggesting an alternative approach to stimulus classification. Accordingly, any auditory stimulus can be described with respect to its sound complexity characteristics specified with respect to changes in time and frequency. This approach represents a comprehensive characterization of sounds that is not limited to specific categories. Therefore, complexity might represent an alternative organizing principle along which to represent auditory cortical response specialization. In this conceptualization, a single activity in primary auditory cortex while complex sound processing is associated with activity including both primary and non-primary auditory cortex. We were also interested in examining whether there was meta-analytic evidence for distinctive patterns of hemispheric specialization for music and vocal sounds.
Next, we more closely examined vocal stimuli and a particular subcategory of vocal sounds: intelligible speech. Vocalizations constitute an ecologically central sound category that includes all sounds with a vocal quality irrespective of phonetic or lexical content. Examples include speech in various languages, non-speech affective vocalizations (e.g., laughter), and laboratory-engineered sounds, such as time-reversed speech, that exhibit distinctly vocal qualities. Vocal sounds include, but are not limited to, intelligible speech. Based on previous findings, we expected to observe bilateral superior temporal gyrus (STG) and the superior temporal sulcus (STS) activity related to vocal sounds Kriegstein and Giraud, 2004), and anterior STG and STS activity on the left related to speech intelligibility (Benson et al., 2006;Uppenkamp et al., 2006).
Finally, we examined whether acoustic complexity, estimated from variations in time (temporal) and frequency (spectral) dimensions, represents a relevant organizing principle for functional response specificity in human auditory cortex. In terms of spectral composition, stimuli can have single or multiple frequency components. In the temporal dimension, stimuli can be characterized as unchanging or, for those containing temporal changes, having either regular or irregular changes. Using this classification, we characterized the cortical response related to each level of acoustic complexity. Then, by comparing the "multiple" to the "single" categories, independent of the temporal changes, and the "changing" to the "unchanging" categories, independent of the spectral composition, we isolated the cortical activity related to variations in the frequency and time dimensions, respectively.

INCLUSION Of STUDIES
A preliminary list of articles was identified using several Medline database searches including both articles published prior to March 2010 [keywords: positron emission tomography (PET), functional magnetic resonance imaging (fMRI), auditory, sound, hear*, speech, and music] and lists of citations within those articles. Studies were included if they fulfilled specific inclusion criteria: (1) the study was published in a peer-reviewed journal; (2) the study involved a group of healthy typical adult participants with no history of hearing, psychiatric, neurological, or other medical disorders, (3) the subjects were not trained musicians; (4) the auditory stimuli were delivered binaurally, with no inter-aural delay because of our focus on non-spatial auditory processing; (5) the task-related activity coordinates were reported in standardized anatomical space; (6) the study used whole-brain imaging and voxel-wise analysis. As our main goal was to determine the spatial distribution within auditory cortical regions, the few studies using incomplete brain coverage, but that included the temporal cortex were not excluded (Binder et al., 1996(Binder et al., , 2000Belin et al., 1999;Celsis et al., 1999;Hugdahl et al., 2003;Stevens, 2004;Schönwiesner et al., 2005;Zaehle et al., 2007). Additionally, some studies specifically included subcortical structures (Griffiths et al., 1998;Hwang et al., 2007; Mutschler that auditory cortex activation to sounds of a given category could reflect a specialized response to the acoustic components characterizing sounds within this category (Lewis et al., , 2009). This suggests a certain level of interaction between the cortical processes involved in the analysis of acoustic features and those showing sensitivity to sound categories. However, recently Leaver and Rauschecker (2010) demonstrated categorical effects of speech and music stimuli even when controlling for changes in spectral and temporal dimensions. The two classification approaches are therefore not mutually exclusive and both methods seem relevant and can complement each other in revealing different aspects of cortical auditory specialization. In vision, cortical representation of stimulus complexity has been described with simple (first-order) information being analyzed within primary visual cortex (V1) and complex (second-order) information processing involving both primary and non-primary visual cortex (V2/V3; Chubb and Sperling, 1988;Larsson et al., 2006). Given that parallels have often been drawn between visual and auditory cortical functional organization (Rauschecker and Tian, 2000), we were interested in examining how well characterization of sounds by their acoustic complexity might reflects new insights into regional functional specialization.
Given that auditory neuroimaging studies exhibit a high degree of stimulus and task heterogeneity, their individual cortical activity patterns are not easily integrated to obtain an unambiguous picture of typical human auditory cortical organization. Neuroimaging meta-analysis offers a potential solution to this problem as it estimates the consistency of regional brain activity across similar stimuli and tasks, providing a quantitative summary of the state of research in a specific cognitive domain (Fox et al., 1998), estimating the replicability of effects across different scanners, tasks, stimuli, and research groups. By revealing consistently activated voxels across a set of experiments, meta-analysis can characterize the cortical response specificity associated with a particular type of task or stimulus (Wager et al., 2009). Activation Likelihood Estimation (ALE) is a voxel-wise meta-analysis method that provides a quantitative summary of task-related activity consistency across neuroimaging studies (Turkeltaub et al., 2002).
In the current study, we use quantitative ALE meta-analysis to examine the spatial consistency of human auditory processing, classified using either conventional sound categories or acoustic complexity. Given the focus of our study on stimulus complexity effects, we excluded studies of spatial auditory processes including localization, and inter-aural delay, as well as those including complex tasks.
First, we classified sounds using typical auditory categories to examine the evidence supporting hierarchically and hemispherically lateralized functional organization for auditory cortical processing. Hierarchical auditory processing has been described as sensitivity to stimulus complexity increasing from primary to non-primary auditory cortex, with simpler perceptual features represented at primary levels (Wessinger et al., 2001;Hall et al., 2002;Scott and Johnsrude, 2003). Relative hemispheric specialization is reflected by predominantly left-hemisphere processing for speech sounds and stronger right-hemisphere responses to music (for a review see . We used typical sound categories, such as pure tones, noise, music, and vocal sounds, to classify auditory material to see if simple sound processing is associated with spectral component numbers; 10 contrasts, 57 foci), (5) "multiple, regular change" (tone sequences and increasing click rate sequences; 6 contrasts, 41 foci), or (6) "multiple, irregular change" (vocal sounds, music, or environmental sounds; 70 contrasts, 517 foci). Each task contrast was classified using the stimulus description provided in each study. Contrasts resulting from covariate effects of a parameter of interest were classified according to parameter complexity. For instance, effects related to parametric increases in temporal modulation rate were assigned to the "single, regular change" complexity level (Schönwiesner et al., 2005). Ambiguous contrasts were excluded from analysis. For example, we did not classify contrasts that used comparison stimuli that had acoustic complexity comparable to the stimuli of interest (Zatorre et al., 1994;Griffiths et al., 1998;Blood et al., 1999;Mummery et al., 1999;Warren and Griffiths, 2003;Giraud et al., 2004;Schwarzbauer et al., 2006;Peretz et al., 2009) nor those using stimuli that could be assigned to more than one complexity level, such as notes, chords, or chord progressions (i.e., stimuli including note/chord/chord progression; Benson et al., 2001).

ALE META-ANALySIS
After the task-related activity maxima were classified, ALE maps (Turkeltaub et al., 2002) were computed using GingerALE 1.1 (Laird et al., 2005). Coordinates reported in MNI space were converted to Talairach space using the Lancaster transform icbm2tal (Lancaster et al., 2007). ALE models uncertainty in localization of each activation focus as a Gaussian probability distribution, yielding a statistical map in which each voxel value represents an estimate of the likelihood of activity at that location, utilizing a fixed effects model for which inferences should be limited to the studies under examination. Critical thresholds for the ALE maps were determined using a Monte Carlo style permutation analysis of sets of randomly distributed foci. A FWHM of 10 mm was selected for the estimated Gaussian probability distributions. Critical thresholds were determined using 5000 permutations, corrected for multiple comparisons (p < 0.01 false discovery rate, FDR; Laird et al., 2005) with a cluster extent of greater than 250 mm 3 . In order to present results in the format most commonly used in the current literature, the ALE coordinate results were transformed into MNI standard space using the Lancaster transform (Lancaster et al., 2007), while ALE maps were transformed by applying spatial normalization parameters obtained from mapping from Talairach to MNI space.

ANALySIS USINg CLASSIfICATION by TypICAL AUDITORy CATEgORIES
First, ALE maps were computed for each of the four typical auditory categories: pure tones, noise, music and vocal sounds. Each resulting map shows regions exhibiting consistent activity across studies for each sound category. For example, the "music" map shows the voxel-wise probability of activity for all "musical stimuli vs. baseline" contrasts.
Next, we examined hemispheric specialization effects by directly comparing the "music" and "vocal" sound categories. We directly compared a random subsample of the "vocal" sounds category ( Table A1 in Appendix; 20 contrasts, 156 foci) to the "music" category (10 contrasts, 175 foci). This procedure ensured that the resulting ALE maps would reflect activity differences between studies rather than the imbalance in coordinate numbers et al., 2010); and (7) the study had to include passive listening or a simple response task, such as a button press at the end of each sound to assess the participants' attentive state, task characteristics that tended to minimize the inclusion of activity related to top-down processes or task difficulty (Dehaene-Lambertz et al., 2005;Dufor et al., 2007;Sabri et al., 2008).
Of over 7000 articles retrieved, 58 (19 PET and 39 fMRI) satisfied all inclusion criteria and were included in the analysis ( Table 1). Several studies reported activity from multiple task and control conditions. For our analysis, only conditions incorporating either no overt task or a simple task used to maintain attention were considered. To maintain consistency among the control conditions, only task contrasts with a low-level baseline (silence, tone, or noise) were included. For some studies, more than one contrast satisfied our criteria and all were included in the analysis. This procedure was employed to maximize the sensitivity of the analysis, but could potentially bias the results toward samples for which more than one contrast was included.

CONTRAST CLASSIfICATION pROCEDURE
One hundred seventeen contrasts, including 768 foci, met the inclusion criteria. These contrasts were classified first by typical sound categories and then according to their variation along either the frequency or time dimension (Table 1).
For the first method, each contrast was classified with respect to one of the typical sound categories: simple sounds or pure tones (9 contrasts, 22 foci), noise (4 contrasts, 31 foci), music (10 contrasts, 175 foci), and vocal sounds (62 contrasts, 370 foci). The pure tones category included only contrasts of single tones vs. silence; the noise category included white, pink, and brown noise (Rimol et al., 2005), noise bursts (Zatorre et al., 1992), and the combination of multiple reversed environmental sounds . Melodies, notes, chords, and chord progressions were classified as music. Finally, all sounds with a vocal quality (syllables, words, voices, reversed words, or pseudowords) were included in the vocal sounds category. Ideally, we would have included other commonly used sound categories such as animal or environmental sounds; however the number of contrasts falling into these categories was not sufficient for quantitative meta-analysis, with only one contrast presenting environmental sounds and only two falling into the animal sound category. The remaining contrasts (30/117), including modulated tones, frequency sweep, harmonic tones, or recorded noise, were not included in this analysis because they did not neatly fit into one sound category.
For the second method, we classified the stimuli with respect to their acoustic features. Two levels of complexity were defined using the frequency dimension (single and multiple frequency components) and three levels in the time domain (unchanging, regular periodic change, or irregular change). Therefore, task contrasts were classified in one of six complexity levels depending on their frequency-and time-related acoustic features (Table 1; Figure 5A): (1) "single, unchanging" (single tone; 9 contrasts, 22 foci), (2)"single, regular change" (frequency or amplitude modulated tone, single formant frequency sweep, parametric variation of modulation rate or rate of presentation; 8 contrasts, 38 foci), (3) "single, irregular change" (1 contrast, 4 foci), (4) "multiple, unchanging" (harmonic tone, square wave tone, vowel, noise, or parametrically increasing      Dalla Barba et al. (1998)                  foci), a randomly selected subsample of 10 contrasts (70 foci, see Table A1 in Appendix) were selected from this level of complexity to facilitate comparison of activity extent between levels. Next, we examined effects related to auditory complexity. For the frequency domain, all contrasts falling in the "multiple" level (26 contrasts, 168 foci) were directly compared to those in the "single" level (18 contrasts, 64 foci), independent of their variation over time, (Figure 5A, bottom row vs. top row, green arrow). For the time dimension, comparisons were made between the contrasts including stimulus changes over time (regular and irregular; 25 experiments, 153 foci) and those who did not (unchanging; 19 contrasts, 79 foci), independent of their frequency composition (Figure 5A, middle and right column vs. left column, blue arrow).

STIMULUS CLASSIfICATION USINg TypICAL AUDITORy CATEgORIES
We observed different patterns of activity corresponding to the typical sound categories of pure tones, noise, music, and vocal sounds (Figure 1; Table 2). For all the categories, the strongest effects were found in auditory cortex (Brodmann areas 41,42,and 22). For the pure tone map, high ALE values were found bilaterally in medial Heschl's gyri (HG). The noise map revealed effects in right medial HG and bilaterally in STG posterior and lateral to HG. Effects related to music were seen in HG, anterior and posterior STG. Finally, vocal sounds elicited large bilateral clusters of activity in HG as well as anterior, posterior, and lateral aspects of the STG. While pure tone effects were restricted to auditory cortex, effects outside temporal cortex were observed for the other categories. Additional activity was seen in frontal cortex for noise (BA 6,9), between those categories (Laird et al., 2005). Then, as lateralization effects are reported for intelligible speech rather than vocal sounds, only contrasts using intelligible speech with semantic content, such as words or sentences, were included. The "music" and the "speech" categories were directly compared to investigate the expected lateralization effects. Given that many contrasts fell into the intelligible speech category, we selected only one contrast per study (Table A1 in Appendix), including a total of 27 contrasts (166 foci).
Finally, we assessed cortical auditory specialization for processing intelligible speech. Given that specialized auditory processes can be more easily isolated when the contrasting stimuli are as close as possible to the stimuli of interest in terms of acoustic complexity (Binder et al., 2000;Uppenkamp et al., 2006), contrasts containing unintelligible spectrally and temporally complex sounds were used as for comparison. Thirteen contrasts (76 foci, see Table A1 in Appendix) selected included reversed words, pseudowords, recorded scanner noise, single formant, environmental sounds, and modulated complex sounds. We directly compared the intelligible speech and complex nonspeech sound categories.

ANALySIS USINg CLASSIfICATION by AUDITORy COMpLExITy
To investigate the relevance of acoustic complexity as a stimulus property predicting functional auditory specialization, we computed ALE maps for each level of complexity. Given that only one contrast fell into the "single, irregular change" dimension, this analysis was not conducted. Moreover, as most of the contrasts were classified as "multiple frequencies, irregular modulation" (70 contrasts, 517 FIguRe 1 | Activation Likelihood estimation maps showing clusters of activity related to sound categories: pure tones, noise, music, and vocal sounds. Maps are superimposed on an anatomical template in MNI space. Axial images are shown using the neurological convention with MNI z-coordinate labels (p FDR < 0.01). music (BA 4,6,44,45,46), and vocal sounds (BA 45). Effects were observed in cerebellum for noise and music as well as in the anterior cingulate gyrus for vocal sounds. Effects related to typical sound categories were lateralized. Qualitative examination revealed larger clusters in right auditory cortex for music and in left auditory cortex for vocal sounds ( Table 2). The direct comparisons between the musical and vocal sounds and between the musical and speech sounds yielded similar findings (Figure 2; Table 3). Greater activity related to music was observed bilaterally in posterior and anterolateral HG, the planum polare, and the most anterior parts of the right STG. We also observed effects related to music processing outside the temporal lobe, in inferior frontal gyrus (BA 45), the middle frontal gyrus (BA 6), and the left cerebellum (lobule IV). On the other hand, the reverse comparisons revealed stronger activity for vocal sounds as well as for speech in lateral HG, extending to lateral and anterior STG. For the vocal sounds, the extent of auditory activity was greater on the left (10312 voxels) than on the right (4952 voxels), however the ALE values were similar on the left (45.66 × 10 −3 ) and on the right (42.24 × 10 −3 ). As for the speech sounds, both the volume of activity and the corresponding ALE were greater on the left (11112 voxels, 61.39 × 10 −3 ) than the right (5736 voxels, 38.21 × 10 −3 ) hemisphere. We observed specialization for speech processing in auditory cortex. The comparison between intelligible speech and complex nonspeech sounds, including vocal sounds without intelligible content, is shown in Table 4 and Figure 3A. Speech was associated with greater activity in non-primary (BA 22) and associative (BA 39) auditory areas, lateral STG, bilateral anterior and middle STS, and the planum temporale (PT). These clusters were larger and had higher ALE values in the left hemisphere. We also observed stronger left prefrontal cortical activity (BA 8) for speech sounds. The reverse comparison yielded stronger activity related to complex non-speech sounds in the right PT (x = 68, y = −27, z = 8, 128 voxels; Figure 3A). The ALE maps associated with speech intelligibility had overlap with the vocal sound category maps ( Figure 3B). While large bilateral clusters were observed along the STG and STS for the vocal sounds, there was specific sensitivity to speech intelligibility in the left anterior STG.

STIMULUS CLASSIfICATION USINg AUDITORy COMpLExITy
Classification of sounds with respect to their spectral and temporal complexity revealed effects in the temporal lobe (Table 5; Figure 4). The "single, unchanging" stimulus class was associated with two clusters centered on medial HG (BA 41). The "single, regular change" stimulus class was associated with two large bilateral clusters of activity in medial and lateral HG, extending around HG into the anterolateral STG. On the left, we observed one additional peak of activity in posterior STG. For the "multiple, unchanging" stimulus class, temporal lobe activity was centered on medial HG and posterior STG. Effects for the "multiple, regular change" stimulus class were observed in HG, extending to the posterolateral STG. Finally, the "multiple, irregular change" stimulus class was associated with large bilateral effects in, and posterior to, HG. The complexity level maps revealed effects outside the temporal lobe, in frontal cortex areas BA 6, 9, 36, and 47 for the "multiple, unchanging" and "multiple, regular change" stimulus classes. We also observed effects in the cerebellum for the "single, regular change" and "multiple, irregular change" stimulus classes.
Effects related to stimulus spectral and temporal variations were identified by comparing, respectively, the multiple to the single stimulus classes (independent of changes over time; Figure 5B, GREEN) and the changing to the unchanging stimulus classes (independent of the number of frequency components; Figure 5B, BLUE). The coordinates of the effects related to increasing auditory complexity are reported in Table 5. Overlapping sensitivity to spectral and temporal effects was observed in the lateral portion of HG. Increasing numbers of frequency components were associated with greater effects in posterior and lateral non-primary auditory fields, specifically bilateral posterolateral STG and PT. Modulatory effects were also seen in inferior frontal gyrus (BA 45,47). In contrast, the effects related to temporal modulations compared to their absence were observed in HG, anterior STG, anterior STS, inferior frontal cortex (BA 46, 47), and right cerebellum (lobule IV).

SUMMARy Of fINDINgS
In a quantitative meta-analysis of 58 neuroimaging studies, we examined the functional specialization of human auditory cortex using two different strategies for classifying sounds. The first strategy employed typical categories, such as pure tones, noise, music, and vocal sounds. The second strategy categorized sounds according to their acoustical (spectral and temporal) complexity.
Activation Likelihood Estimation maps computed for each typical sound category included simple (pure tones) and complex (noise, voices, and music) sounds. This analysis gave results consistent with models describing hierarchical functional organization of the human auditory cortex, with simple sounds eliciting activity in the primary auditory cortex and complex sound processing engaging additional activity in non-primary fields. We observed an expected leftward hemispheric specialization for intelligible speech, while right-hemisphere specialization for music was less evident. Additionally, the comparison of intelligible speech to complex nonspeech stimuli yielded bilateral effects along the STG and STS, with higher sensitivity to speech intelligibility in the left anterior STG.
Examining an alternative classification based on stimulus variation along spectral and temporal dimensions, we observed a withinhemisphere functional segregation, with spectral effects strongest in posterior STG and temporal modulations strongest in anterior to elicit activity in core, belt, and parabelt areas. While belt region responses are thought to be sensitive to acoustic feature variations, the parabelt, and more anterior temporal regions, show greater sensitivity to complex sounds such as vocalizations (Rauschecker, 1998;Hackett, 2008;Rauschecker and Scott, 2009;Woods and Alain, 2009). Our quantitative meta-analysis using typical sound classes confirmed that hierarchical processing is a feature that can adequately describe human auditory cortical organization.
Using an ALE analysis of pure tone processing to investigate the correspondence between the core region and activity related to simple sound processing, we observed ALE extrema values bilaterally in medial HG, the putative location of primary auditory temporal STG. We suggest that acoustic complexity might represent a valid alternative classificatory scheme to describe a novel withinhemisphere dichotomy regarding the functional organization for auditory processing in temporal cortex.

HIERARCHICALLy AND HEMISpHERICALLy SpECIALIzED ARCHITECTURES fOR AUDITORy pROCESSINg
Originally elaborated on the basis of non-human primate studies, the hierarchical functional organization scheme in auditory cortex incorporates three levels of processing: core (primary area), belt and parabelt (non-primary areas). Simple sound processing is thought to solely recruit the core region whereas complex sounds are believed In contrast, we expected ALE analyses of the complex sound categories to show activity in all three levels of the processing hierarchy. We observed overlapping activity among the complex sound maps in medial HG (core) as well as stronger activity related to complex sound processing in regions surrounding medial HG, corresponding to the areas described as the auditory belt/parabelt in primates (Rauschecker, 1998;Kaas and Hackett, 2000;Rauschecker and Scott, 2009;Recanzone and Cohen, 2010) and humans (Rivier and Clarke, 1997;Wallace et al., 2002;Sweet et al., 2005). The fact that the complex sound maps showed effects in medial HG activity cortex. This finding is consistent with previous electrophysiological (Hackett et al., 2001), cytoarchitectural (Sweet et al., 2005), and functional imaging (Lauter et al., 1985;Bilecen et al., 1998;Lockwood et al., 1999;Wessinger et al., 2001) studies of the human auditory cortex that have localized the core region to medial HG. Our findings confirm the existence of functional specialization for simple sound processing in the human core homolog. Consequently, the statistical probability maps obtained here could serve to functionally define primary auditory cortex in a region of interest analysis of functional neuroimaging data.

FIguRe 3 | Activation Likelihood estimation maps showing clusters of activity related to (A)
intelligible speech > complex non-speech sounds (RED-YELLOW) and to intelligible speech < complex non-speech (BLUE-GREEN). Axial images are shown using the neurological convention with MNI z-coordinate labels. (B) Rendering of ALE maps related to vocal sound category (dark blue) and to speech intelligibility (pale blue). The maps are superimposed on anatomical templates in MNI space (p FDR < 0.01). For vocal sounds, we observed strong bilateral temporal lobe activity in anterior and posterior parts of dorsal STG and the STS, findings consistent with earlier studies (Binder et al., 1994;Belin, 2006). STG activity in response to vocal sounds has previously been interpreted as a neural correlate of the rapid and efficient processing of the complex frequency patterns and temporal variations characterizing speech. The human STG is thought to subserve complex auditory processing, such as vocalizations, as is the STG in non-human primates (Rauschecker et al., 1995). Belin and colleagues Fecteau et al., 2004) reported cortical responses to voices along the upper bank of the middle and anterior STS. The anterior STS is selectively responsive to human vocal sounds . Response specificity to vocal sounds and their rich identity and affective information content is of crucial importance, as it reflects a set of high-level auditory cognitive abilities that can be directly compared between human and non-human primates. The regions described as "Temporal Voice Areas" in humans  are thought to be functionally homologous to the temporal voice regions recently described in macaques (Petkov et al., 2008). Our meta-analysis using typical sound categories demonstrates that, in humans, simple sound processing elicits activity limited to the core area while complex sounds elicit effects in all three cortical processing levels.
supports the notion that primary auditory regions participate in the early stages of processing upon which further complex processing is built.
Outside primary auditory cortex, noise elicited activity in posterior temporal non-primary fields such as PT. The spatial pattern was similar to that observed in relation to broadband noise, stimuli that have been used to demonstrate the hierarchical organization of human auditory cortex (Wessinger et al., 2001). The PT is generally believed to be involved in complex sound analysis and participate in both language and other cognitive functions (Griffiths and Warren, 2002).
For music, in addition to primary auditory cortex activity, we observed activity in non-primary auditory fields along the STG bilaterally. This result is consistent with the idea that simple extraction and low-level ordering of pitch information involves processes within primary auditory fields, while higher-level processing for tone patterns and melodies involve non-primary auditory fields and association cortex . Moreover, non-primary regions in anterior and posterior STG are thought to process melody pitch intervals (Patterson et al., 2002;Tramo et al., 2002;Warren and Griffiths, 2003). Music also elicited strong inferior frontal cortex activity, a region thought to process musical syntax (Zatorre et al., 1994;Maess et al., 2001;Koelsch et al., 2002). of understanding the human-specific neural network underlying speech comprehension, and ultimately human language and communication.
In order to identify speech-specific processes, we directly compared intelligible speech to complex non-speech contrasts that included unintelligible spectro-temporally complex sounds. This comparison yielded stronger speech-related activity in lateral non-primary superior temporal regions, specifically in posterior STG, and anterior and middle STS. The effects were stronger and larger in the left hemisphere. Similar effects have been reported in independent studies examining specialization for processing speech sound that did not fulfill our inclusion criteria for this analysis (Scott et al., 2000;Davis and Johnsrude, 2003;Narain et al., 2003;Thierry et al., 2003;Liebenthal et al., 2005). Consistent with the present finding, these previous reports emphasized that speech-specific STS responses are more left-lateralized.
Beyond the auditory cortex, we observed activity in left inferior frontal and prefrontal cortex. These findings support an expanded hierarchical model of speech processing that originates in primary auditory areas and extends to non-auditory regions, mainly within frontal cortex, in a range of motor, premotor, and prefrontal regions (Davis and Johnsrude, 2007;Hickok and Poeppel, 2007;Rauschecker and Scott, 2009). In non-human primates, based on reports of high level of connection between the auditory and frontal cortex, it has been proposed that frontal regions responsive to auditory material should be considered as part of the auditory system Kaas et al., 1999;Romanski et al., 1999).
In addition to the hierarchical organization of auditory cortex, we expected hemispheric asymmetries for music and speech, and observed the expected left lateralization of auditory cortex responses to vocal sounds and intelligible speech. For vocal sounds, lateralization effects were observed only as a larger volume of auditory activity on the left while, for the speech sounds, the left auditory cortical responses were larger and stronger (higher ALE values) than the right-hemisphere responses. Greater lateralization effects for intelligible speech is in agreement with previous independent imaging studies, not included in this meta-analysis, reporting that intelligible speech sounds elicit strong activity in left STG and STS (e.g., Scott et al., 2000;Liebenthal et al., 2005;Obleser et al., 2007). Conversely, we did not see the expected right response lateralization related to music. Possibly, the small number of experiments included in the music category limited the power of this analysis and could have prevented us from observing the expected rightward auditory response. ALE maps derived from small samples are more sensitive to between-study cohort heterogeneity that could limit the detection of hemispheric effects. It is also possible that the right hemisphere is sensitive to particular features of musical stimuli such as fine pitch changes (Hyde et al., 2008) or to specific task demands like contextual pitch judgment (i.e., contextual pitch judgment Warrier and Zatorre, 2004) which were not present in our sample.

RESpONSE SpECIfICITy TO SpEECH INTELLIgIbILITy
Within the general category of vocal sounds, a human-specific category of intelligible speech can be further distinguished. Response specificity to speech intelligibility is an important part mainly observed response selectivity in posterolateral auditory fields. Our results therefore seem to be consistent with previous animal studies.
Second, cortical response specificity to spectral and temporal processing has also been studies in humans. Whereas some studies reported no clear functional segregation between responses to spectral and temporal cues (Hall et al., 2002) or observed neuronal populations tuned to specific combinations of spectrotemporal cues (Schönwiesner and Zatorre, 2009), other studies found the sorts of specific sensitivity to spectral vs. temporal features in human auditory cortex we observed in our meta-analysis. For instance, lateral HG and anterolateral PT activity have been reported in association with fine spectral structure analysis (Warren et al., 2005) and change detection of complex harmonic tones involved the posterior STG and lateral PT (Schönwiesner et al., 2007). Additionally, recent studies examining effective connectivity effects among auditory regions reported that spectral envelope analysis follows a serial pathway from HG to PT and then to the STS Kumar et al., 2007). Conversely for temporal complexity effects, a stream of processing from primary auditory cortex to anterior STG has been observed for auditory pattern analysis such as dynamic pitch variation (Griffiths et al., 1998). Similarly, significant effects of temporal modulation have been reported in anterior non-primary auditory fields (Hall et al., 2000). Some studies therefore report patterns of activity consistent with the current findings, albeit separately for spectral and temporal features.
A more frequently observed feature of spectral vs. temporal processing is between-hemisphere functional specialization. Most studies observed slight but significant lateralization effects with a left-lateralized response to temporal information and rightlateralized activity to spectral information (Zatorre and Belin, 2001;Schönwiesner et al., 2005;Jamison et al., 2006;Obleser et al., 2008). In the current study, lateralization effects were not seen with regard to complexity. However, at higher processing levels, leftward lateralization for speech was observed. Others studies failing to demonstrate the expected lateralization proposed that early stages of processing involve bilateral auditory cortex and that higher cognitive functions, such as speech processing, also rely on these regions but involve more extensive regions in the dominant hemisphere (Langers et al., 2003). Alternatively, Tervaniemi and Hugdahl (2003) reviewed studies showing that response lateralization within the auditory cortex is dependent on sound structure as well as the acoustic background they are presented in. For instance, reduced or absent hemispheric specialization for speech sounds has been reported when the amount of formant structure is not sufficient to establish phoneme categorization (Rinne et al., 1999) or when sounds are presented in noise (Shtyrov et al., 1998). Stimulus heterogeneity among the different experiments included in our meta-analysis could explain why we did not observe asymmetrical hemispheric effects.
To summarize, our meta-analysis demonstrates a clear withinhemisphere functional segregation related to spectral and temporal processing in human auditory cortex, consistent with the known organization of non-human primate auditory system. That such clear spectral vs. temporal complexity gradients are observed (Figure 5), while very few of the included studies have explicitly addressed this issue, illustrates the power of the meta-analysis fUNCTIONAL SpECIALIzATION Of THE AUDITORy CORTEx RESpONSE:

ACOUSTIC COMpLExITy EffECTS
As an alternative to the classical division of auditory stimuli into typical categories like pure tones, noise, voices, and music, we explored how acoustic variations along the temporal and spectral dimensions were represented at the cortical level. This approach for defining auditory material is an efficient and comprehensive characterization of sounds that can be considered as a complement to the more typically studied categorical effects. Possibly, certain aspects of human auditory processes might be better characterized in terms of their capacity to analyze acoustic features rather than having differential sensitivity to typical sound categories. In a meta-analysis Rivier and Clarke (1997) found no clear functional specialization in non-primary auditory fields for a range of complex sound categories, showing that processing sounds of different categories such as noise, words, and music, elicited activity in multiple non-primary fields around HG with no emergence of a specific organizational pattern. Similarly, Griffiths and Warren (2002) reported that activity within the PT, an auditory association region, is not spatially organized according to sound categories such as music, speech or, environmental sounds.
By classifying sounds according to their variations in time and frequency, we isolated different levels of auditory complexity, suggesting a within-hemisphere functional segregation with anterior STG and STS more sensitive to changes in the temporal domain and posterior regions (PT and posterolateral STG) more sensitive to changes in along the spectral dimension. Interestingly, a partial overlap was observed between regions sensitive to temporal and spectral changes in lateral HG, suggesting great sensitivity to variations in acoustic properties within this region, consistent with a recent report of strongest sensitivity to stimulus acoustic features within HG (Okada et al., 2010).
Our observation of differential sensitivity to temporal and spectral features can be interpreted in the light of previous findings. First, in the animal literature, a within-hemisphere model of spectral and temporal processing in the auditory cortex has been proposed (Bendor and Wang, 2008). This scheme suggests two streams of processing originating from primary auditory cortex; an anterior pathway sensitive to temporal changes and a lateral pathway responsive to spectral changes. More precise temporal coding is seen as one progresses from primary to anterior auditory regions in primates (Bendor and Wang, 2007) and greater sensitivity to temporal modulations in anterior non-primary auditory fields is also observed in cats (Tian and Rauschecker, 1994). Possibly, a longer integration window in anterior auditory fields could underlie complex temporal processing (Bendor and Wang, 2008). As regards spectral processing, increasing sensitivity to broadband spectrum noise compared to single tones has been observed in lateral and posterior auditory fields in non-human primates (Rauschecker and Tian, 2004;Petkov et al., 2006). Furthermore, given that the neurons within these regions show strong tuning to bandwidth and frequency, some have suggested their involvement in the early stages of spectral analysis of complex sounds (Rauschecker and Tian, 2004). In our study, sensitivity to temporal changes was observed in anterior temporal regions, while, in response to changes along the spectral dimension, we approach for human neuroimaging studies. Based on the observed regional functional segregation, we argue that acoustic complexity could well represent a relevant stimulus dimension upon which to identify response segregation within the auditory system. Complexity and categorical effects could therefore be considered as two complementary approaches to more fully characterizing the underlying nature of auditory regional functional specialization.

ACKNOWLEDgMENTS
We would like to thank Dr. Laurent Mottron and Dr. Valter Ciocca for providing comments on the manuscript and suggestions regarding the stimulus categorizations. This work was supported by Canadian Institutes for Health Research (grant MOP-84243) as well as a doctoral award from Natural Sciences and Engineering Research Council of Canada to Fabienne Samson.

Samson et al.
Meta-analysis of auditory activation studies