The speech neuroprosthesis.

Loss of speech after paralysis is devastating, but circumventing motor-pathway injury by directly decoding speech from intact cortical activity has the potential to restore natural communication and self-expression. Recent discoveries have defined how key features of speech production are facilitated by the coordinated activity of vocal-tract articulatory and motor-planning cortical representations. In this Review, we highlight such progress and how it has led to successful speech decoding, first in individuals implanted with intracranial electrodes for clinical epilepsy monitoring and subsequently in individuals with paralysis as part of early feasibility clinical trials to restore speech. We discuss high-spatiotemporal-resolution neural interfaces and the adaptation of state-of-the-art speech computational algorithms that have driven rapid and substantial progress in decoding neural activity into text, audible speech, and facial movements. Although restoring natural speech is a long-term goal, speech neuroprostheses already have performance levels that surpass communication rates offered by current assistive-communication technology. Given this accelerated rate of progress in the field, we propose key evaluation metrics for speed and accuracy, among others, to help standardize across studies. We finish by highlighting several directions to more fully explore the multidimensional feature space of speech and language, which will continue to accelerate progress towards a clinically viable speech neuroprosthesis.

A bilingual speech neuroprosthesis driven by cortical articulatory representations shared between languages.

Advancements in decoding speech from brain activity have focused on decoding a single language. Hence, the extent to which bilingual speech production relies on unique or shared cortical activity across languages has remained unclear. Here, we leveraged electrocorticography, along with deep-learning and statistical natural-language models of English and Spanish, to record and decode activity from speech-motor cortex of a Spanish-English bilingual with vocal-tract and limb paralysis into sentences in either language. This was achieved without requiring the participant to manually specify the target language. Decoding models relied on shared vocal-tract articulatory representations across languages, which allowed us to build a syllable classifier that generalized across a shared set of English and Spanish syllables. Transfer learning expedited training of the bilingual decoder by enabling neural data recorded in one language to improve decoding in the other language. Overall, our findings suggest shared cortical articulatory representations that persist after paralysis and enable the decoding of multiple languages without the need to train separate language-specific decoders.

Parallel Encoding of Speech in Human Frontal and Temporal Lobes.

Models of speech perception are centered around a hierarchy in which auditory representations in the thalamus propagate to primary auditory cortex, then to the lateral temporal cortex, and finally through dorsal and ventral pathways to sites in the frontal lobe. However, evidence for short latency speech responses and low-level spectrotemporal representations in frontal cortex raises the question of whether speech-evoked activity in frontal cortex strictly reflects downstream processing from lateral temporal cortex or whether there are direct parallel pathways from the thalamus or primary auditory cortex to the frontal lobe that supplement the traditional hierarchical architecture. Here, we used high-density direct cortical recordings, high-resolution diffusion tractography, and hemodynamic functional connectivity to evaluate for evidence of direct parallel inputs to frontal cortex from low-level areas. We found that neural populations in the frontal lobe show speech-evoked responses that are synchronous or occur earlier than responses in the lateral temporal cortex. These short latency frontal lobe neural populations encode spectrotemporal speech content indistinguishable from spectrotemporal encoding patterns observed in the lateral temporal lobe, suggesting parallel auditory speech representations reaching temporal and frontal cortex simultaneously. This is further supported by white matter tractography and functional connectivity patterns that connect the auditory nucleus of the thalamus (medial geniculate body) and the primary auditory cortex to the frontal lobe. Together, these results support the existence of a robust pathway of parallel inputs from low-level auditory areas to frontal lobe targets and illustrate long-range parallel architecture that works alongside the classical hierarchical speech network model.

Development and validation of a nonverbal consensus-based semantic memory paradigm in patients with epilepsy.

OBJECTIVE: Brain areas implicated in semantic memory can be damaged in patients with epilepsy (PWE). However, it is challenging to delineate semantic processing deficits from acoustic, linguistic, and other verbal aspects in current neuropsychological assessments. We developed a new Visual-based Semantic Association Task (ViSAT) to evaluate nonverbal semantic processing in PWE. METHOD: The ViSAT was adapted from similar predecessors (Pyramids & Palm Trees test, PPT; Camels & Cactus Test, CCT) comprised of 100 unique trials using real-life color pictures that avoid demographic, cultural, and other potential confounds. We obtained performance data from 23 PWE participants and 24 control participants (Control), along with crowdsourced normative data from 54 Amazon Mechanical Turk (Mturk) workers. RESULTS: ViSAT reached a consensus >90% in 91.3% of trials compared to 83.6% in PPT and 82.9% in CCT. A deep learning model demonstrated that visual features of the stimulus images (color, shape; i.e., non-semantic) did not influence top answer choices (p = 0.577). The PWE group had lower accuracy than the Control group (p = 0.019). PWE had longer response times than the Control group in general and this was augmented for the semantic processing (trial answer) stage (both p < 0.001). CONCLUSIONS: This study demonstrated performance impairments in PWE that may reflect dysfunction of nonverbal semantic memory circuits, such as seizure onset zones overlapping with key semantic regions (e.g., anterior temporal lobe). The ViSAT paradigm avoids confounds, is repeatable/longitudinal, captures behavioral data, and is open-source, thus we propose it as a strong alternative for clinical and research assessment of nonverbal semantic memory.

Quantitative Assessment of Preanalytic Variables on Clinical Evaluation of PI3/AKT/mTOR Signaling Activity in Diffuse Glioma.

Biomarker-driven therapeutic clinical trials require the implementation of standardized, evidence-based practices for sample collection. In diffuse glioma, phosphatidylinositol 3 (PI3)-kinase/AKT/mTOR (PI3/AKT/mTOR) signaling is an attractive therapeutic target for which window-of-opportunity clinical trials could facilitate the identification of promising new agents. Yet, the relevant preanalytic variables and optimal tumor sampling methods necessary to measure pathway activity are unknown. To address this, we used a murine model for isocitrate dehydrogenase (IDH)-wildtype glioblastoma (GBM) and human tumor tissue, including IDH-wildtype GBM and IDH-mutant diffuse glioma. First, we determined the impact of delayed time-to-formalin fixation, or cold ischemia time (CIT), on the quantitative assessment of cellular expression of 6 phosphoproteins that are readouts of PI3K/AK/mTOR activity (phosphorylated-proline-rich Akt substrate of 40 kDa (p-PRAS40, T246), -mechanistic target of rapamycin (p-mTOR; S2448); -AKT (p-AKT, S473); -ribosomal protein S6 (p-RPS6, S240/244 and S235/236), and -eukaryotic initiation factor 4E-binding protein 1 (p-4EBP1, T37/46). With CITs ≥ 2 hours, typical of routine clinical handling, all had reduced or altered expression with p-RPS6 (S240/244) exhibiting relatively greater stability. A similar pattern was observed using patient tumor samples from the operating room with p-4EBP1 more sensitive to delayed fixation than p-RPS6 (S240/244). Many clinical trials utilize unstained slides for biomarker evaluation. Thus, we evaluated the impact of slide storage conditions on the detection of p-RPS6 (S240/244), p-4EBP1, and p-AKT. After 5 months, storage at -80°C was required to preserve the expression of p-4EBP1 and p-AKT, whereas p-RPS6 (240/244) expression was not stable regardless of storage temperature. Biomarker heterogeneity impacts optimal tumor sampling. Quantification of p-RPS6 (240/244) expression in multiple regionally distinct human tumor samples from 8 patients revealed significant intratumoral heterogeneity. Thus, the accurate assessment of PI3K/AKT/mTOR signaling in diffuse glioma must overcome intratumoral heterogeneity and multiple preanalytic factors, including time-to-formalin fixation, slide storage conditions, and phosphoprotein of interest.

Human Anterior Insular Cortex Encodes Multiple Electrophysiological Representations of Anxiety-Related Behaviors.

Anxiety is a common symptom across psychiatric disorders, but the neurophysiological underpinnings of these symptoms remain unclear. This knowledge gap has prevented the development of circuit-based treatments that can target the neural substrates underlying anxiety. Here, we conducted an electrophysiological mapping study to identify neurophysiological activity associated with self-reported state anxiety in 17 subjects implanted with intracranial electrodes for seizure localization. Participants had baseline anxiety traits ranging from minimal to severe. Subjects volunteered to participate in an anxiety induction task in which they were temporarily exposed to the threat of unpredictable shock during intracranial recordings. We found that anterior insular beta oscillatory activity was selectively elevated during epochs when unpredictable aversive stimuli were being delivered, and this enhancement in insular beta was correlated with increases in self-reported anxiety. Beta oscillatory activity within the frontoinsular region was also evoked selectively by cues-predictive of threat, but not safety cues. Anterior insular gamma responses were less selective than gamma, strongly evoked by aversive stimuli and had weaker responses to salient threat and safety cues. On longer timescales, this gamma signal also correlated with increased skin conductance, a measure of autonomic state. Lastly, we found that direct electrical stimulation of the anterior insular cortex in a subset of subjects elicited self-reported increases in anxiety that were accompanied by enhanced frontoinsular beta oscillations. Together, these findings suggest that electrophysiologic representations of anxiety- related states and behaviors exist within anterior insular cortex. The findings also suggest the potential of reducing anterior insular beta activity as a therapeutic target for refractory anxiety-spectrum disorders.

Encoding of melody in the human auditory cortex.

Melody is a core component of music in which discrete pitches are serially arranged to convey emotion and meaning. Perception varies along several pitch-based dimensions: (i) the absolute pitch of notes, (ii) the difference in pitch between successive notes, and (iii) the statistical expectation of each note given prior context. How the brain represents these dimensions and whether their encoding is specialized for music remains unknown. We recorded high-density neurophysiological activity directly from the human auditory cortex while participants listened to Western musical phrases. Pitch, pitch-change, and expectation were selectively encoded at different cortical sites, indicating a spatial map for representing distinct melodic dimensions. The same participants listened to spoken English, and we compared responses to music and speech. Cortical sites selective for music encoded expectation, while sites that encoded pitch and pitch-change in music used the same neural code to represent equivalent properties of speech. Findings reveal how the perception of melody recruits both music-specific and general-purpose sound representations.

Microvascular decompression of a vertebral artery loop causing cervical radiculopathy: illustrative case.

BACKGROUND: Vertebral artery loops are a rare cause of cervical radiculopathy. Surgical options for nerve root decompression include an anterior or posterior approach, with or without additional microvascular decompression. OBSERVATIONS: The authors describe a case of a 49-year-old man with a long-standing history of left-sided neck pain and migraines, who was found to have a vertebral artery loop in the left C3-4 neural foramen compressing the left C4 nerve root. The patient underwent a posterior cervical decompression with instrumented fusion and macrovascular decompression of the left C4 nerve root via Teflon felt insertion. In a literature review, we identified 20 similar cases that had also been managed surgically. LESSONS: Although the anterior approach is more frequently described in the literature, a posterior approach for nerve compression by a vertebral artery loop is also a safe and effective treatment. The authors report the third case of this surgical approach with a good outcome.

Amygdala-hippocampus somatostatin interneuron beta-synchrony underlies a cross-species biomarker of emotional state.

Emotional responses arise from limbic circuits including the hippocampus and amygdala. In the human brain, beta-frequency communication between these structures correlates with self-reported mood and anxiety. However, both the mechanism and significance of this biomarker as a readout vs. driver of emotional state remain unknown. Here, we show that beta-frequency communication between ventral hippocampus and basolateral amygdala also predicts anxiety-related behavior in mice, both on long timescales (∼30 min) and immediately preceding behavioral choices. Genetically encoded voltage indicators reveal that this biomarker reflects synchronization between somatostatin interneurons across both structures. Indeed, synchrony between these neurons dynamically predicts approach-avoidance decisions, and optogenetically shifting the phase of synchronization by just 25 ms is sufficient to bidirectionally modulate anxiety-related behaviors. Thus, back-translation establishes a human biomarker as a causal determinant (not just predictor) of emotional state, revealing a novel mechanism whereby interregional synchronization that is frequency, phase, and cell type specific controls emotional processing.

Intraoperative cortical stimulation mapping with laryngeal electromyography for the localization of human laryngeal motor cortex.

OBJECTIVE: The objectives of this study were to describe the authors' clinical methodology and outcomes for mapping the laryngeal motor cortex (LMC) and define localization of the LMC in a cohort of neurosurgical patients undergoing intraoperative brain mapping. Because of mapping variability across patients, the authors aimed to define the probabilistic distribution of cortical sites that evoke laryngeal movement, as well as adjacent cortical somatotopic representations for the face (mouth), tongue, and hand. METHODS: Thirty-six patients underwent left (n = 18) or right (n = 18) craniotomy with asleep motor mapping. For each patient, electromyography (EMG) electrodes were placed in the face, tongue, and hand; a nerve integrity monitor (NIM) endotracheal tube with surface electrodes detected EMG activity from the bilateral vocal folds. After dense cortical stimulation was delivered throughout the sensorimotor cortex, motor responses were then mapped onto a three-dimensional reconstruction of the patient's cortical surfaces for location characterization of the evoked responses. Finally, stimulation sites were transformed into a two-dimensional coordinate system for probabilistic mapping of the stimulation site relative to the central sulcus and sylvian fissure. RESULTS: The authors found that the LMC was predominantly localized to a mid precentral gyrus region, dorsal to face representation and surrounding a transverse sulcus ventral to the hand knob. In 14 of 36 patients, the authors identified additional laryngeal responses located ventral to all orofacial representations, providing evidence for dual LMC representations. CONCLUSIONS: The authors determined the probabilistic distribution of the LMC. Cortical stimulation mapping with an NIM endotracheal tube is an easy and effective method for mapping the LMC and is simply integrated into the current neuromonitoring methods for brain mapping.

Optimal trade-off control in machine learning-based library design, with application to adeno-associated virus (AAV) for gene therapy.

Adeno-associated viruses (AAVs) hold tremendous promise as delivery vectors for gene therapies. AAVs have been successfully engineered-for instance, for more efficient and/or cell-specific delivery to numerous tissues-by creating large, diverse starting libraries and selecting for desired properties. However, these starting libraries often contain a high proportion of variants unable to assemble or package their genomes, a prerequisite for any gene delivery goal. Here, we present and showcase a machine learning (ML) method for designing AAV peptide insertion libraries that achieve fivefold higher packaging fitness than the standard NNK library with negligible reduction in diversity. To demonstrate our ML-designed library's utility for downstream engineering goals, we show that it yields approximately 10-fold more successful variants than the NNK library after selection for infection of human brain tissue, leading to a promising glial-specific variant. Moreover, our design approach can be applied to other types of libraries for AAV and beyond.

Anatomical registration of intracranial electrodes. Robust model-based localization and deformable smooth brain-shift compensation methods.

BACKGROUND: Intracranial electrodes are typically localized from post-implantation CT artifacts. Automatic algorithms localizing low signal-to-noise ratio artifacts and high-density electrode arrays are missing. Additionally, implantation of grids/strips introduces brain deformations, resulting in registration errors when fusing post-implantation CT and pre-implantation MR images. Brain-shift compensation methods project electrode coordinates to cortex, but either fail to produce smooth solutions or do not account for brain deformations. NEW METHODS: We first introduce GridFit, a model-based fitting approach that simultaneously localizes all electrodes' CT artifacts in grids, strips, or depth arrays. Second, we present CEPA, a brain-shift compensation algorithm combining orthogonal-based projections, spring-mesh models, and spatial regularization constraints. RESULTS: We tested GridFit on ∼6000 simulated scenarios. The localization of CT artifacts showed robust performance under difficult scenarios, such as noise, overlaps, and high-density implants (<1 mm errors). Validation with data from 20 challenging patients showed 99% accurate localization of the electrodes (3160/3192). We tested CEPA brain-shift compensation with data from 15 patients. Projections accounted for simple mechanical deformation principles with < 0.4 mm errors. The inter-electrode distances smoothly changed across neighbor electrodes, while changes in inter-electrode distances linearly increased with projection distance. COMPARISON WITH EXISTING METHODS: GridFit succeeded in difficult scenarios that challenged available methods and outperformed visual localization by preserving the inter-electrode distance. CEPA registration errors were smaller than those obtained for well-established alternatives. Additionally, modeling resting-state high-frequency activity in five patients further supported CEPA. CONCLUSION: GridFit and CEPA are versatile tools for registering intracranial electrode coordinates, providing highly accurate results even in the most challenging implantation scenarios. The methods are implemented in the iElectrodes open-source toolbox.

Large-scale single-neuron speech sound encoding across the depth of human cortex.

Understanding the neural basis of speech perception requires that we study the human brain both at the scale of the fundamental computational unit of neurons and in their organization across the depth of cortex. Here we used high-density Neuropixels arrays1-3 to record from 685 neurons across cortical layers at nine sites in a high-level auditory region that is critical for speech, the superior temporal gyrus4,5, while participants listened to spoken sentences. Single neurons encoded a wide range of speech sound cues, including features of consonants and vowels, relative vocal pitch, onsets, amplitude envelope and sequence statistics. Neurons at each cross-laminar recording exhibited dominant tuning to a primary speech feature while also containing a substantial proportion of neurons that encoded other features contributing to heterogeneous selectivity. Spatially, neurons at similar cortical depths tended to encode similar speech features. Activity across all cortical layers was predictive of high-frequency field potentials (electrocorticography), providing a neuronal origin for macroelectrode recordings from the cortical surface. Together, these results establish single-neuron tuning across the cortical laminae as an important dimension of speech encoding in human superior temporal gyrus.

Genomic alterations associated with postoperative nodular leptomeningeal disease after resection of brain metastases.

OBJECTIVE: The relationship between brain metastasis resection and risk of nodular leptomeningeal disease (nLMD) is unclear. This study examined genomic alterations found in brain metastases with the aim of identifying alterations associated with postoperative nLMD in the context of clinical and treatment factors. METHODS: A retrospective, single-center study was conducted on patients who underwent resection of brain metastases between 2014 and 2022 and had clinical and genomic data available. Postoperative nLMD was the primary endpoint of interest. Targeted next-generation sequencing of > 500 oncogenes was performed in brain metastases. Cox proportional hazards analyses were performed to identify clinical features and genomic alterations associated with nLMD. RESULTS: The cohort comprised 101 patients with tumors originating from multiple cancer types. There were 15 patients with nLMD (14.9% of the cohort) with a median time from surgery to nLMD diagnosis of 8.2 months. Two supervised machine learning algorithms consistently identified CDKN2A/B codeletion and ERBB2 amplification as the top predictors associated with postoperative nLMD across all cancer types. In a multivariate Cox proportional hazards analysis including clinical factors and genomic alterations observed in the cohort, tumor volume (× 10 cm3; HR 1.2, 95% CI 1.01-1.5; p = 0.04), CDKN2A/B codeletion (HR 5.3, 95% CI 1.7-16.9; p = 0.004), and ERBB2 amplification (HR 3.9, 95% CI 1.1-14.4; p = 0.04) were associated with a decreased time to postoperative nLMD. CONCLUSIONS: In addition to increased resected tumor volume, ERBB2 amplification and CDKN2A/B deletion were independently associated with an increased risk of postoperative nLMD across multiple cancer types. Additional work is needed to determine if targeted therapy decreases this risk in the postoperative setting.

Closed-loop neurostimulation for the treatment of psychiatric disorders.

Despite increasing prevalence and huge personal and societal burden, psychiatric diseases still lack treatments which can control symptoms for a large fraction of patients. Increasing insight into the neurobiology underlying these diseases has demonstrated wide-ranging aberrant activity and functioning in multiple brain circuits and networks. Together with varied presentation and symptoms, this makes one-size-fits-all treatment a challenge. There has been a resurgence of interest in the use of neurostimulation as a treatment for psychiatric diseases. Initial studies using continuous open-loop stimulation, in which clinicians adjusted stimulation parameters during patient visits, showed promise but also mixed results. Given the periodic nature and fluctuations of symptoms often observed in psychiatric illnesses, the use of device-driven closed-loop stimulation may provide more effective therapy. The use of a biomarker, which is correlated with specific symptoms, to deliver stimulation only during symptomatic periods allows for the personalized therapy needed for such heterogeneous disorders. Here, we provide the reader with background motivating the use of closed-loop neurostimulation for the treatment of psychiatric disorders. We review foundational studies of open- and closed-loop neurostimulation for neuropsychiatric indications, focusing on deep brain stimulation, and discuss key considerations when designing and implementing closed-loop neurostimulation.

"De novo replication repair deficient glioblastoma, IDH-wildtype" is a distinct glioblastoma subtype in adults that may benefit from immune checkpoint blockade.

Glioblastoma is a clinically and molecularly heterogeneous disease, and new predictive biomarkers are needed to identify those patients most likely to respond to specific treatments. Through prospective genomic profiling of 459 consecutive primary treatment-naïve IDH-wildtype glioblastomas in adults, we identified a unique subgroup (2%, 9/459) defined by somatic hypermutation and DNA replication repair deficiency due to biallelic inactivation of a canonical mismatch repair gene. The deleterious mutations in mismatch repair genes were often present in the germline in the heterozygous state with somatic inactivation of the remaining allele, consistent with glioblastomas arising due to underlying Lynch syndrome. A subset of tumors had accompanying proofreading domain mutations in the DNA polymerase POLE and resultant "ultrahypermutation". The median age at diagnosis was 50 years (range 27-78), compared with 63 years for the other 450 patients with conventional glioblastoma (p < 0.01). All tumors had histologic features of the giant cell variant of glioblastoma. They lacked EGFR amplification, lacked combined trisomy of chromosome 7 plus monosomy of chromosome 10, and only rarely had TERT promoter mutation or CDKN2A homozygous deletion, which are hallmarks of conventional IDH-wildtype glioblastoma. Instead, they harbored frequent inactivating mutations in TP53, NF1, PTEN, ATRX, and SETD2 and recurrent activating mutations in PDGFRA. DNA methylation profiling revealed they did not align with known reference adult glioblastoma methylation classes, but instead had unique globally hypomethylated epigenomes and mostly classified as "Diffuse pediatric-type high grade glioma, RTK1 subtype, subclass A". Five patients were treated with immune checkpoint blockade, four of whom survived greater than 3 years. The median overall survival was 36.8 months, compared to 15.5 months for the other 450 patients (p < 0.001). We conclude that "De novo replication repair deficient glioblastoma, IDH-wildtype" represents a biologically distinct subtype in the adult population that may benefit from prospective identification and treatment with immune checkpoint blockade.

DREDge: robust motion correction for high-density extracellular recordings across species.

High-density microelectrode arrays (MEAs) have opened new possibilities for systems neuroscience in human and non-human animals, but brain tissue motion relative to the array poses a challenge for downstream analyses, particularly in human recordings. We introduce DREDge (Decentralized Registration of Electrophysiology Data), a robust algorithm which is well suited for the registration of noisy, nonstationary extracellular electrophysiology recordings. In addition to estimating motion from spikes in the action potential (AP) frequency band, DREDge enables automated tracking of motion at high temporal resolution in the local field potential (LFP) frequency band. In human intraoperative recordings, which often feature fast (period <1s) motion, DREDge correction in the LFP band enabled reliable recovery of evoked potentials, and significantly reduced single-unit spike shape variability and spike sorting error. Applying DREDge to recordings made during deep probe insertions in nonhuman primates demonstrated the possibility of tracking probe motion of centimeters across several brain regions while simultaneously mapping single unit electrophysiological features. DREDge reliably delivered improved motion correction in acute mouse recordings, especially in those made with an recent ultra-high density probe. We also implemented a procedure for applying DREDge to recordings made across tens of days in chronic implantations in mice, reliably yielding stable motion tracking despite changes in neural activity across experimental sessions. Together, these advances enable automated, scalable registration of electrophysiological data across multiple species, probe types, and drift cases, providing a stable foundation for downstream scientific analyses of these rich datasets.

Recommendations for promoting user agency in the design of speech neuroprostheses.

Brain-computer interfaces (BCI) that directly decode speech from brain activity aim to restore communication in people with paralysis who cannot speak. Despite recent advances, neural inference of speech remains imperfect, limiting the ability for speech BCIs to enable experiences such as fluent conversation that promote agency - that is, the ability for users to author and transmit messages enacting their intentions. Here, we make recommendations for promoting agency based on existing and emerging strategies in neural engineering. The focus is on achieving fast, accurate, and reliable performance while ensuring volitional control over when a decoder is engaged, what exactly is decoded, and how messages are expressed. Additionally, alongside neuroscientific progress within controlled experimental settings, we argue that a parallel line of research must consider how to translate experimental successes into real-world environments. While such research will ultimately require input from prospective users, here we identify and describe design choices inspired by human-factors work conducted in existing fields of assistive technology, which address practical issues likely to emerge in future real-world speech BCI applications.

Neural control of lexical tone production in human laryngeal motor cortex.

In tonal languages, which are spoken by nearly one-third of the world's population, speakers precisely control the tension of vocal folds in the larynx to modulate pitch in order to distinguish words with completely different meanings. The specific pitch trajectories for a given tonal language are called lexical tones. Here, we used high-density direct cortical recordings to determine the neural basis of lexical tone production in native Mandarin-speaking participants. We found that instead of a tone category-selective coding, local populations in the bilateral laryngeal motor cortex (LMC) encode articulatory kinematic information to generate the pitch dynamics of lexical tones. Using a computational model of tone production, we discovered two distinct patterns of population activity in LMC commanding pitch rising and lowering. Finally, we showed that direct electrocortical stimulation of different local populations in LMC evoked pitch rising and lowering during tone production, respectively. Together, these results reveal the neural basis of vocal pitch control of lexical tones in tonal languages.

Dissecting neural computations in the human auditory pathway using deep neural networks for speech.

The human auditory system extracts rich linguistic abstractions from speech signals. Traditional approaches to understanding this complex process have used linear feature-encoding models, with limited success. Artificial neural networks excel in speech recognition tasks and offer promising computational models of speech processing. We used speech representations in state-of-the-art deep neural network (DNN) models to investigate neural coding from the auditory nerve to the speech cortex. Representations in hierarchical layers of the DNN correlated well with the neural activity throughout the ascending auditory system. Unsupervised speech models performed at least as well as other purely supervised or fine-tuned models. Deeper DNN layers were better correlated with the neural activity in the higher-order auditory cortex, with computations aligned with phonemic and syllabic structures in speech. Accordingly, DNN models trained on either English or Mandarin predicted cortical responses in native speakers of each language. These results reveal convergence between DNN model representations and the biological auditory pathway, offering new approaches for modeling neural coding in the auditory cortex.