Infants’ speech perception adapts to the phonemic categoriesof their native language, a process assumed to be driven bythe distributional properties of speech. This study investigateswhether deep neural networks (DNNs), the current state-of-the-art in distributional feature learning, are capable oflearning phoneme-like representations of speech in anunsupervised manner. We trained DNNs with unlabeled andlabeled speech and analyzed the activations of each layer withrespect to the phones in the input segments. The analysesreveal that the emergence of phonemic invariance in DNNs isdependent on the availability of phonemic labeling of theinput during the training. No increased phonemic selectivityof the hidden layers was observed in the purely unsupervisednetworks despite successful learning of low-dimensionalrepresentations for speech. This suggests that additionallearning constraints or more sophisticated models are neededto account for the emergence of phone-like categories indistributional learning operating on natural speech.

During speech perception, linguistic elements such as consonants and vowels are extracted from a complex acoustic speech signal. The superior temporal gyrus (STG) participates in high-order auditory processing of speech, but how it encodes phonetic information is poorly understood. We used high-density direct cortical surface recordings in humans while they listened to natural, continuous speech to reveal the STG representation of the entire English phonetic inventory. At single electrodes, we found response selectivity to distinct phonetic features. Encoding of acoustic properties was mediated by a distributed population response. Phonetic features could be directly related to tuning for spectrotemporal acoustic cues, some of which were encoded in a nonlinear fashion or by integration of multiple cues. These findings demonstrate the acoustic-phonetic representation of speech in human STG.

Objective

The superior temporal gyrus (STG) and neighboring brain regions play a key role in human language processing. Previous studies have attempted to reconstruct speech information from brain activity in the STG, but few of them incorporate the probabilistic framework and engineering methodology used in modern speech recognition systems. In this work, we describe the initial efforts toward the design of a neural speech recognition (NSR) system that performs continuous phoneme recognition on English stimuli with arbitrary vocabulary sizes using the high gamma band power of local field potentials in the STG and neighboring cortical areas obtained via electrocorticography.

Approach

The system implements a Viterbi decoder that incorporates phoneme likelihood estimates from a linear discriminant analysis model and transition probabilities from an n-gram phonemic language model. Grid searches were used in an attempt to determine optimal parameterizations of the feature vectors and Viterbi decoder.

Main results

The performance of the system was significantly improved by using spatiotemporal representations of the neural activity (as opposed to purely spatial representations) and by including language modeling and Viterbi decoding in the NSR system.

Significance

These results emphasize the importance of modeling the temporal dynamics of neural responses when analyzing their variations with respect to varying stimuli and demonstrate that speech recognition techniques can be successfully leveraged when decoding speech from neural signals. Guided by the results detailed in this work, further development of the NSR system could have applications in the fields of automatic speech recognition and neural prosthetics.

Speaking is one of the most complex actions that we perform, but nearly all of us learn to do it effortlessly. Production of fluent speech requires the precise, coordinated movement of multiple articulators (for example, the lips, jaw, tongue and larynx) over rapid time scales. Here we used high-resolution, multi-electrode cortical recordings during the production of consonant-vowel syllables to determine the organization of speech sensorimotor cortex in humans. We found speech-articulator representations that are arranged somatotopically on ventral pre- and post-central gyri, and that partially overlap at individual electrodes. These representations were coordinated temporally as sequences during syllable production. Spatial patterns of cortical activity showed an emergent, population-level representation, which was organized by phonetic features. Over tens of milliseconds, the spatial patterns transitioned between distinct representations for different consonants and vowels. These results reveal the dynamic organization of speech sensorimotor cortex during the generation of multi-articulator movements that underlies our ability to speak.

Auditory speech comprehension is the result of neural computations that occur in a broad network that includes the temporal lobe auditory cortex and the left inferior frontal cortex. It remains unclear how representations in this network differentially contribute to speech comprehension. Here, we recorded high-density direct cortical activity during a sine-wave speech (SWS) listening task to examine detailed neural speech representations when the exact same acoustic input is comprehended versus not comprehended. Listeners heard SWS sentences (pre-exposure), followed by clear versions of the same sentences, which revealed the content of the sounds (exposure), and then the same SWS sentences again (post-exposure). Across all three task phases, high-gamma neural activity in the superior temporal gyrus was similar, distinguishing different words based on bottom-up acoustic features. In contrast, frontal regions showed a more pronounced and sudden increase in activity only when the input was comprehended, which corresponded with stronger representational separability among spatiotemporal activity patterns evoked by different words. We observed this effect only in participants who were not able to comprehend the stimuli during the pre-exposure phase, indicating a relationship between frontal high-gamma activity and speech understanding. Together, these results demonstrate that both frontal and temporal cortical networks are involved in spoken language understanding, and that under certain listening conditions, frontal regions are involved in discriminating speech sounds.

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.

How the human auditory system extracts perceptually relevant acoustic features of speech is unknown. To address this question, we used intracranial recordings from nonprimary auditory cortex in the human superior temporal gyrus to determine what acoustic information in speech sounds can be reconstructed from population neural activity. We found that slow and intermediate temporal fluctuations, such as those corresponding to syllable rate, were accurately reconstructed using a linear model based on the auditory spectrogram. However, reconstruction of fast temporal fluctuations, such as syllable onsets and offsets, required a nonlinear sound representation based on temporal modulation energy. Reconstruction accuracy was highest within the range of spectro-temporal fluctuations that have been found to be critical for speech intelligibility. The decoded speech representations allowed readout and identification of individual words directly from brain activity during single trial sound presentations. These findings reveal neural encoding mechanisms of speech acoustic parameters in higher order human auditory cortex.