In this Huberman Lab Essentials episode, my guest is Dr. Eddie Chang, MD, a neurosurgeon and Chair of the Department of Neurological Surgery at the University of California, San Francisco (UCSF). We discuss the neural circuits underlying speech and language, including how the brain controls the larynx, vocal folds and articulators to shape breath into words. We also explore his pioneering work on speech neural prosthetics — brain-machine interfaces that allow paralyzed patients to communicate by decoding neural activity into speech and avatar-driven facial expressions. Additionally, we examine the neurobiology of stuttering, the role of auditory feedback in fluent speech, and the broader ethical questions surrounding brain augmentation technologies. Read the episode show notes at hubermanlab.com. Thank you to our sponsors AG1: https://drinkag1.com/huberman BetterHelp: https://betterhelp.com/huberman Function: https://functionhealth.com/huberman
They distinguish speech as the motor act of producing sound from language as extracting meaning. Language includes pragmatics, semantics, and syntax, while speech is one output modality among others like reading and sign.
Speech starts with exhalation and uses the larynx to generate voiced sound. The pharynx, mouth, tongue, and lips then shape that sound into vowels and consonants.
The larynx brings the vocal folds together during exhalation so airflow makes them vibrate. Typical vibration rates differ by anatomy, contributing to average pitch differences between men and women.
Crying, moaning, and similar vocalizations use overlapping mechanics but rely on different brain systems. People with injuries to speech and language areas can often still vocalize, suggesting older, conserved neural circuits.
They describe conditions like brainstem stroke and ALS where cognition can remain intact but voluntary movement and speech are lost. The resulting inability to communicate creates profound social and psychological isolation.
Chang’s lab targets cortical activity that normally controls the vocal tract to infer intended speech. Signals are routed to computers and decoded with machine learning to produce words for people who cannot speak.
A clinical trial implants electrode arrays over speech-related cortex and connects them to an external port for recording. The system digitizes neural signals and trains algorithms over weeks to map patterns to words.
They discuss a participant paralyzed for about 15 years who previously typed using a stick mounted to a cap. With the implant and decoding, he could generate selected words on a screen by attempting to speak.
The initial system used a 50-word vocabulary to constrain decoding and improve reliability. A language-model-like sentence space and autocorrect logic help compensate for decoding errors and improve output.
Emotional reactions like giggling can disrupt decoding by changing neural and motor patterns. This highlights how real-world behavior can interfere with signal stability and model performance.
They discuss shifting from medical restoration toward potential enhancement beyond normal abilities. Concerns include invasiveness, societal impact, and unequal access, while acknowledging incremental augmentation already exists via everyday tools.
They emphasize nonverbal cues and visual speech information for natural communication. The goal is to decode intended speech and drive an animated face that conveys mouth movements and expressions, not just text.
A realistic avatar could make control feel embodied and provide faster feedback than text-only output. This may improve training and usability for people learning to communicate through a speech neuroprosthetic.
Stuttering is framed as a disruption of fluent articulation rather than a loss of ideas or grammar. Anxiety can worsen stuttering, but it is not presented as the root cause.
Fluent speech requires highly precise timing across larynx, lips, jaw, and tongue, largely outside conscious awareness. Stuttering is described as a breakdown of that coordination, often intermittently.
They note that treatment commonly involves speech therapy strategies that create conditions for smoother output. For some people, the hardest part is initiating the first sounds of an utterance.
Hearing one’s own voice is described as a key feedback loop during speaking. Altering auditory feedback can change stuttering severity, suggesting interactions between motor commands and auditory processing.
They highlight speech as potentially the most complex motor behavior humans perform due to speed and precision across many moving parts. Examples like opera and freestyle rap illustrate the system’s performance demands.