How a Speech Brain-Computer Interface Turns Attempted Speech Into Words

A speech brain-computer interface (BCI) restores communication for a person with severe paralysis by following three sequential steps: it records electrical signals from neurons as the person attempts to speak, decodes those signals into the words they intend to say, and then converts that intent into text or synthesized speech. The device developer Paradromics describes this as a chain of Recording, Decoding, and Action — a mechanism, not magic, binding biology to technology.

Introduction: Restoring Communication with Brain-Computer Interfaces

Why does this matter? Conditions like ALS can paralyze both the body and the vocal tract, cutting off a person's ability to speak even though the speech-related regions of their brain remain active. A BCI creates a direct communication link between the brain and external devices without relying on the body's normal pathways of nerves and muscles, as the healthcare research firm DelveInsight explains. The U.S. Government Accountability Office notes that researchers are developing BCIs that allow people with paralysis to spell words on a computer screen. This article focuses on intracortical BCIs — devices surgically implanted in brain tissue — which the GAO says may be more appropriate for users with severe neuromuscular disorders. It explains the general mechanism, not any specific commercial product, and offers no medical advice.

The Three Stages of Speech BCI Communication

To trace how a thought becomes communication, follow the sequence below. Each stage transforms the data one step further: raw electrical impulses become filtered signals, filtered signals become intended words, and intended words become output you can read or hear. Paradromics illustrates the model with a "brain-to-text-to-speech" design, in which brain signals are translated to text and then converted to speech.

The sequential three-stage model of a speech BCI, from neural impulse to communication output
Stage	What happens	What the technology does
1. Recording	The user attempts to speak; neurons in speech-related brain regions fire.	Implanted electrodes capture electrical signals, amplify the relevant ones, and filter out unwanted components, then move the processed signals onto a computer.
2. Decoding	The raw signals — complex and noisy — are analyzed.	Advanced statistical models, including large language models, extract the actual intended words behind the neural activity.
3. Action	The decoded word intent is converted outward.	Intended words become synthesized text or audible speech, representing the user's neural intent as communication.

Each stage adds latency — the delay between attempted speech and the resulting text or speech — which is why the sequence matters as a whole, not just as separate parts.

Stage 1: Neural Signal Recording

Recording begins inside the brain. When a user attempts to speak, neurons communicate through action potentials — electrical spikes across their cell membranes — and the implanted device captures these signals directly from speech-related regions. According to Paradromics, some BCIs record single-neuron action potentials, while others can only measure the averaged activity of many neurons firing together. The signals are then processed: relevant activity from the neurons of interest is amplified, and unwanted components that do not help the BCI are filtered out. Finally, the processed signals must get out of the brain and onto a computer, where the real interpretation can begin.

Stage 2: Signal Decoding

Decoding is where raw electricity becomes meaning. The incoming brain signals are extremely complex and noisy, so advanced algorithms filter them to identify specific neural patterns — the same principle DelveInsight describes for detecting motor intention or cognitive-task patterns in other BCI applications. In a speech BCI, the system then passes these signals through various advanced statistical models, including large language models, to extract the actual intended words behind them. The GAO adds that new users often undergo an iterative training process: the person learns to produce signals the BCI will recognize, while the BCI's machine learning adapts to translate those signals reliably.

Stage 3: Action and Communication Output

Action turns decoded intent into something the world can perceive. The intended words are converted into a form of communication such as synthesized text or speech, directly representing the user's neural intent. Speed is the central design constraint here. Paradromics notes that natural conversation runs at roughly 160 words per minute, and a written transcript of natural speech conveys about 39 bits of information per second — so a speech BCI must recognize, decode, and output words quickly enough to keep pace with the rhythm of conversation rather than lagging noticeably behind it. A UC Davis-led system published in Nature Medicine in June 2026 shows the stage working in practice: its decoding algorithms translate neural signals into text and also enable cursor control, allowing full interaction with a personal computer from the trial participant's home.

The Impact: Empowering Individuals with Paralysis

The human stakes of this pipeline are easiest to see in the UC Davis trial, developed with Brown University and Mass General Brigham Neuroscience Institute. Trial participant Casey Harrell, a man living with ALS, uses the system independently in his own home. Neurosurgeon David Brandman, co-director of the UC Davis Neuroprosthetics Lab, said the work shows "we may have crossed a threshold, by empowering a person with paralysis to speak on his own terms." The findings highlight the potential of intracortical BCIs as assistive tools for people with severe motor impairments, including ALS, by enabling naturalistic communication and digital access that can improve independence and quality of life.

Empowering Communication: The Future of Speech BCIs

Individuals with severe paralysis can regain independent communication through advanced intracortical speech BCIs, provided the three-stage chain — recording, decoding, action — runs fast and accurately enough to feel like conversation. The measurable indicator to watch is the development and clinical adoption of BCI systems that enable higher words-per-minute rates and lower error rates in real-world use, with natural conversation speed of roughly 160 words per minute serving as the benchmark developers are aiming toward. Each published trial that moves users closer to that speed in home settings, as the UC Davis system did in 2026, marks the next concrete milestone.