High-tech augmentative and alternative communication (AAC) refers to electronic devices and software systems that help people who cannot rely on natural speech to express themselves. These range from tablet-based apps with symbol grids and text-to-speech output to eye-tracking systems, brain-computer interfaces, and AI-powered keyboards that predict entire sentences from a few keystrokes. The technology has advanced rapidly in the past decade, and the gap between what a person thinks and what the device can say is shrinking in ways that would have seemed speculative even five years ago.
Who Uses High-Tech AAC and Why
The population that depends on these tools is broader than most people realize. Amyotrophic lateral sclerosis (ALS) is the condition most publicly associated with AAC, partly because of Stephen Hawking’s iconic speech-generating device. But high-tech AAC also serves children with autism who have limited spoken language, adults who have had strokes or traumatic brain injuries, people with cerebral palsy, and patients in intensive care units who are temporarily voiceless due to intubation or tracheostomy. An estimated 97 million people worldwide have complex communication needs that could benefit from AAC of some kind.
The needs of these groups vary enormously. A child learning language for the first time requires a device that supports vocabulary growth and literacy development. A person with ALS whose motor function is progressively declining needs a system that can adapt as they lose the ability to use their hands, then their head, and eventually their eye movements. An ICU patient recovering from surgery may need a simple touchscreen interface for only a few days. High-tech AAC is not one device but a category of solutions, and the engineering challenge is matching the right combination of input method, language system, and output to each user’s abilities and context.
How People Control These Devices
The input side of AAC is where much of the engineering complexity lives. The simplest high-tech option is a touchscreen tablet running AAC software, where the user taps symbols or types text that the device speaks aloud. But many AAC users cannot reliably touch a screen, so alternative input channels become essential.
Eye-tracking systems follow the user’s gaze to determine which letter, word, or symbol on the screen they are looking at. The user dwells on a target for a set period, and the system registers that as a selection. This works well for people who retain good eye control but have lost use of their limbs. A persistent challenge with gaze-based input, though, is what researchers call the Midas Touch problem: the system cannot easily distinguish between looking at something to select it and simply looking at it while thinking. One recent framework addresses this by separating navigation from selection, letting the user’s gaze move freely across the screen while requiring a deliberate blink to confirm a choice. That same system uses a predictive gaze model trained on roughly six million eye-tracking samples to stabilize cursor movement, reducing the cognitive effort of holding the eyes perfectly still on a small target.
Switch scanning is another common approach. The user activates a single switch, which can be anything from a button pressed by a finger or elbow to a sensor triggered by a head movement or a puff of air. The device highlights options one at a time or in groups, and the user activates the switch when the desired item is highlighted. It is slower than direct selection but works for people with extremely limited voluntary movement. A hybrid approach that combines eye tracking with switch scanning has shown promise for users with severe motor impairment: in one study, the combined system produced significantly fewer errors than eye tracking alone, though typing speed was slower.
Surface electromyography (sEMG) offers yet another pathway. Sensors placed on the face detect tiny electrical signals from facial muscles, and the user selects phonemic targets on screen by activating specific muscle groups. This approach has been evaluated as an input method paired with a phoneme-based interface for speech synthesis, giving users a way to construct spoken output sound by sound rather than letter by letter.
Brain-Computer Interfaces
For people who have lost nearly all voluntary movement, including reliable eye control, brain-computer interfaces represent the most advanced frontier of AAC input. These systems read neural activity directly and translate it into communication output. Most BCI research for AAC has focused on noninvasive methods, particularly electroencephalography (EEG), where sensors on the scalp detect brain signals associated with attention or intention. One well-studied paradigm flashes letters on a screen and detects which letter the user is attending to based on the brain’s response to each flash.
Invasive BCIs, which use electrodes implanted directly in the brain, are far more precise. A striking demonstration published in 2025 reported a “brain-to-voice” neuroprosthesis that synthesized speech in real time from neural signals. The system used 256 microelectrodes implanted in the brain region responsible for speech motor planning in a man with ALS and severe dysarthria. It decoded not just the phonemic content of what he intended to say but also paralinguistic features like intonation, enabling him to modulate the pitch of his synthesized voice and even sing short melodies. This goes well beyond the letter-by-letter spelling that characterized earlier BCI-AAC systems; it begins to restore something closer to natural vocal expression.
How AI Is Speeding Up Communication
Even with good input hardware, communication through AAC is slow. A typical speaking person produces 150 or more words per minute. Eye-gaze typing, even with practice, rarely exceeds 10 to 15 words per minute. That speed gap is socially devastating: conversations move on, jokes lose their timing, and the AAC user is often left responding to a topic that has already passed. This is the problem that artificial intelligence is most aggressively targeting in the AAC space.
Traditional word prediction, the kind built into smartphone keyboards, helps somewhat by suggesting likely next words. But large language models (LLMs) open a different possibility: the user types a highly abbreviated version of their message, and the model expands it into full natural language. A system called SpeakFaster, built around an LLM-powered interface, saved 57% more motor actions than traditional predictive keyboards in offline testing. In real-world trials with two eye-gaze AAC users with ALS, text-entry rates rose 29 to 60 percent above their baselines because the model could predict whole phrases from just a few keystrokes.
The broader landscape of AI in AAC extends beyond text prediction. Computer vision can help devices understand context, such as recognizing that the user is at a restaurant and preloading food-related vocabulary. Natural language processing can improve the grammatical quality of symbol-based outputs, turning a sequence of tapped icons into a sentence that sounds natural when spoken aloud. Machine learning can adapt to a specific user’s patterns over time, prioritizing the words and phrases they use most. These components are still being refined, and researchers have noted that while AI’s promise for AAC is substantial, the field needs to proceed with caution around issues like data privacy and the risk of the system “speaking for” the user rather than expressing their intent.
Sounding Like Yourself
One of the most emotionally significant developments in high-tech AAC is the ability to generate a synthetic voice that sounds like the user’s own voice rather than a generic robotic one. For many people facing progressive conditions like ALS, voice is deeply tied to identity. Losing it feels like losing a piece of who you are. Voice banking, the practice of recording a set of speech samples while the person can still speak clearly, has been available for some time. But modern AI voice synthesis can now produce remarkably natural-sounding clones from those recordings.
A recent study evaluating a HiFi-GAN model for voice generation in ALS patients found that synthetic voices closely matched natural speech in pitch, with an average of about 155 Hz for both. Listeners rated the synthetic samples as highly similar to natural speech, giving them an average quality score above 6 on a 7-point scale. Intensity differed more, with synthetic voices running louder by about 13%, but the overall perceptual similarity was striking.
Voice cloning raises ethical questions that the field is actively working through. If a person’s voice can be synthesized from recordings, who controls that voice data? What happens if the recordings were made before the person lost the ability to consent to specific uses? These concerns are particularly acute for people with ALS, who may bank their voice early in the disease and then lose the ability to provide ongoing consent as their condition progresses. A guiding-principles framework for ethical integration of voice cloning into AAC has been proposed, though the specifics are still evolving. For now, the technology is moving faster than the ethical scaffolding around it.
Children, Literacy, and Language Development
High-tech AAC is not just a tool for producing speech output; for children, it can serve as a platform for language learning itself. A common misconception is that giving a child a speech-generating device will discourage them from developing spoken language. The evidence runs in the opposite direction: AAC use tends to support rather than replace natural speech development in children who have some vocal ability.
For autistic children who use high-tech AAC, literacy instruction delivered through the device has shown encouraging results. A parent-delivered intervention targeting early literacy skills, including sight words, letter sounds, and sound blending, produced large to very large gains in all three areas over just 10 weeks. The children in that study each entered the intervention knowing fewer than ten sight words and fewer than five letter sounds, so the starting point was low, but the rate of progress was notable. Generalization to sound blending, a skill not always directly trained, was an especially promising finding.
The key insight here is that AAC devices are not passive output machines. When designed well, they are interactive language environments. A child navigating a grid of symbols is making linguistic choices, building vocabulary, and learning the structure of communication. The device becomes a scaffold for cognitive and linguistic development, not just a substitute for a missing function.
AAC in Intensive Care
One setting where high-tech AAC is gaining traction is the intensive care unit. Patients who are intubated or have tracheostomies often cannot speak at all, and the experience of being unable to communicate basic needs, pain levels, or emotional distress is a major source of anxiety and frustration. Traditional low-tech solutions like alphabet boards and picture cards help, but they are limited in what they can express and require a communication partner who is patient and attentive.
Systematic reviews of high-tech AAC in ICU settings have found broadly positive results. AAC tools increased the number of communication interactions patients were able to have, improved patient satisfaction with communication, and reduced the frustration and negative emotions associated with being voiceless. Patients generally perceived the technology as useful and easy to learn, and most expressed willingness to continue using it. The devices helped with practical needs like symptom identification and pain management, but also with something less tangible: the feeling of being heard.
ICU use highlights a different design challenge than chronic AAC. These patients often have no prior experience with the technology, are cognitively stressed, and may be using it for days rather than years. The interface needs to be learnable in minutes, not weeks. It also needs to cover a specific vocabulary: pain descriptors, requests for repositioning, emotional states, questions about what is happening. Some ICU-focused AAC apps are designed around exactly this kind of constrained, high-priority vocabulary rather than the open-ended communication that a long-term user needs.
Progressive Conditions and Staging
ALS presents a particularly difficult AAC challenge because the disease does not stand still. A person newly diagnosed may have mildly slurred speech that a listener can still understand. As the disease progresses, intelligibility drops, voluntary movement narrows, and eventually some patients enter a locked-in state where they are aware and cognitively intact but have little or no voluntary motor output. AAC strategies need to evolve alongside the disease.
A stage-based approach to communication rehabilitation in ALS has been outlined in the clinical literature. In the early stage, interventions focus on voice banking and speech amplification while the person can still speak. The middle stage, between the onset of unintelligible speech and the locked-in state, is where most high-tech AAC comes into play: eye-tracking systems, switch-based devices, and predictive text software, chosen based on what motor functions the person still has. Late-stage interventions, after a patient has entered a locked-in state, rely on the most technologically advanced AAC devices, including BCI systems, because no other method can reach the person.
Timing matters enormously. A person who waits too long to start learning AAC may find that by the time they need it, they lack the motor skills to operate the device they were planning to use. Clinicians increasingly advocate for early introduction, not because the person needs the device immediately, but so they can become proficient with it before their abilities change.
Why People Stop Using Their Devices
Device abandonment is one of the most discussed problems in the AAC field. A person is assessed, matched with a device, trained on it, and then stops using it within months. The reasons are varied and not always what you would expect.
A systematic review of barriers and facilitators identified a long list of factors that influence whether someone sticks with a high-tech AAC device. Ease of use and reliability of the hardware are obvious ones. But less obvious factors matter just as much: whether the device’s synthetic voice matches the user’s age, gender, and language; how long it takes to generate a message; whether family members support and engage with the device; how communication partners respond when the person uses it; and whether the user had meaningful input into the decision about which device to get. When a clinician selects a device without consulting the user and their family, abandonment rates climb.
Some researchers have argued that the concept of “abandonment” itself needs rethinking. Rather than framing it as a failure, one proposed model reframes the decision to stop using a device as a rational choice the person and their family make within a broader range of multimodal communication options. A person might set aside their high-tech device not because it failed but because they found that a combination of gestures, low-tech boards, and partner-assisted communication met their needs more effectively in their daily life. Framing this as abandonment pathologizes a reasonable decision.
Barriers Beyond the Technology
Even when the technology works well, access remains uneven. Funding is a persistent obstacle. High-tech AAC devices with dedicated hardware can cost thousands of dollars, and insurance coverage varies dramatically by country and even by region within a country. A survey conducted in Saudi Arabia found that although many AAC users were able to obtain devices, they identified significant barriers including insufficient awareness of AAC’s importance among professionals and the public, high costs, limited funding sources, and a lack of devices adapted to the Arabic language.
That last point, language adaptation, is a global issue. A scoping review of AAC development in languages other than English found only 22 studies in the entire literature, and most focused on linguistic factors like vocabulary selection and symbol design. Cultural factors, such as how different communities conceptualize disability, politeness norms in conversation, or the role of family versus individual autonomy in communication decisions, were rarely addressed. The result is that the vast majority of high-tech AAC systems are designed for English-speaking users in Western cultural contexts, and adapting them to other languages and cultures is not simply a matter of translation.
The Social Reality of Aided Conversation
Technology can put words on a screen and push them through a speaker, but communication is a social act, and the dynamics of conversation change when one participant is using AAC. Turn-taking, one of the fundamental structures of human interaction, works differently. A speaking person produces a response in under a second. An AAC user might take 30 seconds or more. The conversation partner has to learn a different rhythm, resisting the urge to fill silences or finish the AAC user’s sentences.
Research examining spontaneous conversations between young men who used speech-output devices and their mothers found that the participants developed a structured three-part process for managing turn transitions. Despite the inherent slowness, the participants themselves did not consider conversational speed a challenge. The communication partners had adapted, and the conversational flow, while different from unaided interaction, had its own functional rhythm. This finding pushes back against the assumption that slow communication is inherently impoverished communication. Speed matters, but so does the quality of the relationship and the willingness of both parties to adjust.
Training communication partners is, in many ways, as important as training the AAC user. A device that works perfectly in a clinical assessment can fail in a family dinner if nobody at the table knows how to wait for the response, where to look, or how to acknowledge the output. Some AAC programs now include explicit partner training as a core component, recognizing that the technology is only as effective as the social environment around it.
What Is Coming Next
The trajectory of high-tech AAC points toward systems that are faster, more personalized, and more seamlessly integrated into daily life. LLMs are still in early adoption for AAC, and their ability to understand context and intent will continue to improve. Brain-to-voice neuroprostheses are currently experimental, confined to a handful of research participants, but the proof of concept has been established: it is possible to decode intended speech from neural activity in real time and produce intelligible, emotionally expressive synthetic voice. The gap between that laboratory demonstration and a commercially available product remains large, involving questions of surgical risk, long-term electrode stability, and cost. But the direction is clear.
On the software side, multimodal input systems that combine gaze, switch, facial EMG, and even residual speech into a single adaptive interface are being explored. The idea is that the system should meet the user wherever their abilities are on a given day, automatically shifting input strategies if, say, fatigue makes eye tracking unreliable in the afternoon. Personalized AI models that learn not just a user’s vocabulary but their communication style, humor, and conversational preferences could eventually produce output that feels less like generated text and more like the person’s own voice in every sense of the word.