Built on evidence, not assumptions

12+ years of research and development, backed by leading institutions and validated in real classrooms with real children.

What AI Teaches Us About Language

The Most Powerful Language Technology Wasn't Taught with Flashcards

The Traditional Approach

Traditional AAC and autism intervention often relies on discrete-trial approaches:

Show flashcard → Label item → Repeat

Teach words in isolation

Single-exemplar retrieval

Clinical observation: Many children can learn thousands of flashcard labels but struggle to communicate spontaneously. They may know the word "apple" when shown an apple picture, but struggle to ask for one when hungry.

Child looking confused at flashcards, disconnected from real communication
Child overwhelmed by too many choices on a communication board

Hick's Law: The Latency Problem

More choices = longer decision time = more communication latency.

When a child must search through hundreds of buttons to find the right word, the delay triggers:

Listener repairs ("What do you mean?")Anticipation (others finish their sentence)Frustration (giving up)

What Large Language Models Prove

The most advanced AI language systems demonstrate that language acquisition requires several key elements:

What Works

  • •Massive amounts of text in context
  • •Observing language in natural use
  • •Lexical imitation across situations

What Doesn't Work

  • •"Here's every word starting with A"
  • •Individual flashcard pairs
  • •Single-word exemplars in isolation

You cannot teach an LLM by feeding it flashcards. Why would we expect children to learn differently?

Classical vs. Jazz: A Perfect Analogy

Music reveals something fundamental about how we process information: visual symbols persist, while sound fades.

Classical musician reading sheet music

Classical Musician

  • Visual symbols persist: Sheet music stays in view, freeing working memory
  • Rich interpretation: Dynamics, attack, sustain, release all guided by notation
  • Remove the score: Complex pieces become difficult to recreate from memory alone
  • The visual scaffold enables performance that memory alone cannot sustain
Jazz musician improvising freely

Jazz Musician

  • Internalized patterns: Chord changes and forms learned through immersion
  • Generates in context: Complex solos emerge naturally from understood structure
  • Notation paradox: Transcribe the solo, and they may play it worse than improvised
  • Deep contextual learning enables generation that reading cannot replicate

Why This Matters for Language

Sound degrades instantly. When we hear speech, we must hold it in working memory to analyze it, a process demanding significant cognitive load. This is why subtitles help language learners: connected speech moves so quickly that auditory memory alone struggles to keep up, especially in a second language.

Text persists. Like sheet music for a musician, written words maintain their meaning without degrading. InnerVoice uses this principle: pairing speech with text reduces cognitive load and supports language comprehension in real time.

Visual symbols scaffold understanding that sound alone cannot sustain.

What This Means for AAC

→

Buttons/PECS as intro only, not permanent solution

→

Goal: Transition to typing, speech, or signing (lower latency)

→

Teach in chunks/gestalts, not isolated words

→

Support contextual use, not retrieval from massive grids

InnerVoice is built on this research foundation.

Our platform supports gestalt language development, reduces communication latency, and scaffolds transition to natural speech.

4→10
Skills Gained
documented in Missouri study
1,000+
Years
the exposure gap we close
12+
Years
of InnerVoice development
Multiple
Grants Received
from private foundations & government agencies
The 1,000-Year Problem

Why traditional AAC often fails

Neurotypical children in higher-income families are exposed to roughly 45 million words by age 4 (Hart & Risley, 1995). AAC users in twice-weekly therapy receive approximately 3,000-6,000 models annually based on recommended dosages (Binger & Light, 2007)—a staggering disparity that worsens when clinicians struggle to meet modeling targets in real-world practice.

Do the math: it would take over 1,000 years to close that gap through therapy alone. That's not a gap—it's a chasm.

InnerVoice solution: The app itself becomes a modeling partner. Every interaction demonstrates language use. Every button tap shows what words mean in context.

InnerVoice Approach

1Watch Visual StoriesKey step
2Practice with support
3Create own stories

Traditional AAC Approach

1Watch/LearnSkipped
2Practice symbols
3Use in isolation
Featured Implementation Study

When Context MeetsHigh Expectations

A multi-year implementation study documenting InnerVoice outcomes in a Missouri state school serving students with high support needs

Primary Outcome

4
skills (2015)
2 YEARS
10
skills (2017)

mastered communication skills measured by the Communication Matrix

4→10
Words Mastered

More than doubling communication abilities

The Context

Prior to InnerVoice implementation, students at this school had limited access to formal communication systems. Some used picture schedules or basic picture exchange, while others communicated primarily through physical expression—vocalizations, gestures, or leaving situations that weren't working for them.

Teachers could often interpret students' preferences, but students lacked tools to independently initiate communication or respond to questions on their terms.

The Implementation

Teachers discovered InnerVoice through the Autism Speaks website and began daily classroom implementation in Fall 2016. The approach included:

  • 1Morning communication sessions using animated avatars
  • 2Integration across all school activities
  • 3Individual progress tracking using standardized tools

What This Meant in Practice

Expanded Communication Repertoire

The student gained access to new ways of expressing needs and preferences. Where previously his primary communication tools were physical (vocalizations, gestures, leaving the room), he now had additional options through the iPad-based system.

Increased Initiation

Data showed increased instances of the student initiating requests using the device. This gave him more control over his environment and more ways to get his needs met efficiently.

Teacher Observation

They believed that with help our students could do more… and they have.

— Special Education Teacher, Missouri State School

What the Data Showed

Measurable growth in communication skills over two years

The student's measured communication skills expanded from 4 to 10 over two years. More practically: he gained new ways to get his needs met. Where physical expression had been his primary tool—because nothing else was available—he now had additional options through the device.

The data tracked increased instances of him initiating requests. More ways to be understood. More control over his environment.

Documented real-world outcomes in students with high support needs who often show limited measurable progress with traditional interventions

Built on evidence, not assumptions

12+ years of research and development, validated through federally-funded studies with real children.

NSF SBIR Phase I

Grant Number: #1520587

Year: 2015

Description: Synthesized Emotional Communication

Details: Created an emotionally expressive software-based speech-generating communication system for individuals with autism spectrum disorder.

NIH SBIR Phase II

Grant Number: #1920345

Year: 2019

Description: Video Assisted Speech Therapy/VAST

Details: Conducted research on video-assisted speech intervention to improve communication skills in children with autism.