clayoz

Clayoz // starting with autistic children

Speech recognition built for non-standard speech.

Some children develop highly individualized ways of speaking that mainstream speech recognition was not built to understand. Personalized tools can help, but prompt, record, label, and repeat workflows are difficult for autistic children who do not speak on cue. Clayoz learns through everyday communication, with simple parent guidance, to help make needs, choices, thoughts, and ideas understood.

Build with us See what we're building

early stage / building in public

The child's job is to communicate. The technology's job is to learn.

// why_clayoz

Autistic children shouldn't be an edge case.

For autistic children with speech difficulties, being misunderstood can make everyday moments harder, whether asking for help or sharing a joke. General-purpose speech-recognition systems are not trained around each child's individual sounds, patterns, pace, and vocabulary.

Important work on non-standard speech has shown what personalized recognition can do. But today's approaches largely center on adults or speakers who can deliberately complete a structured training process. Autistic children remain underserved because these approaches do not reflect how they naturally communicate.

Clayoz starts from the other direction: build the data collection, device experience, and personalized model around the child's speech, vocabulary, context, and natural way of expressing themselves.

// now_building

Built for communication as it happens.

[01]

Listen naturally

Start a parent-controlled listening session once. The child can communicate when they are ready, without repeated buttons or prompts.

[02]

Capture what matters

Identify useful speech moments from everyday interaction instead of depending entirely on staged recordings and repeated phrases.

[03]

Improve together

Give parents a fast way to confirm meaning, correct mistakes, and improve personalized recognition over time.

// reduce_the_ritual

Children shouldn't have to train like data annotators.

Tools for non-standard speech have shown the promise of speaker-specific learning. But most personalization workflows still ask the speaker to produce clean, labeled recordings on demand. For many autistic children, that process introduces friction and changes the communication we are trying to understand. Clayoz is working toward the next step: natural communication, intelligent data collection, and lightweight parent guidance.

common_personalization_workflow

  1. press record
  2. repeat this phrase
  3. label the recording
  4. repeat many times

clayoz_direction

  1. start listening
  2. communicate naturally
  3. parent confirms useful moments
  4. recognition improves

// product_learning

We started with a button. It showed us the real problem.

Our first prototype used tap-to-talk. Before every expression, an adult had to bring out a device and press record. In our founder's experience, that ritual interrupted natural communication. The next version moved to continuous transcription: start listening once, then allow speech to happen naturally.

working now

  • continuous transcription
  • browser recording and labeling
  • personalized model fine-tuning

working toward

  • intelligent sample capture
  • lightweight parent review
  • speaker-aware listening
  • incremental personalization
  • privacy-preserving local processing

// why_clayoz_exists

It started with our founder's daughter.

Our founder's daughter is neurodivergent and has difficulty being understood by conventional speech models. For their family, this gap isn't abstract. It affects daily connection, independence, and expression.

Clayoz is an early-stage effort to turn advances in AI into something useful in real life. We're building it openly and looking for people who want to help make it work.

// operating_principles

How we intend to build.

// open_connection

This is too important to build alone.

We're looking to connect with people with speech differences, families, speech-language practitioners, accessibility thinkers, speech researchers, and engineers.

Start a conversation