Devices & companions

Put a responsive language coach inside the device.

Embed listening and speaking assessment into connected learning devices so practice feels immediate, conversational and available away from a screen.

Explore device moments

Embedded assessmentStreaming interactionChild-tuned models

Child practising spoken language with a connected learning device

Designed for

Learning tablets, reading pens, robots, headsets and other connected devices

Use case

Low-friction: voice interaction without opening a dashboard
Streaming: feedback aligned to the device moment
Device-aware: logic for microphones and noisy environments

/three device moments

A language coach that lives in the object the learner already holds.

The same compact evidence can power a companion speaker, a reading pen or a headset drill. The device decides how the next second feels.

Adult learner speaking with a tabletop companion speaker and a simple waveform on a tablet

Companion speaker coaching

Device experience: Ask, listen, respond—without opening a dashboard.
Product signal: Streaming evidence and a compact next-step field the firmware can play, light or speak.

Teenager reading a textbook aloud with a scanning reading pen

Reading-pen practice

Device experience: Scan a line, say it, hear whether to continue.
Product signal: Short-utterance scoring mapped to printed text, with capture checks for distance and noise.

Adult learner doing a spoken drill with a wireless headset and a compact learning tablet

Headset language drill

Device experience: A spoken drill that stays in the headset, not on a score page.
Product signal: Low-latency pronunciation and completeness fields, plus a fallback when connectivity drops.

See the assessment engine, without the agent layer.

Choose the product SDK experience or the MCP agent walkthrough based on how you plan to integrate.

Choose a live demo

/built for connected devices

Coaching has to live in the object, not in a dashboard.

Hardware teams need a speech contract that fits firmware, microphones and short attention spans. The device should acknowledge, cue and continue—without turning every turn into a score screen.

Built for these activities

Reading pens
Companion speakers
Learning tablets
Headsets
Story cards
Voice drills

01
Feedback inside the interaction
Return compact evidence fast enough for a spoken retry, a light, a sound or a next card—not a report the learner has to open.
02
A contract across device versions
Keep assessment fields separate from firmware and UI so the same scoring layer can map to pens, tablets, robots and headsets.
03
Microphone-aware responses
Clipping, distance, noise and interrupted turns should change what the device says next. Weak capture is not weak pronunciation.

/on the device vs later

Choose what must feel instant, and what can wait.

Decide which feedback must feel instant on the device and which evidence can be processed or reviewed later.

The situation

A voice-enabled device needs more than speech-to-text to correct pronunciation well.

What changes

On-device experiences that can score, respond and guide the learner’s next attempt.

/who acts on the evidence

The learner talks to the object. Hardware and content teams share one scoring layer.

Learner

Needs: A natural prompt and response that does not feel technical.

Immediate coaching inside the object they already use.

Hardware team

Needs: A predictable speech contract across device versions.

Assessment evidence separated from firmware and interface logic.

Content team

Needs: Reusable tasks that work across stories, cards and lessons.

One scoring layer mapped to different device experiences.

/prompt to coach

Prompt in context, capture cleanly, then respond in the device’s voice.

If the response feels like software, the hardware experience has already failed.

01
Prompt in context
The device asks for a word, sentence or spoken answer at the right moment in the activity.
02
Capture reliably
The microphone path handles wake states, clipping, background noise and interrupted turns.
03
Return compact evidence
The engine sends the score and diagnostic fields the device experience actually needs.
04
Respond naturally
The device acknowledges success, gives one cue, retries or continues without turning into a score screen.