Sport Earbuds That Understand and Act, Without the Cloud
SHOKZ keeps high-frequency voice commands in a local loop with the edge0 on-device speech model, sending complex intents to the cloud on demand — instant, offline-ready and far fewer tokens.
Smart Earbuds
We want voice to be part of the earbuds themselves: no network dependency — the moment a user speaks, the device acts. edge0 delivered natural-language understanding, millisecond-level response and local execution inside our existing hardware resources, so we can ship a stable, consistent experience on every unit while keeping cloud cost under control at scale.
The Cloud Can Complete a Voice Interaction,
but Cannot Guarantee Every Command Works
Open-domain QA can go to the cloud on demand; playback control, sport data and device state are high-frequency core capabilities that must run locally.
The Network Can't Be a Prerequisite
Running and cycling routes cross weak-signal and fully offline zones. A voice experience that depends entirely on the cloud fails the moment the network does.
Control Must Happen Instantly
Skip, pause and volume are instant operations. A cloud round-trip adds visible waiting and breaks the feel of direct control.
Scale Must Not Multiply Cost
Shipped earbuds generate huge volumes of high-frequency, simple requests. Paying cloud inference per call turns into a permanent, growing cost.
Customers Choose edge0 Because It Puts
Near-Cloud Understanding into Existing Consumer Hardware
Traditional on-device solutions are lightweight but locked to fixed commands; cloud models understand deeply but bring network and cost constraints. edge0 connects the two with a higher “intelligence density”.
A Smaller Model, Carrying Denser Intelligence
Inside an 18MB package and 42MB peak memory: speech recognition, natural-language intent understanding and device command callbacks.
*The upper-right region represents models that understand natural phrasing and run directly on consumer hardware. Positions illustrate capability, not test scores.
Near-Cloud Understanding
From fixed commands to natural phrasing: “next track”, “change song” and “skip this one” all map to the same device intent.
Runs on Consumer Hardware
Model compression, INT8 quantization and runtime optimization fit ASR + NLU inside existing memory and compute budgets.
Tuned for Sport Noise
Adapted to wind, footsteps, traffic and music playback for stable outdoor command recognition.
SDK Wired to Device Capabilities
Standardized intents call playback and motion APIs directly — no hardware rework for the brand.
On-Device First, Cloud on Demand —
Every Interaction Spends Only the Tokens It Needs
The edge0 SDK covers the full chain from voice input, recognition and intent understanding to task routing and command callbacks, switching paths automatically. High-frequency tasks such as playback control and motion-status queries close on-device; complex intents and open-domain QA go to the cloud on demand and return through one unified result. Brands never stitch two pipelines together, cloud calls drop, and core local commands keep working on weak or missing network.