An AI service that runs right on the device, with a lightweight LLM on the SoC. In broadcast and telecom set-top boxes, it filters noise and echo and recognizes speech without sending audio to a server. Built for operators and set-top box makers.
Instead of sending audio to a server, a lightweight LLM on the set-top box's SoC filters noise and recognizes speech right there.
The set-top box sends the audio it hears to a server across the network, then acts on the result that comes back.
A lightweight LLM on the SoC strengthens speech recognition and noise removal in broadcast and telecom set-top boxes.
Noise and echo are filtered from the microphone input, then a lightweight LLM understands the command and drives the device.
The set-top box microphone picks up the user's voice together with TV audio and room noise.
Echo coming back from the TV speaker and background noise are removed so only the voice remains.
A lightweight LLM on the SoC recognizes the speech and understands the command on the device.
The set-top box carries out the command, such as changing the channel.
AI speech recognition optimized for noise removal understands commands that mix Korean and English.
AI checks remote control signals to identify errors and malfunctions, and detects and excludes abnormal input.
It collects and parses VOD schedule information and analyzes it with AI optimized for the NPU to make recommendations for each viewer.
AIO2O's lightweight LLM sits on the AI chip (SoC), the set-top box carries both, and broadcasters and telcos deliver the service.
The semiconductor (SoC) that runs AI inside the device
Noise and echo removal plus speech recognition, on the device
The device that carries the AI chip and software
Deliver voice services through the set-top box
We are developing features that make more of the set-top box chip.
Vision recognition on the set-top box chip's NPU spots billboards in the broadcast in real time and produces data on when and how long ads were shown.
Broadcast speech is turned into text in real time, and subtitles in several languages are generated automatically.
Inference runs on the device and learning in the cloud, extending to services such as ad performance reports.
The full patent list is on the intellectual property page.
For companies that build or run voice-controlled set-top boxes and appliances.
Operators that deliver voice services through set-top boxes.
Makers that put AI chips and voice software into devices.
Start with device specifications and the conditions in which people use voice commands. Compare local and server processing, then confirm feasibility and performance test conditions for the target device.
| Topic | Information to prepare | Questions to confirm |
|---|---|---|
| Device specifications and software environment | List the chip, memory, operating system, microphone setup and current voice processing flow of the target device. | Confirm what can run on the device, how software would connect and what resources it would require. |
| Usage conditions and commands | Describe languages, commands, distance from the device, background noise and conditions such as TV playback. | Confirm test conditions for recognition accuracy and response time, including failures and retries. |
| Processing location and data management | Map the current path of voice data and which functions require a server connection. | Confirm the division between device and server processing, network dependencies, and whether voice data is stored or transmitted. |
| Test devices and operational conditions | List target device models, test conditions and schedule, and the teams responsible for updates and operation. | Confirm whether a trial is available, its scope and acceptance criteria, and support for updates and operation. |
Our team will review it and reply.
An AI service that runs right on the device, with a lightweight LLM on the SoC. In broadcast and telecom set-top boxes, it removes noise and echo and recognizes speech on the device, without sending audio to a server.
Cloud AI sends data to remote servers for processing. On-device AI runs on a chip inside the device, so it does not wait on the network and does not need to send voice outside the device.
Speech recognition with noise removal and mixed Korean-English commands, remote control signal analysis that filters out malfunctions and abnormal input, and per-viewer recommendations from VOD metadata analysis.
Broadcasters and telecom operators that deliver voice services through set-top boxes, and set-top box and appliance makers that put AI chips and voice software into devices.
Ad exposure analysis for sports broadcasts using NPU vision recognition, real-time speech recognition with multilingual subtitles, and hybrid AI that splits inference on the device and learning in the cloud.
Last updated
Our team will reply to the contact details you left.
Contact sales@o2o.kr
