On-Device AI
We bring real-time AI to your device. We specialize in deploying audio-centric models on hardware ranging from mobile phones to embedded systems.
Through the development of the timbre transfer algorithm for Neutone which runs in realtime on CPUs, including embedded devices, as well as collaboration with various vendors in deploying AI models on restricted hardware devices we have gained extensive experience in this area. We can provide consulting services for studying the feasibility, or facilitating the deployment of various AI models that run locally on device. We are mainly focusing on audio centric models such as speech recognition, sound event detection, anomaly detection, voice and SFX generation, but image or language models are also possible. Examples of devices range from more powerful ones like laptops and mobile phones to smaller ones like Raspberry Pi or NXP boards.
Data protection and privacy
By keeping all the model inference local, all concerns regarding data protection and privacy can be mitigated because no data transfer to any servers occurs in the first place. This can also help in compliance with regulations.
Offline usage
By enabling offline inference, usage in certain environments such as limited internet access or restrictive access policies is now possible. Low cost of deployment.
Low cost of deployment
Since the model runs locally on the users device, the cost necessary for backend server design, development, deployment and maintenance as well as bandwidth can be greatly reduced.
Intelligent Assistants, Robotics and IoT Solutions Powered by AI
On-Device AI Deployment to Personal Computing Platforms
Interactive AI-Driven Installations
Neutone Hardware Prototype
We are working on bringing our proprietary neural tone transfer model to hardware, running on Raspberry Pi compute models with an astonishing inference time of mere milliseconds. Possible applications are in hardware for electronic musicians, interactive toys etc.
On-Device Real Time Solutions
We developed a bespoke AI solution for Panasonic's Nicobo, optimized to run efficiently in real time on hardware with limited capabilities while maintaining excellent performance.
AI Music Detection in a Chrome Extension
We trained and deployed an AI Music Detection model based on the SONICS research paper by Rahman et al. (2025). The model detects whether a song from the Spotify Web Player has been fully generated with AI services such as Suno or Udio. It runs completely on device in the user’s browser being able to process tens of music tracks in a few seconds.
