Real-Time, Resource-Efficient Voice AI for Edge & Distributed Computing

CandyVoice specialises in real-time, resource-efficient AI and DSP technologies for voice processing across embedded, edge and distributed computing environments.

Our expertise combines advanced Voice AI with strong optimisation for low-latency and resource-constrained execution. We develop technologies that can operate locally on embedded hardware and edge devices, as well as within gateways, servers and distributed infrastructures.

Our core expertise includes:

• AI speech enhancement and noise filtering for extreme acoustic environments, with real-time execution on constrained hardware

• Low-bitrate voice coding and packet-loss resilience, maintaining communication quality under degraded network conditions

• Real-time voice anonymisation, protecting biometric identity while preserving intelligibility and natural communication

• Real-time voice deepfake detection, supporting trustworthy voice communications and content authenticity

• Real-time voice imitation and transformation, using lightweight AI architectures

• Embedded and Edge AI optimisation, enabling low-latency inference with reduced computational, bandwidth and energy requirements

A key CandyVoice capability is the development of high-performance AI models with a very small computational footprint. This enables voice processing to be executed close to the data source instead of systematically relying on centralised cloud resources, supporting lower latency, privacy, resilience and efficient use of computing and network resources.

This expertise is particularly relevant to European R&D projects addressing distributed AI, decentralised computing, the Edge-Cloud Continuum, heterogeneous computing, trustworthy AI, privacy-preserving processing and resilient communications.

For HORIZON-CL4-2027-04-DATA-08 and HORIZON-CL4-2027-04-DATA-03, CandyVoice can contribute both mature Voice AI technology components and demanding real-time workloads/use cases for integration, experimentation and validation within distributed and Edge-Cloud architectures.

CandyVoice technologies reach TRL 6–8 depending on the component and application. CandyVoice received the 2026 International Critical Communications Award (ICCA) – “Best Use of Advanced Technology” for its Mission-Critical AI Voice Stack.