AI-based real-time speech separation on edge devices for noisy industrial environments.

VoxFocus

New on Amplifa

~2021Start of development (karaoke experiment)

About VoxFocus

VoxFocus develops an AI-based audio platform for speech separation: the AI Audio Core separates relevant speech in real time from machine, tool and ambient noise before the signal reaches the ear or digital speech processing systems. Processing takes place locally on edge devices, i.e. without a permanent cloud connection. Two product lines emerge from this common core: COMMAND is a communication layer for hearing protection (hub plus exciter, prototype), while SENSE serves as an acoustic front-end integration for speech recognition, HMI systems and robotics. Target industries include manufacturing, mechanical engineering, robotics, system integration as well as hearing protection and occupational safety. Performance is validated in laboratory and field tests covering speech intelligibility, latency, word error rate and robustness. VoxFocus emerged from a karaoke experiment for separating voice and music and is built on four years of research and development.

Processes: AI-based real-time speech separation on edge devices, AI Audio Core for noisy environments, COMMAND: communication layer for hearing protection (hub + exciter, prototype), SENSE: acoustic front-end integration for speech recognition, HMI and robotics, Laboratory and field tests (speech intelligibility, latency, word error rate), Software and module integration for partner products
Materials: AI Audio Core (edge processing), COMMAND – communication layer for hearing protection, SENSE – acoustic front-end
Industries: Industry/Manufacturing, Mechanical engineering, Robotics, System integration, Hearing protection/Occupational safety, Assistive hearing systems

01
Signal acquisition (microphone)
02
AI Audio Core: speech separation on edge devices
03
Delivery to the human (COMMAND) or to a speech system (SENSE)
04
Laboratory and field tests
05
Software and module integration for partner products

Technology & Processes

EDGE-AI

AI Audio Core (edge processing)

Local, real-time-capable speech separation without a permanent cloud connection.

COMMAND

COMMAND – communication layer for hearing protection

Prototype consisting of hub, exciter and specified hearing protection for noisy workplaces.

SENSE

SENSE – acoustic front-end

Preparation of speech for existing speech recognition, HMI and robotics.

Products & Applications

**COMMAND – hearing protection with exciter and external hub**Prototype concept · Hearing protection/occupational safety

**SENSE – human–machine communication**Acoustic front-end · Robotics/Industry

Inside the production

VoxFocus – We separate the voice from the noise

Industries

Industry/ManufacturingMechanical engineeringRoboticsSystem integrationHearing protection/Occupational safetyAssistive hearing systemsAssistive hearing systems (future)

Additional services

AI-based real-time speech separation on edge devicesAI Audio Core for noisy environmentsCOMMAND: communication layer for hearing protection (hub + exciter, prototype)SENSE: acoustic front-end integration for speech recognition, HMI and roboticsLaboratory and field tests (speech intelligibility, latency, word error rate)Software and module integration for partner products

History

~2021
Started as a karaoke experiment for separating voice and music during the pandemic; four years of research & development
2024
First place in the Rhineland-Palatinate ideas competition (Mainz region, maturity level 1)
2025
Top 50 in the start-up competition – Digital Innovations (Winter 2025)
2025
AI Founder Fellowship (MISSION KI)
2025
Funding through the AI NATION grant program