Industrial AI Podcast Why Robots Need an Ear: The Sound of Industry
Dec 3, 2025
Dr. Finnur Pind, Co-founder and CTO at Treble Technologies, dives into the revolutionary world of synthetic sound data. He explains how simulating sound enhances robotics, enabling smarter machines with spatial awareness. The conversation uncovers the hurdles of accurate sound modeling and how sound could redefine AI applications beyond traditional text processing. Finnur also highlights Reykjavik's vibrant AI scene and Treble's vision for expanding audio-driven innovations in various industries.
AI Snips
Chapters
Transcript
Episode notes
Keep Inference On The Edge
- Run inference as close to the shop floor as possible to avoid latency and determinism issues.
- Use edge servers or local PLC/industrial PCs rather than cloud inference for production control.
Determinism Kills Cloud Inference
- Deterministic timing requirements on the shop floor make cloud inference impractical for many industrial use cases.
- Local recording and local processing reduce the risk of communication glitches affecting real-time control.
Money, Not Just Security, Drives Edge
- Cost and business models now drive the move away from cloud inference, not just data-security concerns.
- Machine builders prefer to ship solutions that run fully on-device to make ROI and sales feasible.