Voice Control Made Simple – Panel Discussion
summary
Panelists report concrete results: running a simple keyword NN on an M55+U55 pair is about 100× speed‑up versus an M4 and can cut overall always‑on power because the core sleeps sooner; a commercial microwave ships with a truly natural recognizer that fits 1.5 MB, 10,500 phrases on an STM H7 (no OS) and works in the field; a Fluent end‑to‑end SLU and wakeword runs on Cortex‑M4 at ~30 MHz, <100 KB RAM; Arm publishes a Cortex‑M voice white paper and a two‑mic frontend PoC on GitHub.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 0:33Panelists
- 4:35Speech Recognition
- 10:41Cambridge Consultants
- 15:15Sensory
- 21:46Fluent
- 28:13Kova
- 29:51Work closely with customers
- 31:00Platform evaluation
- 32:40Audio frontend
- 37:40Demos
- 39:46Frontend partners
- 41:30Ondevice to cloud
- 43:55Internet IoT
- 45:40Accessibility
- 48:10Environment
- 51:25Natural Language
related talks
34mBuilding noise-immune speech interfaces for IoT
31mKeynote: The next big thing in machine learning is tiny
36mVoice Enabling Mobile Apps and Custom Hardware Devices
34mVoice-Controlled Lights on a Raspberry Pi
11mLightning Talk: Accelerating Experimentation...
27mWhat’s TinyML good for
42mWelcome to VOICE: Opportunities and Challenges for Brands
43mSpeeding up Whisper for speech recognition by 10x - Optimisations and walkthrough