More detailed specifications for Luxonis rae?

Shyam

I was looking for (and haven't been able to find) more detailed specs for rae. Does anyone know where I can find them?

I was interested in understanding what sort of code I can run on it, what sort of edgeML capabilities does it have compared to, say, a Jetson Nano. And can I connect it to a Rpi or Jetson as a second brain?

Specifically interested in seeing if I can run a Whisper model on it for voice rec, whether it has a speaker or I can connect speakers to it + whether I can run a 7B Llama/Alpaca model on it for LLM capabilities.

erik

Hi Shyam ,

We will be adding the page for rae on our HW docs soon. It has integrated RVC3 which has 3 TOPS for AI, so I believe it's comparable to Jetson Nano. It has quadcore ARM A53 (same as RPi 3) where you could run your own code, like VIO or sparse SLAM. I am quite sure you won't be able to run LLMs🙂 You could stream the data (via WiFi) to a second brain, and do additional processing over there. Thoughts?

Thanks, Erik

Shyam

Thanks for the response and extra details Erik 🙂 sounds promising.

For context, I have a tiny LLM running on a VM on my PC at the moment 🙂 but I like the idea of streaming via WiFi to a locally hosted second brain.

Any specifications on what sort of microphone and speaker rae would ship with?

erik

Hi Shyam ,

Here are the technical specs of the rae:

Speaker:
Top level specs:
Resonant Frequency: 850 Hz
Frequency Response: 100 Hz to 20000 Hz
Power Rating: 0.7 W (max 1W)
Sound Pressure Level: 91 dBA

Manufacturer: CUI
Manufacturer PN: CMS-16093-078L100A-67

Mics:
Top level specs:
MEMS Microphone
Frequency Range: 60 Hz to 20 kHz
Sensitivity:- 26 dB
Sensitivity Tolerance: 1 dB
SNR - Signal to Noise Ratio: 64 dBA

Manufacturer: TDK InvenSense
Manufacturer PN: ICS-43434

Would that help?

Thanks, Erik