Google released affordable end-user TPUs - Will Mycroft benefit from that?


#1

Hi everyone,

today I’ve read an article about google releasing some end-user TPUs for comparably cheap money.

I never worked with TPU’s myself (or TensorFlow, for that matter), but if I recall correctly both DeepSpeech and Mimic II use TensorFlow for their machine learning models. These two systems require a lot of computational power to run at acceptable speed, and this is the main reason that keeps Mycroft from working without the need of cloud servers for TTS and / or STT.

With TPUs available for end users (even with a raspberry pi per USB, see the link above), could this development help to get us closer to a voice assistant that works well without cloud computing? Maybe not even this first generation of end-user TPUs, but some later product? Maybe someone who understands a bit more of the matter could give us some insights here :slight_smile:


#2

The edge tpu’s are highly limited in what they can accelerate. Right now it’s all image processing, similar to the movidius NCS (same tech in AIY vision kit).


#3

Yeah looks like a way to go yet, but interesting to see Google recognising the need/desire from people for this use case.

I like Hackaday’s take away from it:

“We would like to congratulate the Raspberry Pi foundation for creating something so ubiquitous even Google feels the need to ride the coat tails.”