Chatterbox

2022-09-06 14:00 by Ian

Chatterbox is a evaluation platform for Audeme's MOVI embedded speech recognition board. Not much to say about this one. It was done at the request of my company's director of engineering, who wanted to examine the technical feasibility of doing something similar in our product.

I have this project cross-posted on hackaday.io. The software is written under Espressif ESP-IDF, and I have a driver for my abstraction layer published, as well.

The MOVI board has an Arduino form factor, so I whipped up a one-off support board with an ESP32 DevKit-C, regulator, and audio amp. The MOVI did a great job of being self-contained. Very little glue hardware was required to make it work.

The MOVI takes 12V directly, and has an embedded Linux stack running pocketsphinx for speaker-independent speech recognition. There is enough memory to train concurrently on 100 words. Which is more than enough for many products. If you have it configured to always listen, the current draw will be a gut punch for your batteries (a few Watts). But considering what "always listening" implies, that shouldn't be a surprise.

Boot time is reasonable (~2 seconds), but not quite fast enough to comfortably power the device down and re-enable it on a button press from the user (which would largely defeat the purpose of having speech recognition capability). So unless you want to engage in firmware hacking the MOVI (I do not, for this purpose), you might just have to eat the power draw.

Previous:
Next: