Wearable Voice Assistant

For the first wearable project, I want to examine my struggle with self-censorship. I imagine it stems from a culmination of conservative culture and upbringing, lessons and rules of etiquette my parents imparted on me as a child that continues to prevent me from vocalizing my thoughts, opinions and desires well into adulthood. The saying goes, “If you have nothing nice to say, don’t say anything at all.” This is a rule I generally try to follow, but there are always instances when I’m tried beyond the limits of my patience. Yet, even at my limit, I still find it quite challenging to voice negativity without concern that it comes with repercussions. Therefore, I wanted to explore mediums of expression through visualization and/or layers of encryption.

My original sketch was of an overcomplicated headgear that transmits thoughts from the brain (visualized through Neopixel strips) that travel to the voicebox (wherein a microphone is mounted, symbolic of the lumps that form in our throats in the act of self-censorship) and translates onto an LCD screen mounted over the mouth to both encrypt and visualize the words I want to express.

In essence, it is a very cowardly project as I frantically attempt to cover all grounds to avoid facing my own vulnerabilities and eliminate the possibility of offending others. In the weeks leading up to presentation, I talked to a number of people — Jingwen, Vidia, Elizabeth, Nicolas, Aaron Montoya-Moraga, and Alejandro about the concept to resolve the glaring holes and discomfort of putting this whole thing together. Nicolas was particularly insightful by suggesting I look into Lyrebird, a voice-imitation algorithm capable of generating one’s digital voice by recording a number of audio samples for a minimum of 1 minute. This suggestion opened up a new door of possibilities for me because it allows for self-expression, but shifts the blame onto an alias/machine…


The interesting thing about Lyrebird is it states that the more audio samples you record, the better the quality and likeness to your voice it becomes. This was not the case for me. I was surprised to find that my digital voice sounded more human at about 109 audio samples versus 502, when it sounded downright robotic. My voice has a slight raspiness due to smoking, which is more evident when I’m either just waking up or extremely exhausted. That particular nuance disappeared the more recordings I made. When I tried to test my digital voice to say, “I love you.” the output was muffled and unclear, so I began deleting recordings until those words were discernible again.

Given a tool like this, the obvious thing to do is to get your voice to say naughty things. (I do curse every now and again for emphasis, out of frustration, or when I’m late for an appointment, but I tend to keep my speech clean for the most part.) I began plugging in filthy gangster rap lyrics and it was hilarious. I decided this direction would make for a better interaction, so I continued to develop its potential for humor instead.

To fabricate this wearable, I repurposed an old cardigan that I never wear into a turtleneck crop top. As for my hardware, I used Adafruit’s Audio FX Sound Board, JST connector and LiPo battery, perf board and female headers (so the soundboard can be easily removed and reprogrammed), and a thin plastic speaker. I specifically wanted to use a knit fabric to insulate my soft circuits of conductive thread and conductive fabric buttons to trigger the recordings I edited on Audacity. The fold over of the oversized turtleneck will cover the hardware so everything is hidden in plain sight and virtually indistinguishable from any regular garment. So much of wearable tech prefers to put the technology on display, but I don’t believe the general masses are prepared to embrace that aesthetic quite yet.


I created three scenarios for this wearable to play out and demonstrate its versatility:

1. Introduction to my Voice Assistant
2. Catcalling
3. ER Shift

[Video documentation]

On the day of presentation, I tested it repeatedly and everything worked up until I got in front of the class.  The wearable ultimately failed for the following reasons:

1. My placement of the microcontroller in the back neck was a strategic decision to hide the hardware, but it also made it inaccessible for me to connect the LiPo battery to power it up while wearing it.

2. Although I calculated some ease for the conductive thread circuit, it was not enough to accommodate enough stretch for my head as I was putting it on, which ended up compromising the circuit.

Lessons learned:

1. I could have easily replaced conductive thread with thin, silicone-coated wires instead for purposes of reliability. I have had conductive thread fail me so many times I don’t know why I thought this time would be any different.
2. Always remember to place the microcontroller some place accessible or integrate an accessible on/off switch.

Overall, I’m slightly disappointed that the wearable failed during my presentation, but I’m happy the interaction was successful given the class erupted in laughter!

Special thanks to Jingwen Zhu, Vidia Anindhita, Elizabeth Ferguson, Nicolás Peña-Escarpentier, Aaron Montoya-Moraga, and Alejandro Matamala Ortiz for talking through this project with me and giving me valuable feedback!

Leave a Reply

Your email address will not be published. Required fields are marked *