Vaux AI technology demo
UX, UI Design
VAUX aims to create universal access to quality communication. With advanced voice technology applying extreme compression to voice audio, Vaux makes it small to store, fast to send while retaining its clarity. I was asked design an interactive and educational tool for Vaux to showcase the value of their technology by showcasing the mechanics of their deep learning models in audio and voice.
The advanced voice tech startup was at a critical stage of development and required a solution to showcase their AI-powered voice technology, in preparation for investment and commercialisation.
DISCOVER
Problem exploration
I conducted a deep dive session with the Vaux founder and two engineers – each with different responsibilities and expertise within the business.
We found communication was a blocker as my level of knowledge in the machine learning model was limited and their ability to simplify it was challenging. After a couple of sessions, I found drawing pictures to demonstrate what I had heard to gain their confirmation bridged our communication gap.
​
Drawing pictures and using methods like the 5 whys, helped me gain insights about how they train the deep learning models and the multiple layers of compression and denoising the audio files are processed. This provided context to the problem it was solving and the value it provided to various industries. This also allowed me to extract the language and methods the team used and simplify it for a non-technical audience.
Problem framing
The Vaux team are struggling to demonstrate how their audio compression technology works to potential investors and buyers, who lack high level technical knowledge. Users who have trialled iterations of the demo on the Vaux website, are left confused and can’t see potential or value in the technology. The team risk losing the support of their current investor and could fail to attract new investors and potential buyers.
Researching the demo
I conducted a teardown of the current demonstration to understand how it worked and the meaning of the information. The four dropdowns lacked labels and the implications if a user made a selection. There were also technical terms used within the drop-downs that only made sense to the engineers who were familiar
the data.
​
Choosing from each drop-down menu was like multiple choice but there was no feedback to know if a combination was correct or what happened when a user created a different combination. The audio samples only demonstrated the Vaux technology so it was unclear how it compared to other compression standards like mp3
or Opus. This meant there was no benchmark for a user to know if the results were good or bad.
​
Previous demonstration tests were unsuccessful with users so, I began my research by speaking with these people. I conducted interviews with the current investors, potential investors who were involved and other companies in the accelerator.
Original demo
affinity mapping was used to organise the data from user interviews. this technique identified themes and user pain points.
DEVELOP
Building an audio recipe
I broke down the information, pulled from the current demonstration, and grouped them into categories. I considered the comments from users needing clear direction to complete their task, in addition to feedback from the UI to indicate when a sample was available to listen to or a new selection had been made.
With this in mind, I created steps where the user could create their own audio “recipe”. They could choose their voice, add background noise and hear samples of those sounds along the way.
DELIVER
Testing ideas
Ideas for the UI were sketeched as well as wireframes to pitch to the team before moving onto more detailed mockups. The current Vaux website was very dark with an almost black background so, I was limited to the existing colours and layout when building my designs.
​
We had one day to finalise development and test with the same group of users who were familiar with the original demo and a new user group who had no previous context.
​
Users also needed context to be able to judge whether the Vaux audio was better or worse, in comparison to other compression standards. Therefore, once a user sent their “recipe” through the Vaux generator, the final sample would appear with a selection of other codecs to compare the difference. The audio bars would also animate to demonstrate the bitrate speed, due to the small size.
Iteration 1 - users found the steps unclear and needed more guidance.
Iteration 2 - user feedback found the numbers were still unclear.
Iteration 3 - vertical navigation was added with auto scrolling.
Choose a character
Choose the voice of your character - either male or female
Add background noise
Add music, machinery or traffic background noise to the audio and experience how difficult it is to hear the character speaking
Apply Vaux technology
Send the audio sample through the Vaux generator and listen to the high quality results in comparison to other compression examples
Choose your own adventure
The final outcome was a three step “choose your own adventure” experience , allowing Vaux to bridge the gap between complex technology and the audience they needed to communicate with.
Another round of testing and users reported the demonstration provided a simple breakdown of the information and helped them understand the classification between the audio samples Vaux produced, compared to other codecs. They also found the animated result rewarding and wanted to try different combinations to experience different outcomes.