designer, creative technologist, tinkerer

Talk to Me


a project made by:

mariana tamashiro, sergio rivera, wei miao


“Talk to Me” is an interactive art installation allows a whimsical and eerie experience to talk to inanimate objects. Using open source artificial intelligence connected to an old typewriter and analog phone, answers to questions spoken into a handset are typewritten by an invisible typist. The team became intrigued with the idea of blending AI with mechanical and analog communication technology. The system is activated when the receiver is picked up. Spoken questions are then directed to the open-source AI voice assistant Mycroft via a Raspberry Pi, and responses are conveyed to the “ghost writer”.

(this was a project for the class Design Studio - ATLS5519-002, taught by Joel Swanson at the University of Colorado Boulder)


When brainstorming about our concept, we had the desire to combine our interests, learning goals, and skills in a cohesive idea. Therefore, when the idea of “a typewriter that types itself” really checked the majority of our requirements: it was a magical/whimsical interaction; it would challenge us technically; it was a physical experience, and had the potential to be something impressive. Also, when we thought about the concept in a deeper way, we got fascinated by the idea of combining old and new technology in an unexpected way. It made us reflect about how did we use to interact with objects, how do we interact with them now, and how we could change with an unusual twist.


After having the idea, we structured based on a guideline inspired by Meow Wolf’s Experience Arc (hook, exploration, reward) [1]

  • Hook: when someones passes by our desk, the lamp will flicker

  • Exploration: audience will read some of the prompts that are spread out on the table and will pick up the phone and try saying something.

  • Reward: the typewriter will “magically” type the answers to the audience questions.

Screen Shot 2019-05-02 at 5.56.14 PM.png


Since we wanted the project to express this feeling of “talking to objects”,  we chose to work with already made objects instead of creating all of them from scratch (typewriter, phone, lamp). We found the objects in different places, like flea markets, online stores and online stores for crafts and second handed products. There was also a learning curve to understand how some of those objects, specially the typewriter works.

Below, there are the typewriter, phone and lamp in the process of being hacked.


After we put the microphone connected to the Raspberry Pi and the Open Source AI (Mycroft), we needed to test to observe the performance of the speech recognition system. This is our usability test protocol and the goal was to test not only the concept of the project but specially the reliability of the speech recognition. We had very valuable insights, such as: the system tended to work better with male participants; people would get disappointed if the system did not recognize their voices; the tension of the fishlines to pull the keyboard keys needed delicate adjustments.


To make the typewriter type itself, it required a very iterative process of different technologies, systems and mechanisms. The first started with solenoids which we believed would be strong enough to pull the keys. However, when add together, they require a reasonable amount of space and energy — considering we would use one solenoid per key. Another issue was the budget because it can get expensive quickly when considering a large amount of solenoids. Therefore, we decided to go to another path, that would only require 2 motors, but more complex circuitry and mechanical accuracy. After a few version, we finally ended up with a working prototype that could type based on the messages the speech recognition system was receiving.

Below, are the pictures from the first prototype with the solenoids (left), the 3D model of the second mechanism idea (middle), and the final version (right)