This paper describes Project Spectra, a collective of open source developers that aims to build free and open source voice training technology for transgender people. We demonstrate how a design prioritizing the agency of trans users was made possible through sustained community collaboration. Using an autoethnographic approach, we discuss our community-based design process, which was documented with memos, online meetings and text conversations, sketches, and other data sources. We illustrate how we articulated our values as a group: deciding our programming framework (including a Statement of Principles), elaborating our "Experience Goals" (the feelings we wanted our design to elicit), and determining the features we wanted to implement in our app. We conclude with a reflection on the benefits and challenges of conducting community-based design research through an open-source organizational model.

Increasing attention is being paid to applications of HCI that center on identity and social justice. Much of this work is influenced by Feminist HCI, a set of design orientations that critique dominant research paradigms and support the development of systems informed by feminist theory. However, the field has largely ignored the lived experiences of transgender people. I review the sociopolitical context of this issue, focusing on feminism's exclusion of trans people and critical health disparities in our population. I then present formative qualitative work in the domain of voice training for trans individuals. Interview participants expressed frustration with prescriptive gender norms and revealed opportunities for a personalized health technology that counteracts these norms. Specically, it could adapt to their goals and provide actionable feedback without judgment or reference to gender stereotypes. From these insights, I propose ‘trans competent interaction design’ as a starting point for trans-inclusive design practices.

It has been suggested that computer-assisted instruction (CAI) is a promising method for educating students on the autism spectrum. We aimed to determine whether automated recognition of facial expressions aided in predicting CAI engagement and learning performance. Seven youth with autism (mean age = 12.7, SD = 4.2) interacted with a CAI program, TeachTown Basics, for 15 consecutive sessions. Video recordings of the participants' faces were collected during these sessions and facial expressions from these videos were analyzed using CERT, an algorithm that automatically outputs intensity values for each facial action unit (AU). Using these data, we attempted to operationally define two engagement indices: (1) behavioral engagement, the proportion of time a participant had their face oriented to the computer screen; and (2) emotional engagement, the activation of AUs previously associated with CAI. Our results suggest that both indices strongly correlated with one another, but that emotional (not behavioral) engagement predicted test performance. CAI knowledge domain, participant sex, and developmental age also contributed to the prediction.