In August 2026, the U-GAIN Reading R&D Center hosted Speech Data for Learning Scientists 101, a two-day virtual workshop bringing together researchers, learning scientists, edtech leaders, and school district practitioners to explore emerging technical frontiers with graduate students and early-career scholars. While discussions surrounding artificial intelligence (AI) in early literacy often focus on automated decoding feedback, conversations throughout the workshop highlighted an important shift: rather than treating speech recognition solely as a tool to score reading accuracy, researchers are beginning to explore speech as a rich, multidimensional signal that offers deeper insights into student thinking, engagement, and comprehension.
Three key takeaways from the workshop illustrate how researchers and practitioners are rethinking the role of speech data in literacy research: (1) speech as a signal; (2) balancing skill building with student engagement; and (3) establishing a shared problem space for the field.
Traditional automatic speech recognition (ASR) applications in education focus heavily on verbatim transcription—comparing a child’s spoken audio against a text passage to generate a list of correct and incorrect words. However, when you hear children read, their speech can have a lot more useful information than a simple transcript can capture. Researchers are increasingly looking at new components of ASR, such as acoustic features, as windows into cognition and affect. Elements such as prosody (rhythm), pacing, hesitations, self-corrections, and expressiveness provide valuable context about a child’s reading process. For instance, a pause or a slower reading pace does not always signify a decoding failure; it can reflect thoughtful processing or deeper engagement with the text.
By analyzing these subtle speech patterns, systems can better distinguish between productive struggle and unproductive frustration, laying the groundwork for more responsive instructional support. This shift from a focus on speech recognition to using speech as a signal sets the groundwork for moving toward broader multimodal research.
A central theme throughout the workshop was the ongoing challenge of walking a “tightrope” between delivering targeted, evidence-based phonics and decoding support without making the reading experience overly corrective or mechanical.
If automated tools focus too narrowly on catching every vocal error, students risk becoming discouraged, which can diminish their motivation to read. Reading for pleasure and intrinsic motivation are well-established factors that support long-term literacy development.
To address this tension, researchers and practitioners are co-designing platforms that integrate foundational skill practice into engaging contexts, such as interactive readers’ theater and conversational AI tools. By prioritizing both instructional goals and student agency, the goal is to develop tools that build essential reading skills while preserving the joy of learning.
Rather than presenting finalized solutions, the Speech Data 101 convening was designed to outline a shared problem space for the broader educational ecosystem. Developing speech tools that function effectively in real classrooms requires addressing complex, interdisciplinary challenges that no single discipline or organization can solve alone. Key priorities for future research include:
Whether you are a graduate student exploring new research, a technologist designing speech models, or an educator integrating digital reading tools in your classroom, we invite you to connect with the U-GAIN Reading Center community.
