Talkabel: A Labeling Method for 3D Printed Models
Lei Shi · 2015 · ASSETS '15: Proceedings of the 17th International ACM SIGACCESS Conference on Computers & Accessibility · doi:10.1145/2700648.2811327
Summary
This poster paper presents Talkabel, a low-cost labeling system that makes 3D printed models accessible to blind users by enabling them to access spoken labels through simple taps on tactile markers. Three-dimensional models are valuable educational tools for blind learners, particularly for complex concepts like molecular structures or astronomical scales that benefit from tactile exploration. However, unlike tactile graphics which have established labeling conventions, 3D printed models lack standardized accessible labeling methods. Existing approaches have significant limitations: Braille labels are bulky and readable by less than 40% of functionally blind people in the United States; QR codes work on flat tactile graphics but are difficult to implement on complex 3D surfaces; and embedded electronics with sensors and circuits substantially increase model cost. Talkabel addresses these challenges using commodity hardware—a standard smartphone—and simple auxiliary components. The system consists of tactile markers (small semi-spheres that indicate label locations and provide tactile targets for tapping) and a 3D-printed scaffold that holds the phone in a fixed position relative to the model. When a user taps a marker, the phone's microphone captures the acoustic signal, an algorithm classifies which marker was tapped based on the sound's characteristics, and the phone speaks the corresponding label. To create labeled models, makers simply attach the auxiliary components, then tap each marker while recording its label through the app. The classification algorithm uses a two-phase approach: first detecting taps using Root Mean Square energy thresholding on the audio stream, then classifying the tap using a nearest neighbor classifier with Amplitude Spectrum Density features computed via Fast Fourier Transform. The system restricts analysis to the 0-4800 Hz frequency range, as higher frequencies degraded classification accuracy.
Key findings
Feasibility evaluation with four 3D-printed molecule models demonstrated promising results. With only 5 training samples per marker, Talkabel achieved over 83% classification accuracy across all models. Performance stabilized at 93% accuracy with 25 training samples per marker. The system performed consistently across different marker layouts—models with equidistant markers, varying distances from the microphone, and shared support structures all achieved similar accuracy levels. Model D, which had four markers (instead of three) with smaller inter-marker distances, showed slightly lower accuracy, suggesting that marker density and spacing affect classification performance. The acoustic-based approach offers several practical advantages over alternatives. Unlike embedded electronics, Talkabel requires no special components beyond what can be produced with a consumer-grade 3D printer. Unlike Braille, it does not require literacy in a specialized system. Unlike QR codes, it works naturally on complex three-dimensional surfaces. The training process is quick enough that makers can add labels to models in a reasonable amount of time, and the system uses hardware (smartphones) that blind users already commonly use with accessibility features.
Relevance
Talkabel addresses an important gap in accessible educational materials as 3D printing becomes increasingly common in schools and makerspaces. The approach of using acoustic sensing to distinguish touch locations on physical objects has broader applications beyond 3D models—similar techniques could potentially label maps, museum exhibits, or other tactile educational materials. For accessibility practitioners, this work demonstrates that sophisticated sensing can be achieved with commodity hardware, lowering the barrier to creating accessible physical materials. The finding that only 5-25 training samples are needed per marker means the system is practical for teachers or volunteers without technical expertise. The reliance on smartphone microphones and text-to-speech also means the system integrates naturally with assistive technology infrastructure that blind users already depend on.
Tags: blindness · 3D printing · tactile graphics · acoustic sensing · machine learning · labeling · educational technology · STEM education