Talking Signs

The core interaction is deceptively simple: a transmitter at a fixed location continuously broadcasts a digitally recorded spoken message (e.g., 'Powell Street BART station — escalator to trains') modulated onto a 25 kHz FM subcarrier on infrared light at 850–950 nm wavelength. A blind user carries a pocket-sized receiver (4 × 2 × 1 inches) with a photodiode sensor behind a directional aperture. Scanning the environment by pointing the receiver in different directions, the user hears spoken messages when the aperture is aligned with a transmitter. The signal provides natural directional feedback: clarity and loudness increase as pointing accuracy improves and distance decreases; pointing away causes the signal to become staticky and eventually squelch. Multiple overlapping signals are resolved by FM capture effect — the stronger signal (closer or more directly pointed) captures the receiver, requiring approximately a 20 dB power ratio. Because infrared is line-of-sight and blocked by walls, it naturally matches the behavior of visual signs — you can't 'hear' a sign around a corner any more than you could see one. The transmitter broadcasts continuously, meaning the user chooses when to access the information (a deliberate ADA-conscious design choice to keep the system 'refusable'). The intelligence is entirely in the receiver; the transmitter is a 'dumb' repeater. This split — smart receiver, dumb infrastructure — presages the smartphone-as-sensor paradigm by decades.

The Fall 1991 issue of the Smith-Kettlewell Technical File (SKTF), edited by Bill Gerrey and still available in full online, is one of the richest primary source documents in the museum's research archive. Across dozens of pages, it provides complete schematics for both transmitters and receivers, a bill of materials with prices, beam-dispersion physics calculations, installation geometry guidelines, human-factors deployment philosophy, and circuit-level explanations. The transmitter used a 512K-bit EPROM for ~8 seconds of sampled speech at 64 Kbps, modulating an FM subcarrier with a CD4046 phase-locked loop. The receiver used six Siemens BPW34 photodiodes in parallel behind a Kodak Wratten Filter No. 87C (passing 80% IR, blocking 99.5% visible light), with an LM311 comparator providing squelch with RC hang time. The document addresses practical deployment concerns: how to align transmitters so beams don't bleed between floors, how to handle direct sunlight (which contains IR), how to design street-sign arrays with 48 dual-LEDs for crosswalk coverage. The document even includes a troubleshooting section and a parts-ordering guide with 1991 pricing: $0.60 per LED in small quantities, $150 for a custom-message transmitter, $250 for a receiver. Gerhard Sollner of the SKTF later wrote a supplementary issue with refinements.

Talking Signs is one of the most rigorously evaluated accessibility interventions in HCI history. Bill Crandall and B.L. Bentzen at SKERI, working with UCSB geographers James Marston and Reginald Golledge, conducted multi-year field studies funded by the Federal Transit Administration and PATH/UCTC. At San Francisco's Powell Street Station, 35 of 36 blind participants successfully completed easy routes independently, with 24 completing medium and hard routes — including participants with only written instructions (no hands-on training). At street intersections, the improvement was dramatic: participants using Talking Signs began crossing during the Walk phase 99% of the time (vs. 66% without), started from within the crosswalk 97% of the time (vs. 70%), and needed help finding the crosswalk only 1% of the time (vs. 19%). One participant remarked that in Powell Station he was 'truly equal' to sighted travelers. The formal Steering Committee recommended RIAS as the 'preferred technology enabling print-handicapped persons to travel independently in transit facilities.' These findings directly informed the US Access Board's adoption of RIAS into building code ADAAG 703.7, which specifies the modulation frequency (25 kHz), wavelength range (850-950 nm), and optical power density (26 picowatts/mm² at receiver aperture) as federal requirements.

Despite the proven effectiveness, Talking Signs never achieved universal deployment. SKERI's own assessment is candid: 'The question of who would pay for large-scale transmitter installation and receiver distribution proved insurmountable, and Talking Signs were never adopted broadly enough to make them a universal accommodation.' The capital cost of installing and maintaining hundreds of physical transmitters across a city — each requiring power, weatherproofing, and periodic re-recording — was a barrier that neither transit agencies nor city governments were willing to fund at scale. The 1995 Powell Street installation (93 transmitters, $300/sign for interior deployments) was funded by an FTA demonstration grant, not an operational budget. The receivers, at $250 each, also required distribution and maintenance. This infrastructure problem — brilliant interaction design undone by deployment economics — is a recurring theme in ambient and ubiquitous computing, and Talking Signs is one of its earliest and most instructive case studies.

In 2016, Smith-Kettlewell scientist Joshua Miele — who is himself blind — released 'overTHERE,' a free iOS app that translates the Talking Signs interaction model into a software-only experience. Instead of physical infrared transmitters, overTHERE uses Google Places and OpenStreetMap data as 'virtual transmitters' and the iPhone's compass and GPS as a 'virtual receiver.' The user holds the phone flat and sweeps it around their body; as the phone points toward a nearby destination, that destination is announced through headphones or the speaker, with audio spatialization providing the same directional feedback as the original hardware. The app supports 20+ languages and is used by thousands of blind travelers daily worldwide — solving the infrastructure problem by piggybacking on a device users already carry. Miele developed overTHERE at a Google MakeAthon in 2015, where it won the Improved Mobility prize. The project page at SKERI notes that funding is being sought for a new version as of 2025. This 40-year arc — from modulated IR beams to smartphone compasses, from dedicated hardware to ubiquitous platforms — makes Talking Signs one of the longest continuously researched interaction models in HCI.