Considering Avante, TTS, and OCR
by Noel Runyan

February 2005

Several folks have asked if the Avante DRE, with paper ballot printer and optical character recognition (OCR) scanner, would be a satisfactory solution for both accessible and secure voting. Unfortunately, the Avante DRE uses a synthetic TTS (Text To Speech) system, the IBM Via Voice, that is difficult for most older folks to understand. Use of the poor quality synthetic voice was the primary reason that the Avante scored at the bottom in AFB's Access World review.

Text to Speech (TTS)
Many blind computer users work with the IBM Via Voice TTS speech and can understand it well enough. I have the Via Voice on my own computers, but I prefer to use higher quality TTS voices instead.

Many visually impaired computer users don't appreciate the two main limitations that synthetic speech represents to older voters. One is that the loss of hearing experienced by many of the elderly makes it much harder to understand synthetic speech systems. The other limitation stems from the fact that the synthetic speech systems require learning curves of days or even weeks to reach good understandability. This is especially true for the lower quality synthetic voices like the Via Voice.

Before IBM starts jumping up and down and defending their speech quality, let me say that there are much lower quality synthetic voice systems available on the market. Sadly, Avante did not use one of the available higher quality voices.

Learning to understand synthetic speech systems for your first time is somewhat like trying to learn to understand someone with a very thick foreign accent. Eventually you may be able to understand them, but probably not at first. Trying to learn to use your first strange E-voting machine while also listening to instructions in the same strange voice is especially hard.

Another problem with synthetic voice systems in the DRE voting application is that the ballots are filled with personal names. Personal names are very often mispronounced by synthetic speech systems. This can cause a lot of confusion. Visually impaired folks with their own accessible talking computers usually have the ability to stop and spell out any word that is hard to understand. DREs with synthetic voices could be designed to allow spelling verification of words, but this starts to complicate the button controls and substantially increases the voter's frustration, as well as learning and training requirements.

Optical Character Regognition (OCR)
Avante and other DRE manufacturers have proposed using OCR ballot scanners with the DRE systems, to permit scanning and verification of the printing recorded on printed ballots.

An argument in favor of using OCR ballot scanning for verification of the printed ballot is that DREs might be programmed to cheat on the printed ballots of folks who are using the access options, assuming that these voters would not be as likely to detect the fraud without some way to read the actual print on the ballot.

If OCR is going to be used to verify the printed ballots for visually impaired voters, it could be argued that it should be done by a separate system, not the same DRE unit that might have misrepresented the results on the ballot it just finished printing. This is because it could be possible for perverse portions of a combined vote printing and scanning unit to make the same false or misleading representation of the ballot in the voter's pre-printing as well as post-printing verification reviews.

Background of Noel Runyan:

With his degree in Electrical Engineering and Computer Science, Noel Runyan has been working in human-factors engineering for over 35 years, primarily developing access technologies for helping persons with visual impairments use computers and other electronic devices. During the 5 years he worked for IBM, he was involved in the design and testing of the security systems for both BART ticket machines and ATM credit card systems.

After starting his own company to supply access technologies, he designed and manufactured the Audapter speech synthesizer, to enable computers to talk to visually impaired users. Noel also authored the EasyScan, BuckScan, and PicTac programs that made it easier for visually impaired users to read print books, identify dollar bills, and convert print pictures into raised line tactile drawings.

To help their visually impaired customers access and make use of computer systems, Noel and his wife, Deborah, have personally built hundreds of custom-integrated personal computers with speech, braille, and large print interfaces. More recently, he has been involved in the development of talking internet radios and talking pill bottles and other medical equipment for persons who have difficulties reading print labels and displays.