Technical Reports A Text-to-Speech Converter for Radiology Journal Articles Michael L. Richardson, MD Radiology articles are primarily designed to be read on paper or a screen. Audio versions let users hear this material during activities when reading is not practical. Currently, there are relatively few radiology materials in audio format. However, inexpensive text-to-speech software can easily produce spoken-word versions of digital text. This paper describes a free Web-based program that converts radiology articles to audio format using text-to-speech software. Key Words: Text-to-speech; podcast; audio journal; software; Ruby. ªAUR, 2010 R eading journal articles remains one of the main ways many of us keep up with the latest literature, but it can be difficult to find dedicated time for this. Portable audio players such as the iPod offer one solution. These players make it easy to download and listen to audio files for ‘‘inter- stitial listening’’ during activities that preclude reading, such as driving, exercising, or performing repetitive chores. Currently, there are relatively few radiology educational materials available online in audio format. The editor of Radiology has a monthly podcast (http://radiology.rsna.org/site/podcasts/ archive.xhtml), and faculty members from the Medical Univer- sity of South Carolina (http://itunes.apple.com/us/podcast/ musc-radiology-podcast/id253693973), the Johns Hopkins School of Medicine (http://itunes.apple.com/us/podcast/ ctisus-computed-tomography/id75022309), the Massachusetts General Hospital (http://itunes.apple.com/us/podcast/mgh- radiology-rounds/id216892347), the University of Washington (http://uwmsk.org/podcasts/Site/Podcasts/Archive.html), and Seattle Children’s Hospital (http://web.mac.com/mthapa/Site/ Podcast/Podcast.html) have posted radiology podcasts online. CMEonly.com, a commercial vendor, offers audio reviews of the radiology literature for $349 for 12 issues. Fortunately, the text of many radiology journals is widely available online in digital form, creating a large pool of potential material for an ‘‘interstitial listener.’’ Text-to-speech (TTS) software has been widely available for several years and is currently built into many modern operating systems, such as Apple’s OS X. Such system software makes it easy to read aloud occasional selections but is tedious to use with large quantities of text. Independently developed software, such as TextCast (http://www.bitmaki.com/ textcast/) for the Macintosh, allow easy and efficient whole- sale conversion of Web pages into audible podcasts. Alas, most published Web pages are not optimized for reading aloud. For example, it is quite tedious to listen to a computer reading visual typographic devices, such as tables of content, headers, and footers. In addition, it is a bit jarring to hear TTS software pronouncing ‘‘CT,’’ ‘‘MR,’’ and ‘‘MD’’ as ‘‘Connecticut,’’ ‘‘mister,’’ and ‘‘Maryland.’’ Thus, some form of text filtering is desirable before TTS conversion. Free text-filtering software is currently available for many Web (http://lab.arc90.com/experiments/readability/) and YouTube pages (http://quietube.com). Unfortunately, these two programs may not work very well for journals with special formatting, such as some radiology journals. Therefore, I decided to write my own Web-based TTS program, which is designed specifically to download and filter the text from radiology articles and convert it to an audio file. I wrote this program using HTML, Apache 1.3.41 (an open- source Web server; http://apache.org), Ruby version 1.8.7 (an open-source programming language; http://www.ruby- lang.org/en/), Cepstral Swift version 4.0.3, a proprietary TTS engine sold by Cepstral (http://cepstral.com), and an open- source software module called LAME version 3.98.2 (http:// lame.sourceforge.net). A high-end server is not necessary for this application. My program is currently running on an older Macintosh OS X desktop computer (dual 1.8-GHz PowerPC G5) located in my office. However, all of the program modules used in this project can also be installed on other Macintosh, Windows, or Linux machines as well. The demonstration code included here is designed to work with the American Journal of Roentgenology (AJR) Web site. Here’s how it works. First, find an article on the AJR Web site (http://www.ajronline.org) you would like to hear. Note the volume number, issue, and page number of the desired article. My conversion program (http://uwmsk.org/ ajr.html) will then ask you to enter these three numbers into appropriate fields on a Web-based form (Fig 1). The hypertext markup language (HTML) code is shown in Appendix A. Acad Radiol 2010; 17:1570–1579 From the Department of Radiology, University of Washington, 4245 Roosevelt Way NE, Seattle, WA 98105. Received April 12, 2010; accepted June 30, 2010. Address correspondence to: M.L.R. e-mail: mrich@uw.edu ªAUR, 2010 doi:10.1016/j.acra.2010.06.020 1570