Cognable provides a simple to use SOAP service which quickly creates mp3 audio versions of alphanumeric strings. Various parameters can be adjusted to change the style, security and content of the audio.
The service grew out of a client requirement for a Captcha access solution for visually impaired screenreader users, it is not an approach which is particularly useful to ID audiences - nor is it anything more than a partial solution to the many accessibility issues created by Captcha.
Code examples for PHP are provided, but anything which consumes WSDL should work fine.
This free audio service is intended for low volume commercial or personal sites only. For the sake of efficiency the requests made via SOAP are cached and the mp3 is created only when its url is requested - creation itself takes only a fraction of a second. Please don't attempt to cache the audio on your server, it won't be any quicker for the end user but will use unecessary resources here.
Marginal security can be achieved by varying the mode and phrasing of the Captcha audio. By default the audio is clear and therefore very easy for speech recognition applications to interpret. Further obfuscation options are included to make the audio more challenging for programmatic agents. Obfuscation is achieved mostly by using delayed feedback effects which are varied regularly, as are all the component audio files. In basic testing with desktop recognition software I've found delayed feedback to be a more successful mask than adding pops, tones and noise since all these are really easy to filter out. It pretty much goes without saying that using audio is unlikely to be as secure as a good graphical Captcha if speech is kept clear enough to be widely useful.