viebrock.ca

My article about Turing Protection generated lots of comments about how using image CAPTCHAs restricts access to the visually impaired.

So, I’ve played around a bit, and added an audio component. If you can’t read the CAPTCHA image, you can listen to a .WAV file of our lovely server, spelling out the characters to you.

Again, the code is very straightforward. You just need one new file to generate the file.

You will also need to install some text-to-speech program. The program will need to accept a string to read, and output a sound file, ideally sent to stdout. I searched around for a bit, and flite seemed to mostly fit the bill. It can only output to a file, so we use PHP to generate, stream, then cleanup temporary files.

I also made a slight change to the protection script. Since the sound file doesn’t distinguish between upper and lowercase, and the image is either in all uppercase or all lowercase, I just made the checking routine case-insensitive.

To see it in action, try clicking here. Again, please send your comments and suggestions.

Update: 17-May-2004

Several folks have noted that some of the letters aren’t pronounced very well. Specifically “Q”. This is how flite is generating the sounds, so I suspect that with a small and cheap speech synthesizer, you get what you pay for, so to speak. (“So to speak” ... hah!)

If anyone can suggest a different speech engine, I’ll try that instead. Hopefully, there is a Debian package for it though, since the server I’m using is configured pretty much entirely through packages.

For the record, if your challenge string is “abc123fq”, then the command that gets passed to flite is basically:

/usr/bin/flite -t "a, b, c, 1, 2, 3, f, q."

The commas add a bit of a pause between letters. But it relies on how flite decides to pronounce “q”. I’ve tried spelling out the letters:

/usr/bin/flite -t "aye, bee, see, one, two, eff, kyoo."

That didn’t (to me) produce a better sound than before though. Plus, how do you spell “q”? Kyoo? Keeyou? Queue?