Dialogue Systems Lab using VoiceXML

Responsible teacher for this lab is Gabriel Skantze. If you need any help with this lab, please send an email to gabriel@speech.kth.se, or come by with your laptop to my office.

You will experiment with and extend an existing VoiceXML application. The application is a very simple travel booking system. The application is not connected to any database, so the reservation is never stored. You will use a commercial platform called Voxeo Evolution.

Reference

Preparing

Download the documents

Download booking.xml (the VoiceXML application) and grammar.xml (the grammar of what can be said) and place them on a web server, for example where you have your personal web site or in your public Dropbox folder. Check them and see if you understand their structure.

Register for Voxeo Evolution

Goto Voxeo Evolution and register for an account.

Create a new application (Development, Europe, VoiceXML, Premium ASR/TTS, LON - Prophecy 11 VoiceMXL). Let the Voice URL point to the web address where you put booking.xml.

When you have created the application, you can see how to call the application under Contact Methods. You can either call the landline number (starting with 08), or use a SIP client, such as Blink.

Try to call a few times and say different things. Note that initially, only trips between Paris and London can be performed. Try to fill one slot at a time or several slots at once. Check what can be said in the grammar (which is initially very limited). Examples:

I want to go from Paris to London
I want to go from Paris at three o'clock
I want to go to London on the first of August
I want to go from Paris to London on the third of July at four o'clock

If the voice keeps interrupting itself, it is probably picking its own voice or some noise. Try to use a headset and/or lower the volume of of the speaker and/or microphone.

Edit the documents

Use a text editor of your choice, preferably an XML-aware editor such as EditiX Lite, Eclipse or Microsoft Visual Studio . The files are linked to DTD files which can be used to check that the syntax is correct. If you use an XML editor that does syntax checking (i.e. that the document is both well-formed and validates to the DTD), you will avoid doing a lot of the most common errors, since the Voxeo platform is not very helpful when it comes to debugging.

Change the prompt from ”Welcome to ...” to ”Hello, and welcome to ...” in the vxml-file and call again to see that your changes take effect.

If everything works, you are ready for some exercises!

Exercises

Exercise 1 - error handling

If the system doesn’t understand, it just says ”Sorry, I don’t understand”. If the user is silent, it says “I’m sorry I can’t hear you”.

Add other prompts if the system does not understand the user. Use <nomatch>.
Add other prompts if the system does not understand several times. Use <nomatch count=””>.
Add other answers if the user is silent. Use <noinput>.

Exercise 2 – extending the grammar

The grammar is limited. The user can only travel between Paris and London. Time can only be stated in hours, like “five o’clock”.

Add more cities to the grammar. Call and try to use one of these.
Add the possibility to use phrases like “half past eight”. You don’t have to manage all such phrases, just try a few.
Add the possibility to say “on Christmas” as a date. The system should confirm with ”December the 24th”.

Exercise 3 – confirmation

After the booking is done, it is confirmed by the system, but the user has no chance of correcting it.

Add a confirmation dialogue after all fields are filled. The system could for example say:”You want to go from Paris to London on the 13th of May at three o’clock. Is that correct?” If the user says “yes”, the system should say something like “Thanks for your reservation”, otherwise it could say ”Ok, let’s try again” and start all over again. Tip: Make a last field in the form where the question is asked and clear the form if the user answers “no”. You can use the built-in grammar for yes/no-questions by using the attribute type=”boolean” in the field.

Exercise 4 – speech synthesis

The speech synthesis quality could possibly be improved.

Experiment by adding SSML tags to the prompts in order to improve (or just modify) the speech synthesis.

Report

Write a report about the lab. Include the final vxml and grammar files. Discuss the following issues:

What might be appropriate for the system to say when "nomatch" or "noinput" occurs?
What are the possibilities and limitations of VoiceXML as a standard when developing dialogue systems?
In what way could you (or could you not) improve the speech synthesis by adding SSML tags?

In your discussion, relate to the contents in the lectures and possibly the literature.

The report should be at least 2 pages (normal formatting), excluding the XML code.