Volume 4, Issue 1 - January/February 2004
 
   
 


VoiceXML Helps to Make Holiday Magic

By

We are experiencing the arrival of voice technologies in everyday life. To those in the IT community, this is a natural extension of the technological innovations that have preceded voice. For many in business and industry, this is a new technical “gimmick” that, while interesting, has yet to prove its worth in terms of ROI and the bottom line. For the public at large, awareness is limited at best and interest could realistically be described as wary. Elix, a company that specializes in multimedia contact center and business solutions, wanted to showcase its expertise in deploying a speech and Web application using the new VoiceXML standards and to introduce the public to these new technologies in an amusing and novel manner. Elix decided to use the opportunity afforded by the holiday season to use the traditional Christmas card as the vehicle for this ambitious idea. Elix invited two other companies to participate in what would come to be known as the Santa project: Nü Echo, an Elix partner, specializes in the design and development of speech recognition applications; CONCEPT S2i is a Web design and interactive scripting specialist.

The “Talk to Santa” Project
Elix’s idea was to send electronic Christmas cards that would invite the recipient to register their children and any “child at heart” types they knew for a personalized chat with the man himself—Santa Claus—and his elf, Pixel. All they had to do was visit the Santa Web site to register a child of any age. At the site, they were asked to fill in a questionnaire about the child and to provide some basic information: age and gender of the child, whether the child had already written to Santa, an idea of a gift the child might like to receive from Santa, and, most importantly, advice that Santa should give to the child at the appropriate moment in the interaction (for instance, “You should work harder on your homework and you really have to stop fighting with your brother Michael”). Parents or legal guardians could, if they chose, give permission to have a portion of the call recorded—the part where the child told Santa what gift they would like to receive from him; this audio recording would then be emailed to the parent after the conversation. Following the registration, the child received an email from the North Pole telling them that Santa would like to have a chat with them. The email included the toll-free telephone number to call and a “secret” code to enter. After calling the number and entering the code, the child first talked with Pixel, Santa’s automated elf, and then with Santa.

Creating the Magic
Specialists from Elix, Nü Echo, and CONCEPT S2i, worked furiously to design, develop, and implement the project in time for the Holiday season. The Santa project ran on Elix’s IVS, an interactive voice response platform that supports the VoiceXML and CCXML standards and uses voice technologies like speech recognition and text-to-speech (TTS). Elix managed the project and supplied its IVS voice platform with its VXMLBrowser, which integrates voice and data services using the common client-server paradigm and promotes service portability across implementation platforms. Elix also supplied the telephone and Web infrastructures required for the project and actively participated in implementing, integrating, and testing the application.

Nü Echo designed and implemented the voice application. A key design goal was to make children believe that they were really having a conversation with Santa. This was a challenge since TTS needed to be used to speak the gift idea and the advice that had been entered by the parents on the Web site. Since there was no way a TTS engine could make a convincing Santa, the approach was to use two different characters: Santa, whose voice segments would all be pre-recorded by a voice talent, and his elf Pixel, played by the TTS engine. Santa would handle most of the dialog but, at the appropriate time, would invite Pixel to speak, for instance by saying: “So Pixel, can you think of a gift idea for our young friend?” This all sounded completely natural to the children, who mostly didn’t even notice that the advice and gift idea weren’t actually spoken by Santa. The only negative feedback was from a few people complaining that Pixel’s voice was somewhat “monotonous”.

Another issue was the recognition of children’s speech. Indeed, it is well known that speech recognition engines usually have a very difficult time dealing with children’s voices, especially the younger ones. Although we made sure that the recognition grammars and pronunciation dictionaries were well tuned, we nonetheless had to deal with possibly high error rates in some instances. In order to address this issue, we designed the application so that, whenever recognition problems were detected, the dialog would be able to quickly recover and move on, possibly disregarding the child’s input in the process. This strategy proved quite effective in practice.

The application was developed and deployed using Nü Echo’s VoiceXML application framework, which produces dialog documents in VoiceXML running on Elix’s IVS platform. Although the framework is specifically designed to be multilingual, the support of both English and French necessarily entails additional development efforts. For instance, all recognition grammars had to be developed in both languages, requiring us to conduct coverage and accuracy tuning with children from both language groups. In addition, it was important to us that the sentences spoken by the TTS engine sound as natural and intelligible as possible to the children. In particular, since gift ideas were likely to include all kinds of toys, we wanted to make sure that the most popular toy names would be correctly pronounced. We therefore manually tuned the pronunciation of hundreds of toys listed in recent catalogues. One specific challenge was to get a French TTS engine to correctly pronounce English toy names (such as “Game Boy”). This requires the ability to approximate an English pronunciation with a French TTS engine, a process that we have fortunately mostly automated by using translation rules.

CONCEPT S2i created both the Web site used to personalize the dialogue for each caller and the module used to email management and follow up. All aspects of the project had to be tightly integrated: the Web site design, the creation of dialogues, and the integration of the voice application with both the Web application and Elix’s IVS platform. Visitors to the site could register a child, configure the conversation with Santa by entering the requested information, and immediately call the system to try it for themselves. They then had the option to go back and make changes to the information provided before giving their final approval. Impressively, the design, development, translation, testing, and execution were all accomplished within one month!

Overwhelming Response Rate
Almost 2000 electronic greeting cards inviting people to chat with Santa were emailed. The response was almost overwhelming. People responded from Canada, the US, and Europe. The project was in operation from December 8th to the 24th. During that time, the Web site was visited by over 130,000 people, and nearly 72,000 registered for a chat with Santa. At peak times, the site was handling 20 registrations each minute. In Québec alone 10.8% of all children between the ages of 3 and 9 participated, making this a remarkable marketing success!

The follow-up survey revealed that 96% of the users were very satisfied or satisfied. Many parents, when responding to the survey, said that the interaction with Santa was so realistic that children, especially the younger ones, really believed they were speaking to Santa. Some older children, who had doubts or had stopped believing, began to believe again.

The survey also revealed that although 92% of the site visitors had no knowledge of VoiceXML before their visit, after viewing the site, a full 24% thought it would be useful in their business or on a daily basis. It should also be noted that nearly 16,000 people clicked through to the Concept portion of the site, where the technology was explained and the idea of telephony-Web convergence was outlined.

The Santa project provided a non-threatening way for the public to experience the benefits and ease of using a voice platform for communication. As well, it was a concrete example of how businesses can use these new technologies to handle extreme peak periods without having to increase their staffing levels. After all, if Santa Claus were real, he would be the poster child for handling an intense seasonal peak period! Furthermore, the project showed companies new ways to leverage voice and Web technologies by showing them how they can voice-enable their Web sites, how they can approach their markets with original promotions, and how they can improve their self-service offerings to their customers.


back to the top

Copyright © 2001-2004 VoiceXML Forum. All rights reserved.
The VoiceXML Forum is a program of the
IEEE Industry Standards and Technology Organization (IEEE-ISTO).