.
/v3-uk/analysis/1975898/voicing-concerns-mobile-ecommerce
31 Aug 2000, Liesbeth Evers, Network News , V3
Voice has been touted as the next big thing in ebusiness, but critics say that the technology is not yet up to the job and needs better speech recognition software.
Sunil Soares, programme director of product management at IBM, said that network managers should voice-enable their ebusinesses to reach a wider range of customers, including those who do not have access to computers.
He added that the role of voice is undeniable now that ebusiness is becoming mobile as "fingers aren't getting smaller, but devices are".
"Voice is now booming as a result of the founding of VoiceXML as the standard for writing voice applications, and because of the availability of speedy chips," he said.
However, although voice systems are maturing at a fast pace for desktop applications, their use for ebusiness is still limited to single words or short sentences.
Technology failings
Carey Gray, a research analyst for mobile commerce at Butler Group, said the technology is not up to the task yet, as applications struggle with accents and the often distorted sound from mobile phones. "There is still a real possibility for error, and for use in transactions such as buying shares, that can have financial consequences for users," he said.
Soares said that the next phase of development, natural language understanding, will allow callers to interact and is expected in about three years. This would create the facility for mature voice applications and machine translation. "But that is future. We don't want to get ahead of ourselves," he said.
"For now, benefits are found in the focused use of voice. One customer used voice applications to lower the cost of its sales. They kept the first call as a live conversation, but call back was done using voice recognition, which nearly halved the cost of sales calls," he said.
The limit of voice in ebusiness is caused by the need to 'train' the system to recognise an effective vocabulary. The applications that can deal with a large active word base - up to 250,000 - require substantial training, which renders them useless for short customer contact.
To eliminate the training requirement, ebusiness voice engines reduce vocabulary to the small number of words - about 100 - that are likely to be used for the application.
Voice XML 1.0 was submitted to the W3C in March and was developed from Motorola's VoxML and IBM's SpeechML, both of which have been around since 1998. Without speech recognition or interpretation, it uses ViaVoice for dictation and Java speech grammar format to substitute speech for text.
Easy to adapt
Chris Cross, senior software engineer at IBM voice systems, demonstrated at the developer's conference in Las Vegas how voice applications could work for ebusiness. While a thin electronic voice ran the audience through a routine for requesting a drink with a limited choice, he showed the program behind it and explained how easy it is to change the text script.
"The strength of using text in VoiceXML is that the questions and possible answers can easily be changed, without much knowledge of programming," he said.
However, his audience said it would rather exchange that feature for one having a recognition capability that allows mature use for ebusiness.
One developer said: "People should have VoiceXML imbedded in their phones so that it would get training from regular use and then send that information in text format to the web server."
This suggestion received enthusiastic support from others. When pressed to say what the chances are of getting speech recognition running on a mobile device, Cross said: "That is obviously where we want to be." He added that IBM is currently experimenting with embedding voice recognition in such devices.