meta data for this page
Differences
This shows you the differences between two versions of the page.
| Both sides previous revision Previous revision Next revision | Previous revision | ||
|
personalization [2018/06/15 12:18] miket_forty7ronin.com Update links |
personalization [2019/08/08 12:37] (current) lisa.illgen_concentrix.com Added Anchor Links |
||
|---|---|---|---|
| Line 39: | Line 39: | ||
| **// Don't speak the caller's name unless it's absolutely necessary //**\\ | **// Don't speak the caller's name unless it's absolutely necessary //**\\ | ||
| - | A topic that sometimes comes up in connection with personalization is whether the application should speak the caller's name. Companies who view their IVR applications as extensions of customer service may express a desire to use customer names as they do when scripting interactions between customers and call center agents. Given that in the United States alone there are more than 2,000,000 reasonably common surnames and 100,000 reasonably common first names (Spiegel, 2003a), it's difficult to create and maintain recordings of names. Text-to-speech (TTS) production of names is an alternative, but there are several known difficulties associated with this strategy (Damper & Soonklang, 2007; Henton, 2003; Spiegel, 2003a, 2003b), including: | + | A topic that sometimes comes up in connection with personalization is whether the application should speak the caller's name. Companies who view their IVR applications as extensions of customer service may express a desire to use customer names as they do when scripting interactions between customers and call center agents. Given that in the United States alone there are more than 2,000,000 reasonably common surnames and 100,000 reasonably common first names ([[references#spiegel2003a|Spiegel, 2003a]]), it's difficult to create and maintain recordings of names. Text-to-speech (TTS) production of names is an alternative, but there are several known difficulties associated with this strategy ([[references#dampers2007|Damper & Soonklang]]; [[references#henton|Henton, 2003]]; Spiegel, [[references#spiegel2003a|2003a]], [[references#spiegel2003b|2003b]]), including: |
| * Uncommon patterns of English letter sequences | * Uncommon patterns of English letter sequences | ||
| Line 46: | Line 46: | ||
| * Names that include symbols or punctuation (accents, umlauts) that are not part of standard English | * Names that include symbols or punctuation (accents, umlauts) that are not part of standard English | ||
| - | We know of no research on the impacts of mispronouncing a caller's name, but it's hard to imagine that mispronunciations could do anything other than damage the relationship between an enterprise and a customer, making this a risky approach to creating a positive first impression. Estimates of the pronunciation accuracy of proper names produced by general (untuned) TTS engines range from about 50% (Henton, 2003) to 60-70% (Damper & Soonklang, 2007) to 70-80% (Spiegel, 2003b). After 15 years of research dedicated to improving proper name pronunciation via TTS, Spiegel (2003a, 2003b) reported a tuned system that achieved 99% correct pronunciation for common names and 92-94% for uncommon names, demonstrating that this problem, although very difficult, can be solved. The implications for VUI designers are: | + | We know of no research on the impacts of mispronouncing a caller's name, but it's hard to imagine that mispronunciations could do anything other than damage the relationship between an enterprise and a customer, making this a risky approach to creating a positive first impression. Estimates of the pronunciation accuracy of proper names produced by general (untuned) TTS engines range from about 50% ([[references#henton|Henton, 2003]]) to 60-70% ([[references#dampers2007|Damper & Soonklang]) to 70-80% ([[references#spiegel2003b|Speigel, 2003b]]). After 15 years of research dedicated to improving proper name pronunciation via TTS, Spiegel ([[references#spiegel2003a|2003a]], [[references#spiegel2003b|2003b]]) reported a tuned system that achieved 99% correct pronunciation for common names and 92-94% for uncommon names, demonstrating that this problem, although very difficult, can be solved. The implications for VUI designers are: |
| * Unless an application absolutely requires it, avoid designs that address the caller by name. | * Unless an application absolutely requires it, avoid designs that address the caller by name. | ||