Integration for Delegation

In our recent press release we announced that applications that are richly connected to enterprise backend systems can yield up to three times the automation rate of non-connected dialog systems. The reason for this dramatic increase in automation performance is simple, and is called “delegation”. Rather than having callers perform certain operations or provide certain pieces of information, the dialog system delegates other systems and information repositories to do that. We like to say: “The best question is a question not asked,” to stress on the fact that if there are other ways to collect some pieces of information other that asking the caller, that should be done. By delegating the collection of information or the performing of some actions to external enterprise backends rather than to the caller would lead to better interaction experience and higher automation.

Equipment identification in technical support calls is the perfect example. During a technical support call for internet service it may be important to know the modem type of the caller. In order to provide that information, often the caller has to drop the phone, crawl to some unreachable place, like the bottom shelf of the entertainment cabinet, locate the modem—which is not a simple thing for everyone…with some many boxes and cables around—locate the brand name, go back to the telephone, and speak it. In the meanwhile noise may have triggered the speech recognizer, and there is always the chance of misrecognition. The whole thing may take a few minutes that add to caller frustration and increases the chance of a much dreaded hang-up or “operator!!!” Instead, a dip into the customer account database to locate the caller’s records and the type of modem would take a few seconds, can be done in parallel with other tasks—for instance collecting the reason of the call in natural language—and lead to a much more pleasant, reassuring, and successful interaction.

Let’s face it! Subscribers paid for a service, and asking them to do all this work when something is wrong is not the best possible customer care. Delegating machines to do all the work—rather than callers—is the way to go. We are moving towards a time where an automated customer care call would go like this:

System: Thank you for calling Acme customer care. How can I help you?

Caller: I just got my bill and there is a one hundred and fifty dollar charge I don’t understand.

System: I understand your bill is too high. Is that right?

Caller: Yes

System: I am sorry about that. Let me see what the problem is, and I will call you back in a few minutes.

We are not there yet…but we are moving in the right direction.

What Happened in Vegas

Last week, from March 30 to April 4, Las Vegas hosted ICASSP, the International Conference on Acoustic Speech and Signal Processing. This is one of the largest conferences of the IEEE Signal Processing Society which is held every year in different locations around the world (next year ICASSP 2009 will be held in Taipei, Taiwan). More than 2000 participants, more than 1300 academic papers presented at the conference, 300 of which dedicated to speech and language technologies.

Topics related to speech recognition, spoken language understanding, and dialog technologies are always very hot and represent the majority of the presentations in the speech and language area. That is a clear indication that the interest of academic and industrial research in the human-machine interaction using speech recognition has not diminished, but has been growing steadily during the past years.

So, what is hot in speech and language technology research this year? Certainly voice search and its applications is one of the areas that attracted a lot of attention, with a special session entirely dedicated to the topic. However, besides basic research on speech recognition and language modeling, there were also several interesting presentations on the most recent advances on dialog learning, emotion detection, speech translation, audio mining, spoken information retrieval, and spoken language understanding.

The Show & Tell session was one among the most interesting events; it attracted hundreds of people for a whole afternoon. SpeechCycle presented a demo of our “third generation” customer care systems, including the advanced tools that participate into what we call “the cycle”, such as authoring, reporting, learning and optimization of dialog, speech accuracy tuning, annotation, SLM training, and online behavior modification, based on our latest QuickTouch product. Our demo gave to the many people that stopped by our booth a clear sense of the advancements that a company like SpeechCycle has brought to the commercial world. We at SpeechCycle have a good tradition of technological innovation, and we cherish our strong links with the worldwide research community in the speech and language areas. Contrary to the popular say, this time we hope that what happened in Vegas won’t stay in Vegas.

-------

SpeechCycle’s papers presented at scientific conventions and workshops:

· Evanini, K., Suendermann, D., Pieraccini, R., Call Classification for Automated Troubleshooting on Large Corpora, ASRU 2007, Kyoto, Japan, December 9-13, 2007

· Albalate, A., Dimitrov, D., Pieraccini, R, Unsupervised Categorisation Approaches for Technical Support Automated Agents, Interspeech 2007, Antwerp, Belgium, August 27-31, 2007.   

· Acomb, K., Bloom, J., Dayanidhi, K., Hunter, P., Krogh, P., Levin, E., Pieraccini, R., Technical Support Dialog Systems, Issues, Problems, and Solutions, HLT 2007 Workshop on “Bridging the Gap, Academic and Industrial Research in Dialog Technology,” Rochester, NY, April. 26, 2007.

· Levin, E., Pieraccini, R., Value-Based Optimal Decision for Dialog Systems, Proc. of IEEE/ACL 2006 Workshop on Spoken Language Technologies (SLT 06), Aruba, Dec. 10-13, 2006.

50,000,000 Calls

We recently announced that SpeechCycle has processed 50 million calls. Beyond the successful execution of a process that delivers high levels of automation to the largest US cable companies, this achievement can be attributed to several levels of technological advancement and innovation brought by SpeechCycle. Here are some:

- Our analytic and reporting tools that enable the analysis of millions and millions of calls and help provide higher and higher levels of automation.

- The capability delivered by our integration software to seamlessly exchange information with enterprise web services and deliver what we characterize as an “immersive caller experience”.

- The flexibility of our SaaS platform with respect to different customers, backends, applications, call-centers, and its scalability towards high volumes of calls.

-  A special and deep understanding of “data driven” speech technology for continuously improving grammars, statistical language models, and the Voice User Interface.

- A data bank of millions of utterances, transcribed and semantically annotated, that represent one of the richest linguistic inventories of caller expressions and responses.

All of this allowed SpeechCycle to create its “rich phone applications” which are beginning to move beyond the area of technical support and expand into new industries as well.