Verint’s Speaker Separation and Automated Theme Discovery Buoys Speech Analytics

Verint Systems recently launched a revamped workforce optimization (WFO) solution that leverages speech analytics to provide better insight into customer engagements.

The WFO suite has three separate pieces: customer analytics, workforce optimization, and engagement management. As part of the customer analytics suite, the totally rebuilt speech analytics engine offers faster speech-to-text conversion while providing deeper insights into interactions. Additionally, the WFO suite uses voice biometrics technology that Verint gained when it acquired Victrio in 2013.

Kristyn Emenecker, global vice president of enterprise workforce optimization at Verint, says the company spent three years rebuilding its speech analytics engine after studying its contact center–specific language model that it has made commercially available for the past 10 years. "We knew it was time to get rid of any commercially available language models under the cover and build the whole thing using the knowledge that we've gained," Emenecker says.

As a result of that heavy re-engineering effort, Verint improved its transcription accuracy across every language. The levels of improvement vary by language model, she says, but a 17 percent increase in precision is the average. "We were able to get a lift in transcription accuracy just by using the model of spoken conversation in contact centers," Emenecker explains.

The upgrade also provided a boost in processing speeds. "We were able to literally cut the hardware requirements in half, which will have a significant impact on our customers' total cost of ownership," she says.

Verint also added a software speaker separation solution. Now, contact centers can see where a customer is saying something in the interaction relative to when an employee is speaking. Emenecker says that even without the benefit of a stereo audio or mono audio feed, both speakers are clearly heard and identified. "Other companies will claim to have speaker separation, but what they may not tell you is that you can only support separation if you have a phone system in stereo and if you do your call recording in stereo," she says.

Emenecker calls the solution "groundbreaking" in that the phonetics—the sounds themselves—are able to be distinguished.

"In the past, contact centers probably listened to just 1 percent of customer conversations, mainly listening for employee compliance, and maybe they could get a couple of nuggets of customer insight. Now you can get trends and patterns across one hundred percent of calls," she says.

That ability helps provide customer context, since the solution can tell the difference, for instance, between an angry customer and an angry agent. "There's a big difference whether it's the customer or the agent who is upset," Emenecker says. "This really influences the actions you're going to take or the importance you put on the interaction."

Another addition to the speech analytics engine is automated theme discovery. The system looks for words that are occurring with statistical frequency in the same calls and spots trends. The theme is created when words are heard together and frequently, allowing the solution to self-learn. "It continues to learn based on the spoken words in a contact center's conversations over time. Every single day, it's refreshing the theme and searching," Emenecker says. "It [goes] beyond what you can do with manual searches that are innately biased because people tend to look for different words.

"We really wanted to take the challenge out of speech analytics," she continues. "We wanted to bring the benefits without any barriers. I think this is a huge step forward. We think this is a game changer and a differentiator for Verint."

The new release also incorporates voice biometrics technology that Verint gained from its acquisition of Victrio. Emenecker points out that since Verint does a great deal of voice recording, the company holds a lot of voiceprints and realized it could put them to use.

With the new capability, the agent can carry on a conversation with the customer while the system compares the voice with a voiceprint on file. A notification tells the agent when she doesn't have to ask any more security questions.

"It makes for a natural conversation for the customer. It's completely frictionless," Emenecker says. "It offers better fraud protection, shorter average handle time, and leads to happier customers."