Speech Technology - October 2008 - (Page 22)

NO ONE UNDERSTANDS ME! SPEECH & EMOTION to-consumer (B2C) environment, but the technology is very limited. Given the lack of real-world applications, claims that speech technology vendors make about their emotion detection solutions are just that—claims, Fluss says. Nevertheless, she professes to be a champion of emotion detection solutions. “I’m anything but cynical about this,” she says. “I think it has great potential.” But as it is today, emotion detection as a standalone cannot stand. When used as a component of a complete speech analytics and workforce optimization (WFO) suite, however, it can provide tremendous insight into understanding the critical components behind why a customer is emotional and how to react. It contributes to the bigger puzzle when it comes to deciphering a customer call. With this added dimension, companies can diagnose and treat the pain points of not only products or services, but the performance of CSRs. The knowledge feeds into a perpetual cycle of improved customer service. Techno-Voice The potential of emotion detection has gradually pushed it into the spotlight, becoming one of the top three features customers ask for from call monitoring software provider Autonomy etalk. It follows behind concerns about speech processing and speaker recognition—most commonly used to distinguish between the agent and the caller. Still, Roger Woolley, vice president of marketing at Autonomy etalk, admits emotion detection is not “prevalent” in most call centers. He estimates that only two out of 10 companies even inquire about real-time capabilities, while most opt instead to perform analytics on recorded speech. But don’t be too quick to blame technology. Reasons behind low market penetration revolve primarily around budget and operational priorities, Woolley says. For its part, Autonomy etalk converts calls into a .WAV format and then dives in to determine their audio makeup. The solution has an emotion detection component that evaluates the call for the speaker’s tone, loudness, pitch, speed of sound, and cross-talk, which happens when more than one person is speaking at the same time. Based on these factors, the solution can then be set to highlight areas of emotional inflection, making it easier and more efficient to identify instances during calls in which disputes might have occurred. “In the early days, companies had to staff organizations with quality assurance teams and supervisors and manually listen to those calls,” Woolley says. By listening to calls, call centers derive a voice?” After all, voice analysis happens almost instinctively in everyday conversations, cluing us into the intricacies that aren’t directly verbalized. Happy? Sad? Angry? While it’s convenient to assume, say, an increase in volume as indicative of heightened emotion, it’s also easy to see the obvious drawbacks of such thinking. What’s the difference between happyloud and angry-loud? What about those who aren’t volatile, but rather, become very quiet when they’re angry? Emotion detection in speech analytics systems often doesn’t recognize these intricacies, Fluss says. Autonomy etalk looks for large inflections in a call. “If I just say, ‘I’m really getting tired of you asking me questions,’ in a low, monotone way, it might not pick that up,” Woolley says. Therefore, emotion detection cannot depend on acoustics alone, says Daniel Ziv, vice president of customer interaction analytics at Verint. Even in everyday speech, emotion is defined by a combination of acoustics and linguistics, or what Woolley refers to as the context. By looking just at the acoustics (i.e., loud and soft), analytics systems can often be thrown off by the interference of background noise (e.g., rushing traffic, a loudspeaker announcement, or a crying baby). When the technology tags a call as characteristically emotional, other tools in a speech analytics solution can take that call, mine it for specific words and phrases, and compare it to calls that were considered nonemotional. This helps pinpoint what’s uniquely driving emotional calls, whether it’s the product, the agent, or a billing issue. Emotion detection can even be used to compare what a customer seemed to feel with what she says she felt. Most experts recommend pairing calls with postcall surveys that ask the customer to rate the quality of the call: How did we do? Did we resolve your problem? Would you recommend us to a friend? “THE PROMISE OF THAT YOU LET THE TECHNOLOGY DO THE LISTENING FOR YOU.” SPEECH ANALYTICS IS plethora of lessons that impact best practices, coaching opportunities, or even tracking the development of a particular campaign. “The promise of speech analytics,” he adds, “is that you let the technology do the listening for you…you no longer have to look for the needle in the haystack. You can go directly to the calls that you’re most interested in.” This isn’t to suggest, however, that the early days of speech analytics are over, especially when it comes to emotion detection. Even if the technology is available, there’s still a learning curve to overcome. Amir Liberman, CEO of Israel-based speech analytics solutions provider Nemesysco, recounts how one company he spoke to recently questioned the value in analyzing voice. His response? “How can you not analyze 22 | Speech Technology OCTOBER 2008 www.speechtechmag.com http://www.speechtechmag.com

Table of Contents Feed for the Digital Edition of Speech Technology - October 2008

Speech Technology - October 2008 Contents Editor’s Letter Industry View Inside Outsourcing Interact Keynoter Highlights the Shrinking Technological World Former Hacker Tackles IVR and Biometrics ‘Press 1’ for Caller Thoughts Soundbytes Voice Vote A New Dragon Emerges Overheard/Underheard An Emotional Mess Emotional Intelligence The Case for Call Recording Unified in Care and Communications An Education in E-Learning Guest Column Standards Speech Solutions Voice Value Forward Thinking

Speech Technology - October 2008

For optimal viewing of this digital publication, please enable JavaScript and then refresh the page. If you would like to try to load the digital publication without using Flash Player detection, please click here.