Speech Technology Magazine


CallMiner Releases New Eureka Analytics Engine

The latest version of CallMiner Eureka offers improved speech recognition and reporting.
By Tye Pemberton - Posted Jun 30, 2016
Page1 of 1
Bookmark and Share

Making good on promises made at its 2015 user conference, customer service analytics provider CallMiner has released an improved speech-to-text engine for its Eureka product line.

The new engine, which will be available in July as a part of CallMiner's Eureka version 10 upgrade, also features improved performance for low-fidelity call recordings, up to 10 times real-time processing speed, support for more than 20 languages and dialects, and other advanced speech recognition capabilities.

The new engine represents a completely new iteration, advancing from the 9.x series of the Eureka engine (which had to this point advanced into its fifth or sixth upgrade) to the 10.x series, representing a new paradigm for user interface, self-service, and integration, with new benchmarks in speed and accuracy.

"Since we focus on analytics, we've always been about speed to intelligence and speed to insight, and so we're constantly upgrading the performance of our services," said CallMiner's founder and chief technology officer, Jeff Gallino.

With less than 10 percent of CallMiner's business coming from mid-market firms, Gallino notes that larger companies had been asking for an option between handling their analytics completely in-house and using full-featured solutions like earlier versions of Eureka. "The new version lets companies implement our technology in the way that best suits them, whether that be through our upgraded user interface or by integrating our tech with their own using our SDKs and API," he says.

Supporting the call-center functions of Eureka, CallMiner's speech recognition technology has also been considerably enhanced. "Clients' ability to wait is fast approaching zero when it comes to transcription, but, luckily, so is our latency," Gallino says. "Most often, managers and executives are looking to have their call transcriptions ready for analysis by the shift after they were recorded at the latest. Eureka begins transcription with only a four- or five-second latency, now."

Eureka's recognition engine can transcribe 10 times real time, with a high-end accuracy of 80 percent to 90 percent. According to Gallino, this represents at least twice the speed of the previous iteration of Eureka, while still maintaining the improved accuracy.

Faster processing speeds will improve performance for real-time automated monitoring in EurekaLive and near real-time feedback in Eureka Express Mining.

Also included in the recognizer's upgrades is the new add-word feature, which enables CallMiner clients to add words that they want to capture by updating the dictionary to include proper nouns, product names, people, and companies. No utterance examples are needed to add new words to the recognition and transcription functionality.

"CallMiner remains committed to continuous improvement, and the upgraded recognition engine is just the latest example," CallMiner CEO Terry Leahy said in a statement. "Improvements in accuracy as well as speed-to-intelligence can have an immense impact on our customers' ability to rapidly respond to a consumer's question or address an issue. And the capability to support more languages and dialects will allow us to better support our global customers and partners and expand into other regions of the world."

Gallino has other hopes for the technology. "I hope this technology finally puts the 'new and expanding vocabulary' debate to rest," says Gallino, who feels it unnecessarily favors a phonetic approach to transcription. "With machine learning and context models, anxieties over accuracy in an add-word model are unfounded. Users can expand the vocabularies themselves, and with their consent that new vocabulary can be shared with all of our customers via the cloud."

"Some other companies offer APIs," says Gallino, emphasizing the flexibility of the new rollout, "But they're less invested in allowing you access to the tools they've developed. CallMiner wanted to get away from what I call 'data-only' APIs that only let customers strip the data with none of the analytic capabilities. Our API allows new companies to build in our entire feature set to whatever platform they're currently using."

Page1 of 1