With customers including Amazon, Twilio, Nasdaq and C-SPAN, VoiceBase is the world’s premier API developer for speech analytics. And as the only speech analytics provider designed with an open architecture, they’ve revolutionized the industry by using a consumption-based model to take prices down to a tenth of what they were at the time of their founding.

Every month VoiceBase processes millions of recordings that allow users to search from the web or their mobile device into the timeline of a recording and play the precise parts of phone calls, conference calls, webinars, educational lectures, podcasts, or videos. And their machine learning technology takes it all a step further by allowing a business, for example, to predict a caller’s intent based on tone, sentiment, and other factors.

They use Solace messaging technology to intelligently route voice, video and text events through their system. The Solace infrastructure orchestrates the flow of events as VoiceBase’s processing cluster transcribes, indexes, summarizes, redacts and conducts predictive analytics on audio and video.

Selection

VoiceBase initially used Amazon Simple Queue Service to orchestrate workflow, but operational challenges and growing event volumes made that approach unrealistic. They evaluated several software-based Java Message Service (JMS) alternatives, but realized that scaling JMS brokers to meet their needs would be complicated and expensive.

“We looked at the architecture we wanted to build and we settled on an architecture that used an orchestration of services, and a key part of that was the concept of queuing,” said Jeff Shukis, the VP of Engineering and Technical Operations at VoiceBase.

Shukis chose Solace for its ability to route millions of messages per second in a small footprint with low operating expenses.

“Only one enterprise service bus worked, and it didn’t even break a sweat.”

“Solace, in that architecture, touches every single task twice. You know, when the task is sent out for orchestration it lands in a Solace queue, when the results are put back into the orchestration environment it lands in a Solace queue. That means that for every job we get it touches Solace between 30 and I think about 1200 times.”

Prior to his discovery of Solace, Shukis always had problems at scale with queuing systems. He even “dreamed of a queuing system that worked like a network switch,” was done in hardware, and could scale exponentially.

And his past problems spelled especially bad news when he projected what future VoiceBase requirements might be: 2.7 billion minutes per month of run rate.

To evaluate the appliances on the market, Shukis and his team built a prototype that simulated those loads by pushing that much data through the queuing system.

Solace won, hands down.

“Only one enterprise service bus worked, and it didn’t even break a sweat.”

Deployment

“The Solace boxes were slated for deployment in Virginia and someone had to do the install. So that was me,” said Shukis. “And I haven’t been back since.”

That was in 2014.

“It’s definitely required the least amount of management of any component in our architecture.”

“We haven’t had to make any changes, fix anything, or change any of the configurations.” 

Support

“I know that anyone I’m dealing with at Solace is going to be a professional, is going to know the way I think about queuing, what I need out of the product and how to help me,” began Shukis.

“My fear is calling up a support organization and getting somebody in the call center whose sole job is to take a note and have someone call me back in four hours.”

“When I call Solace I get someone who, even at the ungodly hours I tend to call, says: “Alright, let me get online and help you. They instantly help us. So it’s definitely enterprise-grade product, enterprise-grade support, which has been fantastic.”

voicebase

Industry:

Speech Analytics

“We built a prototype that simulated the kind of loads we expected to see a few years out. And then we built a prototype that actually pushed that much data through our system and through the queuing system. There was no way that any of the queuing systems we tried could hold up to that.”

Jeff Shukis, VP Engineering & Technical Operations