Video Transcription

Why Choose LumenVox - Part One

Hi, I am Stephen Keller and welcome you to the first part of our why choose LumenVox series here at This video series is going to cover some of the reason why you would want to choose LumenVox as your speech recognition provider.

This assumes that you have already made the decision to use speech recognition. If you haven't, we have other videos here talking about why you would want to use speech recognition in the first place. If you already have a speech recognition solution and you're looking to switch recognizer providers, this video is also appropriate.

We're going to talk about some things that makes LumenVox as a company unique, talk a little bit about our technology, licensing costs, and our overall cooperate philosophy and how that relates to the product and our relationship with our customers. In this video, particularly, we're going to focus on what it is that makes LumenVox different from the other speech vendors.

Speech accuracy is mostly the same.

One thing I will say is the technology, especially as far as accuracy is concerned, is largely similar. This sounds weird coming from a technology company but here we have some numbers showing our accuracy versus one of our top competitors. We have this published in more detail on our website.

The fact is for the most part us and the other speech recognition providers have a very similar sort of product that gets similar accuracy, in some domains they might be a point better than us and we may be a point below them, 2 points here, whatever.

The truth is that it's all pretty close, to be honest. This trend will continue, because as technology gets better, as computers get faster, as speech recognition matures, we're going to see a gradual growth to a 100 percent accuracy or just below it. You can't ever get 100 percent accuracy but 99.59 percent or something like that eventually.

As time passes, technology will become even more similar.

We're already starting to see this. We have this graph here showing the accuracy over time and we see that we're already starting to reach this point. Where everyone is in about the high eighties, ninety percent for most domains in terms of accuracy and we're going to just begin slowly diminishing returns.

We have seen a real drop off in terms of gains in accuracy. Between speech recognition generations, the gain is much less than what it used to be. This is just a normal technology curve that we would expect to see, so we're all becoming similar in terms of accuracy.

Similarity in technologies is due to the adoption of open standards.

Another thing that has really fueled the similarity in terms of technology has been the adoption of open standards. This has been true throughout the information technology industry over the last ten or fifteen years, particularly in computer telephony since the late nineties and early two thousands.

We have seen things like the media resource control protocol (MRCP). This describes how a speech recognizer and a voice platform communicate. It outlines certain functions that a recognizer has to support and describes the communication between them, and has to follow a set pattern. We've also seen things like the speech grammar specification (SRGS) or the semantic interpretation specification (SISR), which sort of goes with grammars.

These determine what grammars can look like when your building speech applications and everyone supports these. Us and all the other speech vendors support MRCP, SRGS, and SISR, and these kind of constrain our behaviors. We have to support these specs and we have to behave the way they tell us to and this makes us all similar.

Also, on the platform side of things the VoiceXML movement, which we have really seen become important in the last few years, describes a standard method for writing applications. This of course also relates to speech recognition because it says your speech recognizer has to support certain functions that can be in your VoiceXML documents. So we support these, the other guys support them, and these open standards.

The whole point of them is to make things easy for you to switch between speech providers, so that if you have your solution you can just rip out your speech provider and plug in a different one and everything still works correctly. That's the whole idea and at LumenVox we completely support this. You should have choice, freedom, and we completely believe in offering this and supporting these. Our competitors have also done the same, so it makes it easy to move in between providers.

However, while the technology is similar, the companies couldn't be more different.

So why would you want to pick one provider over the other? The fact is, even though the technology is similar the companies that provide them are quite different. LumenVox as a company is much different than our competitors. Let me give you a great example. We have a project down in South and Central America, it is the Spanish equivalent of what we call OnStar here in North America.

Do not take our word for it.

It's a product by General Motors and it's an in car kind of computer, it does voice dialing, navigation and other kinds of stuff. You press this little button in your headliner and it activates a microphone in the car you can issue commands to and do all sorts of cool stuff. Down in south Central America it's called ChevyStar and a customer of ours is the one actually implementing this for GM down south, this customer is called Redmond Software and they had an existing solution built on one of our competitor's speech technology.

They took our competitor out and put us in. And after working with Redmond we actually achieved a 42 percent accuracy increase, but this all wasn't from the core technology. Like I said, accuracy is very similar so if the raw accuracy is similar how is it that we were able to achieve this increase over our competition? As Redmond's CEO says, "It was largely trough a close collaboration and support from LumenVox that we were able to do this."

Accuracy Gains

What happened is they had a very challenging application in that they had a whole lot of noise in the car. You're in your car, you're using basically a speakerphone pressing this button and it's difficult to hear. There's noise, there's people in the car, back seat talking, yelling, whatever. So what we did is we took a bunch of audio from them and we worked with Redmond. We had them describe their situation, we listened to their audio, we looked at their grammars, we looked at their settings.

We said, here we go: we can revamp these grammars to make them tighter and more accurate, we can tweak these engine settings, voice activity detection to make it more accurate, and then we finally did something that I think most other companies wouldn't do for our customer — or at least wouldn't do without charging them an arm and a leg.

We actually took their transcribed audio and we built it into our recognizer. We made it a part of the acoustic model, the data that we train our speech engine on to understand speech. We took their noisy data with their Spanish speakers and the dialects of Spanish being spoken and improved our own acoustic models. Therefore, when we redeploy our new Speech Engine with this new technology in place, they actually have exactly what their environment is inside of our Speech Engine perfectly represented and their accuracy goes way up. We have worked with them, understood their problem, and provided them a solution.

Redmond Software and General Motors aren't alone.

Redmond and General Motors are not the only companies that know about LumenVox and see the difference. Over the last couple of years, we have seen a large increase in the number of developers working with us because they too see that difference, they know that LumenVox difference.

Why choose LumenVox?

So why choose LumenVox? Because we're focused on you, our customers. I'm assuming if you're watching this you're probably going to tend to be a platform provider, systems integrator, application developer, a professional services company, any one of these things. You work with our technology to build stuff that other people are going to use and we are really focused on helping you succeed, our competition not so much.

Our competitors compete with their customers.

The fact is that other speech vendors are looking at providing end-to-end speech solutions; they tout this in their press releases. They want to be every link in the supply chain from core technology to finished application to professional services.

Ultimately where does that leave you, if your speech vendor is doing everything, they are trying to sell you everything, trying to sell your customers? If you have to bid against your speech recognition provider where do you stand? I mean, eventually you're gone and they're in. Do you really want to work with companies like that?

LumenVox is your partner, not your competitor.

LumenVox, we want to be your partner. We do not want to compete with you. We want to sell you core technology, you buy these licenses from us, you build a neat application with it, you resell that, when you resell that you resell licenses. We make money, you make money and your customer is happy. There is room in the chain for everyone.

We're not trying to squeeze you out or nickel and dime you to death, so that's a real key reason to choose LumenVox. First and foremost, we want to be your partner and not your competitor. Now in the next video will look at some other things that make us different especially as it relates to licensing and cost factors.


  • Choosing a speech recognition company to work with can be difficult. Not only must you evaluate their technology, but you must evaluate the company itself. In the first of three parts, we discuss the main reasons for choosing LumenVox as your speech vendor, particularly how our corporate philosophy is one of true partnership with our users.


  • Video Playtime:8:54



  • Contact Us
  • +1-858-707-7700
  • Toll Free: (877) 977-0707,
    say "Sales"

© 2016 LumenVox, LLC. All rights reserved.