What’s Mobivox Really Listening For?

Nitzan Shaer from Mobivox a company I’ve written about before does “calling by name”. So it’s nice to learn something new and glimpse into the model of how they tune their IVR - voice recognition. Turns out they listen to more than just me saying your name.

He starts with a story that brings to life something I thought about a few weeks ago. What happens when you swear at Mobivox; what happens when you are trying to make that call and she gets it wrong. I’ve joked with a few that it would be cool to have the equivalent of hidden emoticons… Make the wrong statements and you get a response. Well, turns out that Wildfire an early Mobivox like system had just this in place. Example response “. You think your life is hard.. think about me I’m stuck in the box.” Now that maybe more backtalk than I’d really want.

I for one have sworn at Mobivox during a demo; more for effect and because the demo was in a noisy environment. Turns out that is one of the things Mobivox corrects for.

Once you go beyond the traditional 100 year old telephony experience you want a little personalization. It hasn’t really existed so far and certainly hasn’t reached the mass market. So Mobivox is aimed at using voice and working to add the accuracy that humanizes the experience. “What I really want is an assistant!” without creating the animosity that goes along with she doesn’t listen to me. According to Nitzan these are the key elements that impact on the user experience. Environment, Speaker, Device, Channel, Receiver. That’s the path that voice takes… There is some stuff in the middle.

Mobivox’s approach to adapting and personalizing the service to overcome the challenges; where we are, how we talk or shout, what device we use etc. First they create a personal profile for each person. Then they listen to your personal noises (do you snort when you speak?) more likely where you often speak (eg in a school vs in the car). For each person they characterize them and then set thresholds. Second they have created a pool of common noises (streets, bars?) and then trained these models to recognize these signatures based on over 20 features. Mobivox characterizes them and then sets thresholds. It requires rapid sampling of voice vs noise.

To make it all easier they then focus on reducing the vocabulary and set the tone of the conversation so the user tends to say the right things. Frankly, I’m just happy she says “Hi Stuart” with a smile each time I call. At least I think she has a smile.

Lastly they are looking to expand funtionality that enable “Say it and Save it”, Thus say and save a contacts details in your own voice. That’s not completely new, then it does demonstrate that she’s getting smarter; prepared to listen and respond. Personally, as long as she like me and will forgive me the odd swear word I’m in for continuing the relationship.

At the end of the day I’m not sure if these strategies are “generic” to the IVR industry or specific only to Mobivox. Still if they make her more human overtime; if I get my assistant that will always work harder for me then I’m probably going to keep on working with her and not just because they are also a current client.