Hello All,

I am a new participant in the chatbots.org world. This has been a topic of interest to me since I visited the Jet Propulsion Laboratory in 2001 to review a technology they called Digital Personnel. It was capable of accurately producing human avatars and unique speaking voice making the avatar capable of realistically impersonating an individual in the digital world. They envisioned a day when you could “rent” a celebrity to do a commercial and they would license their avatar to give a pitch. Cheaper, faster and no travel…sort of a visual “voice over”.

My question to the group: “What do you see as the role of the biometrics software companies in chatbot development”?

I added a poll for your input but the explanations are the important part.

Thanks and I look forward to contributing to the group.

Regards,
Barry

Hi, Barry, and welcome to chatbots.org.

I think another category would have made a nice addition to the poll, in the 3rd position: “All of the above”. After all, to place “role limits” is to reduce functionality, and makes a system or company less robust than it could/should be, so why restrict it’s role to an “either/or” scenario? That’s why I voted “other”.

Dave,

Good input. Let me see if I can edit the responses.

Best,
Barry

I would say “Facial Recognition and 3D morphing/avatar generation” and “Speaker verification/voice duplication” are the most important, especially in a commercial product where the primary purpose of the bot is not simply chatting. Emotion recognition might help the bot deduce if it is providing satisfactory or unsatisfactory responses, but since there are so many varied causes of a user’s mood, this would need to be coupled closely with a sense of context (concerning the conversation and the user’s situation in general) to be useful.

@Barry: Welcome here as well!

My answer would be :‘All of the above’, but the timing is different.

From a pessimistic point of view, I believe that the CAPTCHA’s (those annoying characters to be retyped all the time), will be cracked soon. Then, we’ll need another type of identifying a human, and I believe this will be speech to start with. Once we have speech integrated in our browsers (ever heard about the Voice Browsing working group in the W3C? http://www.w3.org/Voice/) human authentication will be next step.

During the introduction, TV’s & game consoles will gradually integrate (i.e. split into a display and intelligence components connected through fiber) into new devices having stereo camera’s built in as a standard.

Then, image recognition, facial recognition and emotion recognition will be next steps.

Eventually, we’ll have virtual agents, artificial human beings, as servants, as friends, companions, advisors, negotiators, coaches, and they will all communicate with us, observing us like any other human would do.

Erwin,

Even though we spend the majority of our development efforts in the facial recognition arena, I agree that voice recognition/speaker verification has the most near term potential in many applications. The key inhibitors are perfecting the performance over disparate networks and inconsistencies in managing very large populations. One of the earliest commercial biometric access control projects took place in the Philadelphia area about 10 years ago and used voice biometrics. You simply spoke to your door and it opened. For verification i.e. replacing CAPTCHAs, the technology is sufficient today. The templates are small and suitable for any device with a built-in microphone.

I am not familiar with the voice browser project you mention but I will check it out.

Best,
Barry

It’s just a matter of time. I’m not running through my LinkedIn group updates and I’ve noticed that some groups are full of spam. It get harder to fight spam as Captcha’s are being hacked all the time. So we need something else.

Do you know whether TTS is already so good today, that ASR can’t distinguish a real voice from a synthetic voice?

Erwin,

There is an inconsistency in my previous message I need to clarify. I stated the technology was good to go but pointed out the limitations when applied to large databases. To be more specific, I believe the technolgy is sufficient for 1:1 verification but would have difficulty with 1:many identification apps. To replace CAPTCHAs, it wouldn’t help much with new enrollments to a site but could be an improvement for existing users who are enrolled with a voice biometric.

An additional issue that I do not know the answer to today is the ability of the algorithms to deal with a recorded voice. If the software is text independent, the systems could be susceptible to someone recording your voice and playing it back through the ID system to gain access.

I have been away from the development on the voice side for several years but I still have a network of subject matter experts in the field if you wish to dive deeper. Check LinkedIn for Luigi Bojan. If you like I will send him a recommendation to connect to you and we get him on this thread. He would be a real asset on all things ‘Voice” related.

Best,
Barry

Barry Hodge - Nov 4, 2011:

To replace CAPTCHAs, it wouldn’t help much with new enrollments to a site but could be an improvement for existing users who are enrolled with a voice biometric.

That would mean, once they are logged in, but when additional security is needed, for completing a transaction, voice identification might be used?

Barry Hodge - Nov 4, 2011:

I have been away from the development on the voice side for several years but I still have a network of subject matter experts in the field if you wish to dive deeper. Check LinkedIn for Luigi Bojan. If you like I will send him a recommendation to connect to you and we get him on this thread. He would be a real asset on all things ‘Voice” related.

I’d be delighted to meet him in this thread!