Speech Processing for Disease - Dr Ami Moyal, President, Afeka Tel Aviv College of Engineering - Voice Tech Podcast ep.076

Automatic TRANSCRIPT

Symptoms. So covid nineteen. Vocal the. And throat infection. These affect human localization patterns. We are modeling samples of speech coughing and breathing from both symptomatic and Azima the night to carry us to compare with models taken from healthy subjects. AUTOMO-. We are also modeling vocalisations subject the tested negative for covid nineteen yet are exhibiting similar symptoms shot such as those infected by the common flu. DC will a loaded the commodity who differentiate between someone who is likely a carrier and someone? Well. That makes perfect sense I mean obviously someone who is infected and is showing symptoms is going to sound different to someone who is isn't so I completely understand how you can train these machine learning models to recognize the difference I'm fascinated by the ability to be able to distinguish between different diseases though because I would imagine the you know if you have one type of disease or another, the reaction of your body would largely be the same the produce flam in the lungs which would come up in a fatal throw in different ways I'll be really really interested to know what the difference is. In how that sounds depending on the on which disease that you have this this is really on the cutting edge of things. I wanted to ask you what what you think more. Generally what you think the the frontiers of discovery are in speech processing and what you think we'll be able to do with our voice in the future anything and everything. The early usage of technology was mainly focused on human machine interaction. People were initially hesitant speech recognition even for simple voice commands, but this perception has changed over the years. In the not. So distant future, we will be able to communicate with any object machine using our voice. And they will be able to communicate with us in return using artificial voices in a way that precisely simulates human-to-human direction. If we take, for example, the current project on Covid nineteen step partner. I imagined that one day of us will be analyzed continuously by ourselves. which will not defy ause. In real time. went to go to see a doctor because he discovered the change in our voice that may result from. This episode is brought to you by Manning publishing an independent publisher of a huge range of self development books. At Manning Dot com, you'll find books on all the technologies you need to learn in order to create world class voice applications. All the books are available at Manning Dot com, and right now, voice podcast listeners a massive forty percent of all the books with the Promo Code, pod voice tech- nineteen to go check it out at money dot com. That reminds me of another study I saw detecting Alzheimer's disease or the onset of Alzheimer's disease using voice signals as well. Similar kind of thing that continuous monitoring program that would listen to the words you say every day all day every day. And measure the pauses in between the words as you know, Alzheimer's patients often struggle to find the right word. This is the kind of thing that only machine pick up over the long term looking at all your utterances in aggregate. So. Yeah. Completely understand the the need for to that continuous monitoring and it sounds like it has You know many many applications. Love the fact that you're an optimist as well with the gods to everyone using voice in the future. This is exactly what we talk about on this show. Every day voice is going to be used in all sorts of different ways every part of the day, all these tiny little use cases that we do now touching tapping and swiping will gradually convert towards a hands-free modality. Before you know it will be using voice all day every day. Clearly you guys in the universities are looking ahead to the future you're optimists. And you're renovating, of course through three research who asked you think the biggest innovators in the voice tackles speech technology world today since major companies such as facebook comments on and Google have defined severe speech recognition as Petitional I. Expect that we'll see major advances in technology and its use. Human Machine Interaction will become realistic in the sense that we will be able to communicate with four boats with the central entity that we communicate with each other. Even maintaining an intelligent dialogue in fact, we will communicate freely read any machine or device whether it's our mobile phone, a refrigerator, our robots all call. In addition I think the entire world of search change the ability to automatically transcribe audio and video accurately using speech recognition. will allow us to easily search content within large quantities of video and audio data. I also believe the medical family greatly influence tools that analyze speech will be available for general use. This will be able to detect. Anomalies in our speech of voice patterns, the thunderstorms, potential illness, providing further agnostic analysis that alert us when we should pay a visit to the doctor. Well I'm completely with you there on both points. On the point of. The human computer interaction becoming much more lifelike, seeing that all the time with these Texas beach voices with a dialogue management improving all the time. And on the search side is while the googlies is increasingly prioritizing content as a first class citizen. Turning in search results, using it for SEO purposes to determine the best web page to return and much more besides I'm sure. And this is exactly what we're doing with this tool that I'm recording this podcast with now rumble studio. We think that over time these kinds of interactions that we're having right now human-to-human acing Crinsley willing will gradually be overtaken by human to machine conversations where the machine is asking the questions or responding with follow ups in order to be able to get more of that information from guests, and then on the search side wants that audio content is published of course. Google, then more easily find it because that'd be transcribing all of this audio and by recording the questions blocks atomic Louis, and that audio would be in the post for him to return as a as a search was out. So I really think that's something that a lot of people should pay attention to now is how that generating the content so that it's picked up by the search. Algorithms of the future which are going to be instigated through voice search queries. Of course, this intuition is a common knowledge everybody outside of the voice and speech technology world's appreciates this. So what do you think every startup or startup employ should know about the voice and speech industry and is there any bad advice or any bad ideas that you come across that? You feel people should ignore. I think that speech processing technologies whether being speech recognitions equal recognition. ignition or evenly motion detection as which the stage where eight the is application programming interfaces or closed models that include human machine interface are readily available. This means the programmers emphasis than using which colleges. Can simply access dysfunctions and integrate them within their application alongside this researchers in the Philo speech assessing whether from academia or industry. We've continued to advance the core technology in

Coming up next