 So Han, welcome to ODSC India. Pleasure to have you here. Thank you so much. My pleasure. So Han, you just gave a talk here at ODSC India I was part of the audience myself a very packed room So please do tell us a little bit about yourself and a little bit about your talk So my name is Sohan. I work in Amazon in the Alexa team as a developer evangelist Right, so I spoke a little bit today about the machine learning and the data science that goes into Alexa right Alexa is is At the forefront of voice tech right now There's a lot of speech to text a lot of natural language understanding There's a lot of text to speech all of this involves a lot of machine learning a lot of data science So that was the focus of my talk today Yeah, and Amazon Alexa as many people know extremely impressive product. I remember the first time I brought one home plugged it in and I spoke that understood my Irish accent and amazingly it understood my daughter's Seven-year-old accent at the time and that was very impressive. So what are the breakthroughs in the last couple of years that enabled These models were instead of train them for long periods of time and training your voice all of a sudden now You can just go straight into giving them voice commands Yeah, so two things one is I think the general speech to text has gotten so much better In fact, I remember reading a report saying we are almost or nearly as good as any human Right, which is like a 95 96% of speech to text conversion And a lot of small improvements have happened on the natural language understanding front as well Right and because we have access to so much more data right now we're able to process it and you know hold context and things like that and It's really improved and and we've had a lot of good talks today about how you know Just small changes in like say a data set or representing certain ways of what a user is saying has led to so much of an increase in You know conversions and speech to text improvement. So And for someone interested in this area, what are some of the models? They should be looking at machine learning or deep learning models. Do you recommend that they take a look at and study? Yeah, so I'd say it depends on what you're really interested in There's a lot of so I spoke a little bit about This thing called a hypothesis re-ranker, right? Which is how the best skill is chosen for what a user saying which uses like a recurrent neural network and Bidirectional LSTMs, right? So if they interest in that, maybe that's that's the way to go There's a lot of things that happen in the speech front as well, right? Like like let's take India for example There's just so many different accents and dialects and ways to represent things So there's a lot of advancements that's happened in speech recognition And one thing I personally like people to really do is maybe pick Indian languages and you know natural language understanding in Indian languages Because that's something that's not been explored too much But I think has a lot of potential to change billions of lives. So excellent and what are some of the other interesting areas of Text on elixir NLP research that you are coming that they're coming to the front for front right now at the moment Let's see now so One what I spoke about earlier is the Indian language support Like in India is a fairly large country and not everyone speaks English. There's just so many different languages So the first step is a lot of people speak English with Words from their native tongue throne and so that is like a challenge in itself And second is conversing in like a Hindi or a Kannada for example, right? Unfortunately Indians have started coming online and very recently So there's not that much data available for like NLP in these languages But now they are right. So there is access to the data. So, you know, there's a lot of advancements I know a lot of Universities and academies are working on this So would really like to see some sort of innovation in Excellent and you mentioned LSTM models like the idea of these models having a memory, but how far away do you think you we are from? moving away from question-and-answer platforms to True dialogue system. So when I ask Alexa, what's the weather? It tells me it's raining. I ask it. Should I bring an umbrella or not with a note answer? Big ask true. Yeah, and bit of what I spoke today about was, you know, how Changing how we represent what the user said helps in like these sort of contextual answers, right? Where if I say something like Find me a restaurant near the cricket game It's very difficult because you have to sort of relate two different intents But now with like the Alexa meaning representation language, which is something that we're working on It becomes easier to sort of represent this interaction So hopefully that will help with you know contextual things like what you said Yeah, context is gonna be very very exciting and so to come so people who are interested in Doing NLP machine learning around that. What are some of the tools platforms? Alexa's got an API. I know that that you'd recommend So I'm more I work more on the Alexa side of things. So to be honest, I'm not too sure about about the you know The machine learning parts of things, but we have an open API and an open SDK We have like a bunch of templates for skills that you can build as well So I just say, you know explore some of that stuff again. I'm not too sure on the other side of things No, no worries and I want to ask you about how your ODSC ODSC India experience has been so far, but you were on the selection committee. So Yeah, so to be honest, I work as an evangelist, so I do a lot of conferences This is one conference that I know is purely about providing value to the attendees, right? It's not about the money It's not about the sponsors It's not about any of that stuff in the sense that I was part of the committee every talk was chosen democratically based on the sort of value it would provide to the people attending which I think is fantastic not too Many other conferences do that and ODSC did that really really well And you know, I was really touched and I was really inspired by how they approached it and so far It's been amazing. I met some great people. So really looking for a great talk Great job in the selection committee and to you and everyone else who helped organize ODSC. So thank you very much Thank you so much. Thank you