On this episode of 10 minutes with TalkVia we speak to senior developer El Lee about her favourite features of TalkVia One
10 Minutes with TalkVia Ep.3
P: Hi and welcome to 10 minutes with TalkVia
the podcast where we chat to our favourite TalkVia clients and users about their conversational AI projects.
P: I’m Paige Harkness
P: Today I’m pumped to be speaking to the original TalkVia superuser. We are going behind the screens to chat to one of our lead developers, El Lee. El has been with TalkVia since the beginning and is the developer behind some of our most high profile projects including the Webby Award winning “Last Seen”, The Hit Networks “The Local’s Guide”, “Anzac Stories”, “First 5 Forever” and the “ask me anything” bot for QTAC. El really understands conversational AI and has a gift for taking complex problems and turning them into code. She’s spent the majority of her time this year working on the TalkVia platforms, so I stole 10 minutes
P: El thanks so much for taking the time to join us.
P: So, TalkVia One. For those listeners who are not familiar with our entry level platform, can you please tell us a bit about how you use it? Maybe focusing on your favourite features and what makes it such a great tool for developing Conversational AI projects.
E: Sure, one of my favourite features, and something that really makes the TalkVia platforms stand out is our fuzzy keyword engine, it utilizes a variety of AI algorithms to make keyword matching easier and quicker than if you were building something on the native platforms.
When a user asks something we use keywords to understand a user's intention and take them down the appropriate pathway. TalkVia's keyword algorithms use fuzzy matching to try to identify phonetically similar phrases, which can compensate for things like inaccurate voice recognition due to background noise.
Also, our engine looks at word roots to recognize suffix variations such as dog vs dogs, wish vs wished, run vs running, so you don't need to supply each variation. At TalkVia you can just use the keyword "Bike", there is no need to add bikes, biked, biking or biker as separate keywords.
We have also leveraged the power of AI so it’s constantly learning and getting smarter. So words which are misheard by the device are corrected once they're passed over to the voice engine, making the end user experience much more enjoyable.
A good example of how to use keywords is the QTAC ask me anything bot.
P: Ah yes, great example. For those listeners who are not familiar, QTAC or The Queensland Tertiary Admissions Centre, provides entry and application services for 17 major universities and tertiary education institutions in Queensland and northern New South Wales. So they have high callcenter volumes and a very broad range of questions being fired at them each day. We’ve been working with QTAC to introduce voice and conversational AI as a solution to triaging calls and delivering answers quickly to students and educators.
E: That’s right. The solution we built out for them will form the foundations of their call center conversational AI and includes over 100 questions. TalkVia’s fuzzy keyword technology allowed us to build a voice solution with an enormous scope that is not possible (or at least extremely difficult) to accomplish relying solely on the AI of native platforms.
One of the other big points for QTAC is that exactly the same project was deployed on their website chatbot, Alexa, Google and Siri smart assistants. As the project is being maintained, future changes only need to be made on TalkVia One in one project, and they are updated on all platforms.
Another feature I love is the ability to upload real audio quickly and share it across multiple accounts. For example, The Locals' Guide which was developed by the team at SCA uses professional voiceover for the entire project. The shared media library allowed assets to be reused, which is an integral part of our enterprise media toolset.
P: Oh, I absolutely love that as well! Using voiceover artists makes the experience so much more enjoyable.
Here’s a little snippet of what that voiceover actually sounded like:
As I mentioned in your intro, you’ve built numerous high profile and award winning projects. What are your main learnings and what is your number one piece of advice when it comes to building a high fidelity conversational AI experience?
E: Remember to create clear and straightforward experiences. Conversations are multidimensional and so is a great voice app. Go beyond directed dialog where you read out all the options, like a menu. Instead, try to intuitively inform users on open questions by providing available options as suggestion chips. With TalkVia you can randomise these suggestions to keep things fresh and easily inform users of more things to say. A combination between directed and open ended questions provides a more rich user experience. If your analytics tell you that users often don't know what to say at a certain point then try to improve the experience by supplying more directed questions after the user has been misheard or after they said "uhhh". Always be ready to provide a more hand held experience when nothing has been said or something is misunderstood. TalkVia is really good at easily handling these situations.
P: That’s really interesting.
That’s nearly all the time we have left, but just quickly… Can you give us a sneak peek at what else you’re working on and what can we expect to see in the coming months?
E: Yeah, so we've been working hard on TalkVia Pro, a more advanced platform that provides greater control and fidelity to build complex experiences. I could spend at least an hour talking about the extra AI and features that TalkVia Pro offers, some big ones are natural language understanding and contextual awareness to accommodate for complex phrases like "what's the cheapest flight to LA next Thursday?" With Pro you can easily integrate streams, videos and API calls to your voice app. My favourite part about Pro is that one person can mock up the conversation using our AI driven visual graphing tool, while another person fills in the meat of the voice app. There is something to be said for being able to see the conversational flow of your voice app from a glance and know every pathway a user can go down. It makes future additions or extensions really easy when you come back to your app a couple months later. There's a lot to look forward to with TalkVia Pro and I'm excited to work on natural language generation.
P: This episode featured content and audio from Andrew Fritsh and Tony Byrne from SCA
I’m Paige Harkness and thanks for listening to 10 minutes with TalkVia. You can find more episodes on our website talkvia.com or wherever you find your favourite podcasts.