The cold doesn't bother us anyway...
On Tuesday our 8th Meet AI session took place in our Cocoon Networks eventspace and despite the snow and freezing cold, our AI enthusiasts and and four-legged friend 'Lola' (pictured below) couldn't be deterred from attending Meet AI 8 to discuss speech and language tech in AI!
To create Meet AI, intended as a multicultural, open and free AI knowledge-sharing platform for Londoners to exchange the latest AI insights and network with likeminded peers, we joined forces with the award-winning London-based AI startup Emotech in May of last year. Our goal continues to be to offer the newest exciting AI content. We invite researchers who are presently tackling AI-related issues and are willing to share their insights and discuss their experiences, successes and difficulties in applying AI theory to practice.
Our 8th session was titled 'Scaling Speech & Language Technology: from Academia to Application' and our two speakers of the evening were Dr. Raymond NG, Research Scientist at Emotech and Team Leader of the Speech & Language Technology research group at the University of Sheffield, and Shawn Wen, Co-Founder and Chief Scientist at machine learning platform PolyAI.
left to right: Dr. Raymond ND (Emotech) and Shawn Wen (PolyAI)
Dr. Raymond NG: "Speech technology: An overview and recent developments"
In his presentation, Dr. Raymond ND traced the journey of speech technology from a purely academic approach to practical application. He stressed how the technology behind machines' synthesised voices had become more natural sounding than ever before: "Over the decades, there has been a concerted effort from linguists, statisticians and computer scientists to help advance speech technology...since the 2010s, deep learning has brought a revolutionary change". Dr Raymond also provided a forecast of how these developments might pave the way to various applications that will greatly impact our daily lives.
Shawn Wen: "Democratising Conversational AI - Scaling Academic Research to Industrial Applications"
Shawn Wen, Cambridge grad and Co-founder and Chief Scientist at machine learning platform PolyAI, shared his personal experience of applying academic theory to practice, i.e. building conversational agents for his business with the aim to democratise and scale these across multiple application domains and languages. He emphasised the"profound impact" deep learning had had on "conversational AI research" in recent years and laid out how models such as CNN intent extractors, DQN policy networks and LSTM language generators have formed the centre of spoken dialogue systems research in recent years. However, as Shawn stressed, these methods are often struggling to scale to complex, real-world conversational scenarios. Shawn concluded his presentation by defining the pillars of democratic conversational AI as a healthy view of Machine learning ("ML is powerful, but not omniscient!"), a focus on generating and owning one's data, and a passion for bringing the most natural user intervoice to modern technology.
In true Meet AI fashion, our audience was very engaged and fired away with their most pressing questions:
"You mentioned democratization, what does it mean?"
Shawn (PolyAI): "Democratization is already used by lots of people in the AI sector. It is the attempt to bring AI to our daily lives and enable any developer to build it. At PolyAI, we want to make it possible for anyone to have access to conversational AI and include it in their daily lives"
"Can you speak about the privacy concerns?"
Raymond (Emotech): "Speech application is one of the applications which draws privacy concerns. Some of our features for the robot Olly, for example the voice trigger and speaker recognition, are completely offline so they don't involve any streaming. And the data won't be stored on the cloud or anywhere, so privacy is fully protected. On top of that we have feature-extraction and encryption so it's very difficult to reconstruct any signal. We know that as a community, the EU is very concerned about privacy in speech application, so new laws will be rolled out in the middle of this year and we are fully compliant with that"
"What platforms are you using and can your model be used on different platforms?"
Shawn (PolyAI): "We are currently building our own platform. In the future we hope to support all kinds of platforms including Google system"
"Do you see any future beyond phonemes or wavenets for speech synthesis, as there are some sounds that aren't actually operating properly at the moment?"
Shawn (PolyAI): "That's a good question. WaveNet and other end-to-end training programs, where researchers are trying to bypass the traditional, classical linguistic structure in learning, are actually quite promising...the hurdles we see are the huge complexity of computation time it takes to generate very high frequency outputs... In the context of this talk I would like to advocate the importance of classical knowledge, particularly in application-driven research. If we know about classical theory and also the structure of the model, it saves us a lot of time and space complexity when we create it. There will then be a lower entry barrier for us to have a state of the art performance"
Both companies are currently hiring, you can get in touch with Emotech via: firstname.lastname@example.org and PolyAI via: email@example.com
Are you an AI enthusiast come rain, storm or snow? Are you keen to join us next time?
To become part of this lively community of AI aficionados, join us for our next session, Meet AI 9! Registration will open soon on Eventbrite and you can be the first to find out by signing up for our newsletter or checking our Eventpage.
Conducting AI research? Keen to speak at our next event? Contact us here!
Also don't forget to engage with us on social media and leave us your feedback, so we can continue to offer these free networking sessions.
We also have a range of interesting partner events coming up, such as our Hardware Pioneers' monthly event and free 1:1 PR sessions with our partner Luminous PR. And if you'd like to find out more about our community and what to expect at our socials, you can read about our Cocoon Networks Chinese New Year and Breakfast Meetup.