Tyler Weitzman, Co-Founder & Head of AI at Speechify – Interview Collection
Tyler Weitzman is the Co-Founder, Head of Synthetic Intelligence & President at Speechify, the #1 text-to-speech app on the earth, totaling over 100,000 5-star evaluations. Weitzman is a graduate of Stanford College, the place he obtained a BS in arithmetic and a MS in Pc Science within the Synthetic Intelligence observe. He has been chosen by Inc. Journal as a Prime 50 Entrepreneur, and he has been featured in Enterprise Insider, TechCrunch, LifeHacker, CBS, amongst different publications. Weitzman’s Masters diploma analysis targeted on synthetic intelligence and text-to-speech, the place his remaining paper was titled: “CloneBot: Customized Dialogue-Response Predictions.”
You started coding if you have been solely 9 years outdated, what initially attracted you to pc science?
I used to be fairly obsessed as a child with Dragon Ball Z, and I needed to be taught to animate myself. I discovered Adobe Flash and Photoshop and put my very own animations of Goku on a fan webpage I constructed. It was quickly after I started studying about techniques and algorithms, and once I discovered I might really program for a residing that was fairly thrilling. I believed it was only a interest like taking part in video games.
You then started constructing iphone apps if you have been solely 12 years outdated, what have been a few of these apps?
One app is known as Black SMS that permits folks to ship encrypted textual content messages to one another. One other app was known as Frontback that allows customers to take selfies and pictures of what’s in entrance of them at the very same time.
May you talk about your analysis at Stanford College and the way it was centered round pure language processing and speech synthesis?
My analysis spanned a number of makes use of for transformer networks, together with language technology fashions for chat, part-of-speech tagging, punctuation prediction, and text-to-speech. Optimizing neural community inference for cellular CPUs was a main focus and that straight translated to the offline voices accessible on Speechify, which work even on airplane mode.
May you share the genesis story behind Speechify?
I’m blind in a single eye and my brother Cliff is dyslexic. We’ve used audiobooks and textual content to speech audio know-how for so long as we will keep in mind to get by college and after we have been younger for studying books like Harry Potter. As we bought older and began to make use of extra know-how merchandise, we began to appreciate there was a possibility to construct higher textual content to speech apps on internet and cellular with higher voices due to developments in AI and a greater consumer expertise. So we determined to go for it in Speechify.
What are among the totally different machine studying applied sciences which are used at Speechify?
We’ve adopted cutting-edge methods for superior generative architectures— transformers/conformers, large-scale pretraining, distributed coaching, gradient accumulation, auto-encoded latent areas, diffusion, adversarial networks, and language modeling. We make use of supporting methods for function processing surrounding phonemization, pitch, and emotion, to higher mannequin speech particularly.
What are among the challenges behind constructing a text-to-speech app?
One key problem is constructing top quality voices that sound like actual people somewhat than robots. Our aim is for folks to not have the ability to inform the distinction between how our voices sound and the way people sound, in order that our customers are snug listening to content material on Speechify for lengthy durations of time. A second problem is distributing our AI fashions to thousands and thousands of customers. It’s one factor to construct top quality AI voices and one other to ensure thousands and thousands of customers internationally really discover out about them and use them.
Speechify is the #1 app in its class within the app retailer, what do you attribute this success to?
We imagine we’ve constructed one of the best merchandise out there for individuals who need to hearken to the studying they should eat – whether or not it’s college students with homework, professionals who’re studying for work, or leisure readers who simply need to be entertained. Now we have one of the best collection of voices, together with celebrities like Snoop Dogg, and one of the best consumer interface for folks to simply add and entry the content material that they need to eat. And our consumer expertise is seamless throughout the Speechify ecosystem – you can begin listening to an article in your pc after which simply zap it to maintain listening in your cellphone.
What are among the largest use instances for this app?
Speechify’s generative AI solves actual issues for college kids who need to get by numerous homework sooner, actual folks with Dyslexia and ADHD who’ve bother studying, seniors with low imaginative and prescient, professionals who need to learn extra and be extra productive, writers who need to hearken to their work, auditory learners, and numerous others.
What’s your imaginative and prescient for the way forward for AI?
We would like AI – and particularly AI textual content to speech voices – to eradicate boundaries to studying no matter your revenue stage, studying variations, geography, or language. We see AI as a device for social good to raise the standard of life people can dwell by enhancing their training.
Thanks for the nice interview, readers who want to be taught extra ought to go to Speechify.