The Conception of PaperPlayer
The beginnings of a revolution in scientific content?
PaperPlayer is a revolutionary new service that converts scientific papers to audio for easy listening. This is the brief story from the early founders of how PaperPlayer came to be and where we’re headed.
Oh by the way, we used our PaperPlayer neural model to convert the blog into a pleasant voiceover if listening to the story is more your speed.
With nearly 3 million peer reviewed articles published digitally per year, keeping up with the ever evolving state of science has become a daunting challenge. Add thousands of preprints and open scientific content to your reading list and staying in touch with your science becomes almost impossible. Information overload is real. So what can we do to increase access and discoverability of new data and studies?
How PaperPlayer came to be
The running joke is that Taylor wrote more code than papers in grad school and keeping up with the literature was always a moving target. After grad school, Taylor was offered the opportunity of a lifetime at Oxford and moved across the Atlantic to the UK. The principal job was writing software to solve technical problems in neuroimaging that the group was facing in various large-scale studies. Working in the codebase and writing new features was gratifying work but again made keeping up with the neuroimaging literature… almost impossible.
At the time Taylor and his wife were living in a small but beautiful country village called Bladon (this is also where Sir Winston Churchill is buried). It was in Bladon one day while walking the surrounding country trails and listening to an audiobook that the idea hit Taylor for a way to transform scientific articles into a more accessible format.
Over a few nights and weekends, he spun up a prototype that pulled open access preprint abstracts from bioRxiv and converted them from text-to-speech using some cloud APIs. Google’s Neural voices were actually really good!
What would the experience be like if cutting edge research could be consumed as easily as listening to a podcast?
Taylor was doing this on a budget, so the entire setup ran on a Raspberry Pi that was hidden in a shoebox in his living room. Tuning some of the voice model parameters and syncing with pod APIs like Spotify he was able to publish the audio abstracts as podcasts. The podcast was self-updating and would automatically check for new papers every hour.
The experiment produced some interesting data.
There were over 4,000 episodes with nearly 10,000 streams at an almost 10% conversion rate in subscriptions on Spotify. Organic traffic was bringing science content to those that needed it most, and was solving a real problem.
The PaperPlayer podcast experience also introduced novelty and discoverability compared to other experiences for consuming science through blogs or audio readers that required manual copy and pasting.
Being admittedly a bit biased to neuroscience, Taylor coded podcasts for every other bioRxiv category to assay the data. Neuroscience turned out to be the most streamed topic, and is also one of the most popular categories on bioRxiv.
Giving it another go
The experiment satisfied a need to stay up to date with the literature and get a feel for the market. But after nearly a year it needed a rewrite. After all, it was quickly written code running on a small computer in a shoebox! That was not a production grade setup and it had its issues.
Not long after hitting pause to rework the code, Taylor was chatting with his colleagues in a group meeting about the overwhelming amount of scientific content. With the global pandemic and the rise of preprints information exhaustion was taking its toll on everyone working in science.
With a new sense of inspiration, Taylor reached out to the first person he often thought of when it comes to startups. His colleague and friend Christian Graves from grad school had spent the last 10+ years starting and investing in companies. He also shared a passion for science communication and was active in the open science space. Hopping on evening calls across timezones showed us we might be onto something and we rebranded, updated the code, and last week released a new look for PaperPlayer on Spotify and Apple Podcasts.
We’re starting with our roots and the most popular category from bioRxiv: Neuroscience. We already have a few hundred abstracts and several hours of content, check it out on Spotify or Apple Podcasts.
We’ve got a lot of plans in the product roadmap and have a new native app experience we’ll be launching soon that will transform the way people consume scientific content.
We will also be writing more about the models we use and how we hope PaperPlayer will help promote open science, inspire new scientists, and reduce information overload. Audio can be a truly transformative medium if delivered properly. With the advances in open AI products like GPT-3 and Dall·E we think the neural models for text-to-speech are poised to be trained on more technical scientific data. We personally enjoy listening to abstracts while on road trips, running, walking our dogs, and doing household chores. We’re excited to build serendipity and discoverability, and to make open science more accessible.
Taylor and Christian