It’s a cliché at this level to say that podcasts (and, to a much less publicized extent, audiobooks) have a discovery drawback. To assist repair that, Spotify introduced it’s utilizing Google Cloud’s AI instruments to enhance each content material discovery and personalised suggestions for its audio choices.
With this newest transfer, Spotify is utilizing Google Cloud’s LLMs (giant language fashions) to research the roughly 5 million podcasts and 350,000 audiobooks in its content material library with a objective of attempting to “increase” the metadata, in accordance with a press launch. (The Verge has reached out to Spotify to make clear precisely what “increase” means on this case). A podcast or audiobook’s metadata contains data just like the title, the identify of the host or creator, present notes, and different particulars — all of which come up within the search outcomes of podcast apps like Apple Podcasts or Spotify.
Google Cloud is Spotify’s sole cloud supplier, and as Protocol reported again in 2021, the 2 corporations have been working collectively on extra providers.
Figuring out “dangerous” content material seems to be one other intention of Spotify’s Google Cloud AI experiment, although the corporate didn’t specify a lot about what that may entail. The platform’s rules for podcasters and musicians ban a number of varieties of “delicate” and “harmful” content material — and it at present makes use of a mixture of automated expertise and human reviewers to implement these guidelines. The Verge has reached out to Spotify for extra particulars and can replace once we hear again.
Google Cloud’s AI instruments can even be used to enhance Spotify’s personalised suggestions for podcasts and audiobooks. LLMs shall be used to “higher perceive patterns behind customers’ favourite spoken content material,” which can presumably lead to higher tailor-made suggestions.
At the moment, Spotify will counsel new podcasts and particular person episodes to listeners on their homescreen. On every podcast’s present web page, there’s additionally a “Extra like this part” with related podcasts. Each options could be hit-or-miss, as many customers have famous. (For instance, Spotify’s suggestions for Overheard at Nationwide Geographic embody a podcast about nursing, one on bipolar dysfunction, and a real crime podcast.)
Earlier this yr, Spotify rolled out an AI device for songs, AI DJ, which makes use of OpenAI’s generative AI instruments to create customized playlists. To this point, critiques of the characteristic have been mixed. Spotify’s use of AI for spoken audio isn’t a standalone characteristic like AI DJ, however it’s an indication that the corporate is critical about enhancing the consumer expertise for its non-music choices.