Hi! I'm Freeman, a research engineer at Sesame. I'm the developer of the turnbench.sesame.com website and also a co-author. For the past bit, I've been working on models for real-time conversation at Sesame. We wanted an evaluation for turn-taking, the understanding of when to speak and when to yield in a natural dialogue.
I wanted to make sure the benchmark felt very accessible and understandable to people, so I built this in-browser dataset viewer at https://turnbench.sesame.com/conversations - it lets you inspect the audio directly with overlaid annotations. I thought HN might find it cool. As far as I'm aware, such an interactive web platform is a first for the academic speech community!
With TurnBench, we're also releasing a hand-annotated benchmark corpus, training set, public leaderboard, and our full eval protocol - which is based on human annotation.
In Lyft's defense, they are providing it anonymized under the NYCBS Data Use Policy. They also aren't providing the exact GPS routes, which is why OSRM is used to calculate the shortest path instead.
Is MapLibre GL a cheaper (free?) open source alternative?
Cool stuff btw. I’m trying to visualize weather model data myself (millions of points) at https://futureradar.net and have been researching client-side techniques like yours.
I see, yeah I use Next.js pretty regularly, you mean on hot reload then?
You still need to write your content as Markdown or something else, right? I feel the editors of Substack give some nice features like shortcuts for bold, italics. What do you do about that?
I launched this on HN over the summer, but it's millisecond-precise audio synchronization for multiple devices, performed purely in the browser! I'm sitting at around 5K daily active users now. Also, it's open-source!
I wanted to make sure the benchmark felt very accessible and understandable to people, so I built this in-browser dataset viewer at https://turnbench.sesame.com/conversations - it lets you inspect the audio directly with overlaid annotations. I thought HN might find it cool. As far as I'm aware, such an interactive web platform is a first for the academic speech community!
With TurnBench, we're also releasing a hand-annotated benchmark corpus, training set, public leaderboard, and our full eval protocol - which is based on human annotation.
Happy to answer questions!
Paper: https://arxiv.org/abs/2608.25218
Code: https://github.com/SesameAILabs/turnbench