lily clifford
@lilyjclifford
founder, chief celery man officer @rimelabs, human-centered speech technologies also @stanfordnlp
our mission is to make ai voices sound like real people, with all the messy imperfections, hi-fi mouth sounds, breaths that carry unique meanings. arcana captures the rhythm of real talk by real people
Introducing Arcana: AI Voices with Vibes 🔮 We just launched the most realistic spoken language (TTS) model like ever! At Rime, we're dedicated to capturing the authentic nuances of real human speech, accents, laughter, sighs, and everything in between. Arcana also lets…
Why do kids keep saying — “unalive”? Why is “rizz” — somehow — a real word now? Linguist Adam Aleksic — aka @etymology_nerd — unpacks it all in his new book — Algospeak. It’s not just about dodging censors — or gaming the algorithm. Algorithms — now — are shaping how we…

Early video of Twitter Founder, Jack Dorsey on obsession:
Early video of Twitter Founder, Jack Dorsey on obsession:
how is it that competitors with 10x our resources provide pretty much no support? just yesterday had an inbound customer come in getting no help on bugs for months and also a hackathon winner who got prizes from both rime and this other company getting told he had to pay to…
📞 Why do we say “B as in Bravo” or "B as in Boy"? Because over the phone (at 8kHz) B, D, P, T all blur together. In WWII, this led to dangerous miscommunications. So in 1956, NATO standardized the phonetic alphabet: Alpha, Bravo, Charlie... Each word chosen to cut through…

Not all AI is a monster. Not all AI is created equal. Training an LLM looks a lot like this 👇 A surreal, chaotic creature fed on the internet, refined through supervised tuning, and finally force-fed a smiley face of human feedback (RLHF) as a cherry on top. Meanwhile, voice…

unexpected finding we’ve uncovered at rime: the “low-energy” voice everyone worries about? they actually outperform this is another reason to not randomly pick voices but based on A/B tests and outcomes
🔐 How does Mirage Security simulate real social engineering attacks with Rime? 📞 They call your employees. 🗣️ They use cloned executive voices. 📹 They even simulate Zoom invites. These aren't fake phishing emails. They’re full-blown, multi-channel social engineering…
if there's one thing i've learned in scaling LLMs and spoken language models (SLMs) it's that loss curves mean very little. degenerate solutions are everywhere
The most human part of a voice is the hesitation. At first glance, “um” and “uh” sound meaningless. Research has shown that listeners do notice the difference. For instance, in typical child development, kids learn that “uh” typically signals a short pause, while “um” signals…