Raising 4 children to be good citizens of planet earth. Product research fan. a hiker , a ( former) mountain biker. admire vegetarians . growing vegetables.
@seansong@fal@NVIDIAAI agree. fitting 30+ in a 600m parameter model require data and tooling to optimize for a language bundle :
https://t.co/m6vktnXAmk
Everthing matters : chunk size, langID or Auto and the ability to easily customize
🚀 Introducing NVIDIA PersonaPlex 🚀
🕷️ With great control comes great naturalness.
PersonaPlex explores the limits of naturalness and controllability in full-duplex conversational AI, and demonstrates that controllability is key to fully leveraging naturalness.
By prompting both voice and role, we can finetune a full-duplex model on real human conversations together with synthetic data in a single stage. This allows the model to learn natural tone and speaking style from real speech while covering a wide range of roles and behaviors from synthetic data, without getting confused. The result is human-level conversational naturalness with emergent generalization to new personas.
🎧 Demos and details
👉 https://t.co/M46zeTjhTG
🤗 Model weights
👉 https://t.co/CbSP3deztZ
💻 Code
👉 https://t.co/MxCZevEkJ7
📄 Paper
👉 https://t.co/TmhBOG1aGg
Grateful to all co-authors and collaborators. Looking forward to presenting at #ICASSP2026! 🎉
🚨 BREAKING NEWS: NVIDIA DROPS MIND-BLOWING CANARY-1B-FLASH & CANARY-180M-FLASH! 🔥🤯
Get ready to have your mind blown, AI enthusiasts!
@nvidia just unleashed two game-changing variants of their Canary model—Canary-1B-Flash and Canary-180M-Flash—supercharged for lightning-fast training and inference!
💨⚡ These multilingual beasts tackle speech recognition and translation in English, French, Spanish, and German like never before! 🌐🎙️
Here’s why this is HUGE:
🔥 Several TIMES FASTER than the original Canary-1B—blazing through tasks at breakneck speed!
📉 MORE ACCURATE—nailing precision like a pro!
��️ WORD-LEVEL TIMESTAMPS—pinpoint accuracy for every spoken word!
💰 CC-BY LICENSE—now open for commercial use, no restrictions! 🚀
Canary-1B-Flash snagged 2nd place on the
@huggingface
Open ASR Leaderboard with a jaw-dropping 1045 inverse real-time factor—faster than you can say "speech AI revolution"! 🤩 The 180M variant isn’t far behind, crushing it in 7th place while supporting translation and timestamps too! 🌟
Check out the demo on HF Spaces and dive into the tech paper for the juicy details: https://t.co/FOYWr7mVJe & https://t.co/gSXHqLKkwh.
Big props to @PiotrZelasko and the NVIDIA NeMo team for this epic release—AI’s future just got a whole lot brighter! 🌞👏
Last week, the world watched in horror at the images coming from Israel.
This was an act of pure evil. To the people of Israel, and to Jews around the world, be strong and brave. California stands with you.
Officially the snowiest Oct-Feb period since 1970 as recorded by the Central Sierra Snow Lab near Donner Summit, and likely the highest since 1952. It's also the coldest winter recorded so far for Tahoe City since 1952, but I doubt we're getting to 800 inches again.
@UCB_CSSL
@DutchFlyBird@omkamal@FlySJC@flySFO Ain't it great to get updates from Twitter rather than the airline itself?
We live 5 mins from SJ airport but I've been waiting for my family at SFO for the last four hours. At this point I'm not even mad, I'm *impressed* by the levels of incompetence. #LH458#Lufthansa
The Jan 6 hearings were once again amazing, stunning and magnificent. In a related story, because it is my responsibility and my job, I have recorded the valiant attempts made by the closed-captioning software to spell Pat Cippolone.
When you are asked to work on a new product initiative within a startup or a large company, probably the worst thing that could happen is that you are also given a whole team of engineers right away. Few understand this.
Ketanji Brown Jackson has put in a lot of time and effort to get to this point. ASR vendors for live TV, let's put a little effort into getting her name right.
#ASR#SpeechRecognition